diff --git a/.gitignore b/.gitignore index 3938196ce0..f0f1f29f3e 100644 --- a/.gitignore +++ b/.gitignore @@ -16,6 +16,7 @@ m4/ltoptions.m4 m4/ltsugar.m4 m4/ltversion.m4 m4/lt~obsolete.m4 +autom4te.cache # downloaded files gtest @@ -61,3 +62,13 @@ src/test_plugin src/testzip.* src/zcg*zip ar-lib + +test-driver +compile + +src/**/*.log +src/**/*.trs + +# JavaBuild output. +java/target +javanano/target diff --git a/.gitmodules b/.gitmodules deleted file mode 100644 index 5351ea07a5..0000000000 --- a/.gitmodules +++ /dev/null @@ -1,3 +0,0 @@ -[submodule "upb"] - path = upb - url = https://github.com/haberman/upb diff --git a/.travis.yml b/.travis.yml new file mode 100644 index 0000000000..93013b8bdf --- /dev/null +++ b/.travis.yml @@ -0,0 +1,4 @@ +language: cpp +script: ./autogen.sh && ./configure && make distcheck +notifications: + email: false diff --git a/Makefile.am b/Makefile.am index dadccf2529..95316eb5d1 100644 --- a/Makefile.am +++ b/Makefile.am @@ -195,7 +195,6 @@ python_EXTRA_DIST= \ python/google/protobuf/internal/wire_format.py \ python/google/protobuf/internal/wire_format_test.py \ python/google/protobuf/internal/__init__.py \ - python/google/protobuf/internal/import_test_package/BUILD \ python/google/protobuf/internal/import_test_package/__init__.py \ python/google/protobuf/internal/import_test_package/inner.proto \ python/google/protobuf/internal/import_test_package/outer.proto \ @@ -240,15 +239,26 @@ python_EXTRA_DIST= \ python/README.txt ruby_EXTRA_DIST= \ - ruby/ext/defs.c \ - ruby/ext/encode_decode.c \ - ruby/ext/extconf.rb \ - ruby/ext/message.c \ - ruby/ext/protobuf.c \ - ruby/ext/protobuf.h \ - ruby/ext/repeated_field.c \ - ruby/ext/storage.c \ - ruby/ext/test.rb + ruby/README.md \ + ruby/Rakefile \ + ruby/ext/google/protobuf_c/defs.c \ + ruby/ext/google/protobuf_c/encode_decode.c \ + ruby/ext/google/protobuf_c/extconf.rb \ + ruby/ext/google/protobuf_c/map.c \ + ruby/ext/google/protobuf_c/message.c \ + ruby/ext/google/protobuf_c/protobuf.c \ + ruby/ext/google/protobuf_c/protobuf.h \ + ruby/ext/google/protobuf_c/repeated_field.c \ + ruby/ext/google/protobuf_c/storage.c \ + ruby/ext/google/protobuf_c/upb.c \ + ruby/ext/google/protobuf_c/upb.h \ + ruby/google-protobuf.gemspec \ + ruby/lib/google/protobuf.rb \ + ruby/tests/basic.rb \ + ruby/tests/stress.rb \ + ruby/tests/generated_code.proto \ + ruby/tests/generated_code.rb \ + ruby/tests/generated_code_test.rb all_EXTRA_DIST=$(java_EXTRA_DIST) $(python_EXTRA_DIST) $(ruby_EXTRA_DIST) diff --git a/README.md b/README.md index a79bfef17f..ae406792f5 100644 --- a/README.md +++ b/README.md @@ -1,6 +1,8 @@ Protocol Buffers - Google's data interchange format =================================================== +[![Build Status](https://travis-ci.org/google/protobuf.svg?branch=master)](https://travis-ci.org/google/protobuf) + Copyright 2008 Google Inc. https://developers.google.com/protocol-buffers/ diff --git a/autogen.sh b/autogen.sh index 1e97a72ae7..08966c631b 100755 --- a/autogen.sh +++ b/autogen.sh @@ -15,12 +15,6 @@ __EOF__ exit 1 fi -if test -e .git; then - # pull down git submodules. - git submodule init - git submodule update -fi - # Check that gtest is present. Usually it is already there since the # directory is set up as an SVN external. if test ! -e gtest; then diff --git a/javanano/pom.xml b/javanano/pom.xml index 3d98a5e041..409d8d6156 100644 --- a/javanano/pom.xml +++ b/javanano/pom.xml @@ -94,6 +94,7 @@ + diff --git a/javanano/src/main/java/com/google/protobuf/nano/CodedInputByteBufferNano.java b/javanano/src/main/java/com/google/protobuf/nano/CodedInputByteBufferNano.java index b147f69deb..b4f20fde71 100644 --- a/javanano/src/main/java/com/google/protobuf/nano/CodedInputByteBufferNano.java +++ b/javanano/src/main/java/com/google/protobuf/nano/CodedInputByteBufferNano.java @@ -638,4 +638,44 @@ public final class CodedInputByteBufferNano { throw InvalidProtocolBufferNanoException.truncatedMessage(); } } + + // Read a primitive type. + Object readPrimitiveField(int type) throws IOException { + switch (type) { + case InternalNano.TYPE_DOUBLE: + return readDouble(); + case InternalNano.TYPE_FLOAT: + return readFloat(); + case InternalNano.TYPE_INT64: + return readInt64(); + case InternalNano.TYPE_UINT64: + return readUInt64(); + case InternalNano.TYPE_INT32: + return readInt32(); + case InternalNano.TYPE_FIXED64: + return readFixed64(); + case InternalNano.TYPE_FIXED32: + return readFixed32(); + case InternalNano.TYPE_BOOL: + return readBool(); + case InternalNano.TYPE_STRING: + return readString(); + case InternalNano.TYPE_BYTES: + return readBytes(); + case InternalNano.TYPE_UINT32: + return readUInt32(); + case InternalNano.TYPE_ENUM: + return readEnum(); + case InternalNano.TYPE_SFIXED32: + return readSFixed32(); + case InternalNano.TYPE_SFIXED64: + return readSFixed64(); + case InternalNano.TYPE_SINT32: + return readSInt32(); + case InternalNano.TYPE_SINT64: + return readSInt64(); + default: + throw new IllegalArgumentException("Unknown type " + type); + } + } } diff --git a/javanano/src/main/java/com/google/protobuf/nano/CodedOutputByteBufferNano.java b/javanano/src/main/java/com/google/protobuf/nano/CodedOutputByteBufferNano.java index 37982b575f..2777f34cca 100644 --- a/javanano/src/main/java/com/google/protobuf/nano/CodedOutputByteBufferNano.java +++ b/javanano/src/main/java/com/google/protobuf/nano/CodedOutputByteBufferNano.java @@ -876,4 +876,128 @@ public final class CodedOutputByteBufferNano { // Note: the right-shift must be arithmetic return (n << 1) ^ (n >> 63); } + + static int computeFieldSize(int number, int type, Object object) { + switch (type) { + case InternalNano.TYPE_BOOL: + return computeBoolSize(number, (Boolean) object); + case InternalNano.TYPE_BYTES: + return computeBytesSize(number, (byte[]) object); + case InternalNano.TYPE_STRING: + return computeStringSize(number, (String) object); + case InternalNano.TYPE_FLOAT: + return computeFloatSize(number, (Float) object); + case InternalNano.TYPE_DOUBLE: + return computeDoubleSize(number, (Double) object); + case InternalNano.TYPE_ENUM: + return computeEnumSize(number, (Integer) object); + case InternalNano.TYPE_FIXED32: + return computeFixed32Size(number, (Integer) object); + case InternalNano.TYPE_INT32: + return computeInt32Size(number, (Integer) object); + case InternalNano.TYPE_UINT32: + return computeUInt32Size(number, (Integer) object); + case InternalNano.TYPE_SINT32: + return computeSInt32Size(number, (Integer) object); + case InternalNano.TYPE_SFIXED32: + return computeSFixed32Size(number, (Integer) object); + case InternalNano.TYPE_INT64: + return computeInt64Size(number, (Long) object); + case InternalNano.TYPE_UINT64: + return computeUInt64Size(number, (Long) object); + case InternalNano.TYPE_SINT64: + return computeSInt64Size(number, (Long) object); + case InternalNano.TYPE_FIXED64: + return computeFixed64Size(number, (Long) object); + case InternalNano.TYPE_SFIXED64: + return computeSFixed64Size(number, (Long) object); + case InternalNano.TYPE_MESSAGE: + return computeMessageSize(number, (MessageNano) object); + case InternalNano.TYPE_GROUP: + return computeGroupSize(number, (MessageNano) object); + default: + throw new IllegalArgumentException("Unknown type: " + type); + } + } + + void writeField(int number, int type, Object value) + throws IOException { + switch (type) { + case InternalNano.TYPE_DOUBLE: + Double doubleValue = (Double) value; + writeDouble(number, doubleValue); + break; + case InternalNano.TYPE_FLOAT: + Float floatValue = (Float) value; + writeFloat(number, floatValue); + break; + case InternalNano.TYPE_INT64: + Long int64Value = (Long) value; + writeInt64(number, int64Value); + break; + case InternalNano.TYPE_UINT64: + Long uint64Value = (Long) value; + writeUInt64(number, uint64Value); + break; + case InternalNano.TYPE_INT32: + Integer int32Value = (Integer) value; + writeInt32(number, int32Value); + break; + case InternalNano.TYPE_FIXED64: + Long fixed64Value = (Long) value; + writeFixed64(number, fixed64Value); + break; + case InternalNano.TYPE_FIXED32: + Integer fixed32Value = (Integer) value; + writeFixed32(number, fixed32Value); + break; + case InternalNano.TYPE_BOOL: + Boolean boolValue = (Boolean) value; + writeBool(number, boolValue); + break; + case InternalNano.TYPE_STRING: + String stringValue = (String) value; + writeString(number, stringValue); + break; + case InternalNano.TYPE_BYTES: + byte[] bytesValue = (byte[]) value; + writeBytes(number, bytesValue); + break; + case InternalNano.TYPE_UINT32: + Integer uint32Value = (Integer) value; + writeUInt32(number, uint32Value); + break; + case InternalNano.TYPE_ENUM: + Integer enumValue = (Integer) value; + writeEnum(number, enumValue); + break; + case InternalNano.TYPE_SFIXED32: + Integer sfixed32Value = (Integer) value; + writeSFixed32(number, sfixed32Value); + break; + case InternalNano.TYPE_SFIXED64: + Long sfixed64Value = (Long) value; + writeSFixed64(number, sfixed64Value); + break; + case InternalNano.TYPE_SINT32: + Integer sint32Value = (Integer) value; + writeSInt32(number, sint32Value); + break; + case InternalNano.TYPE_SINT64: + Long sint64Value = (Long) value; + writeSInt64(number, sint64Value); + break; + case InternalNano.TYPE_MESSAGE: + MessageNano messageValue = (MessageNano) value; + writeMessage(number, messageValue); + break; + case InternalNano.TYPE_GROUP: + MessageNano groupValue = (MessageNano) value; + writeGroup(number, groupValue); + break; + default: + throw new IOException("Unknown type: " + type); + } + } + } diff --git a/javanano/src/main/java/com/google/protobuf/nano/Extension.java b/javanano/src/main/java/com/google/protobuf/nano/Extension.java index 5d18ae6e5e..c29b030fb6 100644 --- a/javanano/src/main/java/com/google/protobuf/nano/Extension.java +++ b/javanano/src/main/java/com/google/protobuf/nano/Extension.java @@ -55,24 +55,24 @@ public class Extension, T> { * PrimitiveExtension // for primitive/enum typed extensions */ - public static final int TYPE_DOUBLE = 1; - public static final int TYPE_FLOAT = 2; - public static final int TYPE_INT64 = 3; - public static final int TYPE_UINT64 = 4; - public static final int TYPE_INT32 = 5; - public static final int TYPE_FIXED64 = 6; - public static final int TYPE_FIXED32 = 7; - public static final int TYPE_BOOL = 8; - public static final int TYPE_STRING = 9; - public static final int TYPE_GROUP = 10; - public static final int TYPE_MESSAGE = 11; - public static final int TYPE_BYTES = 12; - public static final int TYPE_UINT32 = 13; - public static final int TYPE_ENUM = 14; - public static final int TYPE_SFIXED32 = 15; - public static final int TYPE_SFIXED64 = 16; - public static final int TYPE_SINT32 = 17; - public static final int TYPE_SINT64 = 18; + public static final int TYPE_DOUBLE = InternalNano.TYPE_DOUBLE; + public static final int TYPE_FLOAT = InternalNano.TYPE_FLOAT; + public static final int TYPE_INT64 = InternalNano.TYPE_INT64; + public static final int TYPE_UINT64 = InternalNano.TYPE_UINT64; + public static final int TYPE_INT32 = InternalNano.TYPE_INT32; + public static final int TYPE_FIXED64 = InternalNano.TYPE_FIXED64; + public static final int TYPE_FIXED32 = InternalNano.TYPE_FIXED32; + public static final int TYPE_BOOL = InternalNano.TYPE_BOOL; + public static final int TYPE_STRING = InternalNano.TYPE_STRING; + public static final int TYPE_GROUP = InternalNano.TYPE_GROUP; + public static final int TYPE_MESSAGE = InternalNano.TYPE_MESSAGE; + public static final int TYPE_BYTES = InternalNano.TYPE_BYTES; + public static final int TYPE_UINT32 = InternalNano.TYPE_UINT32; + public static final int TYPE_ENUM = InternalNano.TYPE_ENUM; + public static final int TYPE_SFIXED32 = InternalNano.TYPE_SFIXED32; + public static final int TYPE_SFIXED64 = InternalNano.TYPE_SFIXED64; + public static final int TYPE_SINT32 = InternalNano.TYPE_SINT32; + public static final int TYPE_SINT64 = InternalNano.TYPE_SINT64; /** * Creates an {@code Extension} of the given message type and tag number. @@ -338,42 +338,7 @@ public class Extension, T> { @Override protected Object readData(CodedInputByteBufferNano input) { try { - switch (type) { - case TYPE_DOUBLE: - return input.readDouble(); - case TYPE_FLOAT: - return input.readFloat(); - case TYPE_INT64: - return input.readInt64(); - case TYPE_UINT64: - return input.readUInt64(); - case TYPE_INT32: - return input.readInt32(); - case TYPE_FIXED64: - return input.readFixed64(); - case TYPE_FIXED32: - return input.readFixed32(); - case TYPE_BOOL: - return input.readBool(); - case TYPE_STRING: - return input.readString(); - case TYPE_BYTES: - return input.readBytes(); - case TYPE_UINT32: - return input.readUInt32(); - case TYPE_ENUM: - return input.readEnum(); - case TYPE_SFIXED32: - return input.readSFixed32(); - case TYPE_SFIXED64: - return input.readSFixed64(); - case TYPE_SINT32: - return input.readSInt32(); - case TYPE_SINT64: - return input.readSInt64(); - default: - throw new IllegalArgumentException("Unknown type " + type); - } + return input.readPrimitiveField(type); } catch (IOException e) { throw new IllegalArgumentException("Error reading extension field", e); } diff --git a/javanano/src/main/java/com/google/protobuf/nano/InternalNano.java b/javanano/src/main/java/com/google/protobuf/nano/InternalNano.java index e08bb4b79d..c92ec41761 100644 --- a/javanano/src/main/java/com/google/protobuf/nano/InternalNano.java +++ b/javanano/src/main/java/com/google/protobuf/nano/InternalNano.java @@ -30,8 +30,13 @@ package com.google.protobuf.nano; +import com.google.protobuf.nano.MapFactories.MapFactory; + +import java.io.IOException; import java.io.UnsupportedEncodingException; import java.util.Arrays; +import java.util.Map; +import java.util.Map.Entry; /** * The classes contained within are used internally by the Protocol Buffer @@ -43,6 +48,26 @@ import java.util.Arrays; */ public final class InternalNano { + public static final int TYPE_DOUBLE = 1; + public static final int TYPE_FLOAT = 2; + public static final int TYPE_INT64 = 3; + public static final int TYPE_UINT64 = 4; + public static final int TYPE_INT32 = 5; + public static final int TYPE_FIXED64 = 6; + public static final int TYPE_FIXED32 = 7; + public static final int TYPE_BOOL = 8; + public static final int TYPE_STRING = 9; + public static final int TYPE_GROUP = 10; + public static final int TYPE_MESSAGE = 11; + public static final int TYPE_BYTES = 12; + public static final int TYPE_UINT32 = 13; + public static final int TYPE_ENUM = 14; + public static final int TYPE_SFIXED32 = 15; + public static final int TYPE_SFIXED64 = 16; + public static final int TYPE_SINT32 = 17; + public static final int TYPE_SINT64 = 18; + + private InternalNano() {} /** @@ -329,5 +354,203 @@ public final class InternalNano { } return result; } + private static final byte[] EMPTY_BYTES = new byte[0]; + private static Object primitiveDefaultValue(int type) { + switch (type) { + case TYPE_BOOL: + return Boolean.FALSE; + case TYPE_BYTES: + return EMPTY_BYTES; + case TYPE_STRING: + return ""; + case TYPE_FLOAT: + return Float.valueOf(0); + case TYPE_DOUBLE: + return Double.valueOf(0); + case TYPE_ENUM: + case TYPE_FIXED32: + case TYPE_INT32: + case TYPE_UINT32: + case TYPE_SINT32: + case TYPE_SFIXED32: + return Integer.valueOf(0); + case TYPE_INT64: + case TYPE_UINT64: + case TYPE_SINT64: + case TYPE_FIXED64: + case TYPE_SFIXED64: + return Long.valueOf(0L); + case TYPE_MESSAGE: + case TYPE_GROUP: + default: + throw new IllegalArgumentException( + "Type: " + type + " is not a primitive type."); + } + } + + /** + * Merges the map entry into the map field. Note this is only supposed to + * be called by generated messages. + * + * @param map the map field; may be null, in which case a map will be + * instantiated using the {@link MapFactories.MapFactory} + * @param input the input byte buffer + * @param keyType key type, as defined in InternalNano.TYPE_* + * @param valueType value type, as defined in InternalNano.TYPE_* + * @param value an new instance of the value, if the value is a TYPE_MESSAGE; + * otherwise this parameter can be null and will be ignored. + * @param keyTag wire tag for the key + * @param valueTag wire tag for the value + * @return the map field + * @throws IOException + */ + @SuppressWarnings("unchecked") + public static final Map mergeMapEntry( + CodedInputByteBufferNano input, + Map map, + MapFactory mapFactory, + int keyType, + int valueType, + V value, + int keyTag, + int valueTag) throws IOException { + map = mapFactory.forMap(map); + final int length = input.readRawVarint32(); + final int oldLimit = input.pushLimit(length); + K key = null; + while (true) { + int tag = input.readTag(); + if (tag == 0) { + break; + } + if (tag == keyTag) { + key = (K) input.readPrimitiveField(keyType); + } else if (tag == valueTag) { + if (valueType == TYPE_MESSAGE) { + input.readMessage((MessageNano) value); + } else { + value = (V) input.readPrimitiveField(valueType); + } + } else { + if (!input.skipField(tag)) { + break; + } + } + } + input.checkLastTagWas(0); + input.popLimit(oldLimit); + + if (key == null) { + // key can only be primitive types. + key = (K) primitiveDefaultValue(keyType); + } + + if (value == null) { + // message type value will be initialized by code-gen. + value = (V) primitiveDefaultValue(valueType); + } + + map.put(key, value); + return map; + } + + public static void serializeMapField( + CodedOutputByteBufferNano output, + Map map, int number, int keyType, int valueType) + throws IOException { + for (Entry entry: map.entrySet()) { + K key = entry.getKey(); + V value = entry.getValue(); + if (key == null || value == null) { + throw new IllegalStateException( + "keys and values in maps cannot be null"); + } + int entrySize = + CodedOutputByteBufferNano.computeFieldSize(1, keyType, key) + + CodedOutputByteBufferNano.computeFieldSize(2, valueType, value); + output.writeTag(number, WireFormatNano.WIRETYPE_LENGTH_DELIMITED); + output.writeRawVarint32(entrySize); + output.writeField(1, keyType, key); + output.writeField(2, valueType, value); + } + } + + public static int computeMapFieldSize( + Map map, int number, int keyType, int valueType) { + int size = 0; + int tagSize = CodedOutputByteBufferNano.computeTagSize(number); + for (Entry entry: map.entrySet()) { + K key = entry.getKey(); + V value = entry.getValue(); + if (key == null || value == null) { + throw new IllegalStateException( + "keys and values in maps cannot be null"); + } + int entrySize = + CodedOutputByteBufferNano.computeFieldSize(1, keyType, key) + + CodedOutputByteBufferNano.computeFieldSize(2, valueType, value); + size += tagSize + entrySize + + CodedOutputByteBufferNano.computeRawVarint32Size(entrySize); + } + return size; + } + + /** + * Checks whether two {@link Map} are equal. We don't use the default equals + * method of {@link Map} because it compares by identity not by content for + * byte arrays. + */ + public static boolean equals(Map a, Map b) { + if (a == b) { + return true; + } + if (a == null) { + return b.size() == 0; + } + if (b == null) { + return a.size() == 0; + } + if (a.size() != b.size()) { + return false; + } + for (Entry entry : a.entrySet()) { + if (!b.containsKey(entry.getKey())) { + return false; + } + if (!equalsMapValue(entry.getValue(), b.get(entry.getKey()))) { + return false; + } + } + return true; + } + + private static boolean equalsMapValue(Object a, Object b) { + if (a == null || b == null) { + throw new IllegalStateException( + "keys and values in maps cannot be null"); + } + if (a instanceof byte[] && b instanceof byte[]) { + return Arrays.equals((byte[]) a, (byte[]) b); + } + return a.equals(b); + } + + public static int hashCode(Map map) { + if (map == null) { + return 0; + } + int result = 0; + for (Entry entry : map.entrySet()) { + result += hashCodeForMap(entry.getKey()) + ^ hashCodeForMap(entry.getValue()); + } + return result; + } + private static int hashCodeForMap(Object o) { + if (o instanceof byte[]) { + return Arrays.hashCode((byte[]) o); + } + return o.hashCode(); + } } diff --git a/javanano/src/main/java/com/google/protobuf/nano/MapFactories.java b/javanano/src/main/java/com/google/protobuf/nano/MapFactories.java new file mode 100644 index 0000000000..98fa4877bc --- /dev/null +++ b/javanano/src/main/java/com/google/protobuf/nano/MapFactories.java @@ -0,0 +1,67 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2013 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +package com.google.protobuf.nano; + +import java.util.HashMap; +import java.util.Map; + +/** + * Utility class for maps support. + */ +public final class MapFactories { + public static interface MapFactory { + Map forMap(Map oldMap); + } + + // NOTE(liujisi): The factory setter is temporarily marked as package private. + // The way to provide customized implementations of maps for different + // platforms are still under discussion. Mark it as private to avoid exposing + // the API in proto3 alpha release. + /* public */ static void setMapFactory(MapFactory newMapFactory) { + mapFactory = newMapFactory; + } + + public static MapFactory getMapFactory() { + return mapFactory; + } + + private static class DefaultMapFactory implements MapFactory { + public Map forMap(Map oldMap) { + if (oldMap == null) { + return new HashMap(); + } + return oldMap; + } + } + private static volatile MapFactory mapFactory = new DefaultMapFactory(); + + private MapFactories() {} +} diff --git a/javanano/src/test/java/com/google/protobuf/nano/NanoTest.java b/javanano/src/test/java/com/google/protobuf/nano/NanoTest.java index 442f0b7472..aa279da2af 100644 --- a/javanano/src/test/java/com/google/protobuf/nano/NanoTest.java +++ b/javanano/src/test/java/com/google/protobuf/nano/NanoTest.java @@ -30,31 +30,11 @@ package com.google.protobuf.nano; -import com.google.protobuf.nano.CodedInputByteBufferNano; -import com.google.protobuf.nano.EnumClassNanoMultiple; -import com.google.protobuf.nano.EnumClassNanos; -import com.google.protobuf.nano.EnumValidity; -import com.google.protobuf.nano.EnumValidityAccessors; -import com.google.protobuf.nano.FileScopeEnumMultiple; -import com.google.protobuf.nano.FileScopeEnumRefNano; -import com.google.protobuf.nano.InternalNano; -import com.google.protobuf.nano.InvalidProtocolBufferNanoException; -import com.google.protobuf.nano.MessageNano; -import com.google.protobuf.nano.MessageScopeEnumRefNano; -import com.google.protobuf.nano.MultipleImportingNonMultipleNano1; -import com.google.protobuf.nano.MultipleImportingNonMultipleNano2; -import com.google.protobuf.nano.MultipleNameClashNano; +import com.google.protobuf.nano.MapTestProto.TestMap; +import com.google.protobuf.nano.MapTestProto.TestMap.MessageValue; import com.google.protobuf.nano.NanoAccessorsOuterClass.TestNanoAccessors; import com.google.protobuf.nano.NanoHasOuterClass.TestAllTypesNanoHas; -import com.google.protobuf.nano.NanoOuterClass; import com.google.protobuf.nano.NanoOuterClass.TestAllTypesNano; -import com.google.protobuf.nano.NanoReferenceTypes; -import com.google.protobuf.nano.NanoRepeatedPackables; -import com.google.protobuf.nano.PackedExtensions; -import com.google.protobuf.nano.RepeatedExtensions; -import com.google.protobuf.nano.SingularExtensions; -import com.google.protobuf.nano.TestRepeatedMergeNano; -import com.google.protobuf.nano.UnittestMultipleNano; import com.google.protobuf.nano.UnittestRecursiveNano.RecursiveMessageNano; import com.google.protobuf.nano.UnittestSimpleNano.SimpleMessageNano; import com.google.protobuf.nano.UnittestSingleNano.SingleMessageNano; @@ -67,6 +47,8 @@ import junit.framework.TestCase; import java.util.Arrays; import java.util.HashMap; +import java.util.Map; +import java.util.TreeMap; /** * Test nano runtime. @@ -3754,6 +3736,320 @@ public class NanoTest extends TestCase { assertTrue(Arrays.equals(new boolean[] {false, true, false, true}, nonPacked.bools)); } + public void testMapsSerializeAndParse() throws Exception { + TestMap origin = new TestMap(); + setMapMessage(origin); + assertMapMessageSet(origin); + + byte[] output = MessageNano.toByteArray(origin); + TestMap parsed = new TestMap(); + MessageNano.mergeFrom(parsed, output); + } + + public void testMapSerializeRejectNull() throws Exception { + TestMap primitiveMap = new TestMap(); + primitiveMap.int32ToInt32Field = new HashMap(); + primitiveMap.int32ToInt32Field.put(null, 1); + try { + MessageNano.toByteArray(primitiveMap); + fail("should reject null keys"); + } catch (IllegalStateException e) { + // pass. + } + + TestMap messageMap = new TestMap(); + messageMap.int32ToMessageField = + new HashMap(); + messageMap.int32ToMessageField.put(0, null); + try { + MessageNano.toByteArray(messageMap); + fail("should reject null values"); + } catch (IllegalStateException e) { + // pass. + } + } + + /** + * Tests that merging bytes containing conflicting keys with override the + * message value instead of merging the message value into the existing entry. + */ + public void testMapMergeOverrideMessageValues() throws Exception { + TestMap.MessageValue origValue = new TestMap.MessageValue(); + origValue.value = 1; + origValue.value2 = 2; + TestMap.MessageValue newValue = new TestMap.MessageValue(); + newValue.value = 3; + + TestMap origMessage = new TestMap(); + origMessage.int32ToMessageField = + new HashMap(); + origMessage.int32ToMessageField.put(1, origValue); + + TestMap newMessage = new TestMap(); + newMessage.int32ToMessageField = + new HashMap(); + newMessage.int32ToMessageField.put(1, newValue); + MessageNano.mergeFrom(origMessage, + MessageNano.toByteArray(newMessage)); + TestMap.MessageValue mergedValue = origMessage.int32ToMessageField.get(1); + assertEquals(3, mergedValue.value); + assertEquals(0, mergedValue.value2); + } + + /** + * Tests that when merging with empty entries, + * we will use default for the key and value, instead of null. + */ + public void testMapMergeEmptyEntry() throws Exception { + TestMap testMap = new TestMap(); + byte[] buffer = new byte[1024]; + CodedOutputByteBufferNano output = + CodedOutputByteBufferNano.newInstance(buffer); + // An empty entry for int32_to_int32 map. + output.writeTag(1, WireFormatNano.WIRETYPE_LENGTH_DELIMITED); + output.writeRawVarint32(0); + // An empty entry for int32_to_message map. + output.writeTag(5, WireFormatNano.WIRETYPE_LENGTH_DELIMITED); + output.writeRawVarint32(0); + + CodedInputByteBufferNano input = CodedInputByteBufferNano.newInstance( + buffer, 0, buffer.length - output.spaceLeft()); + testMap.mergeFrom(input); + assertNotNull(testMap.int32ToInt32Field);; + assertEquals(1, testMap.int32ToInt32Field.size()); + assertEquals(Integer.valueOf(0), testMap.int32ToInt32Field.get(0)); + assertNotNull(testMap.int32ToMessageField); + assertEquals(1, testMap.int32ToMessageField.size()); + TestMap.MessageValue messageValue = testMap.int32ToMessageField.get(0); + assertNotNull(messageValue); + assertEquals(0, messageValue.value); + assertEquals(0, messageValue.value2); + } + + public void testMapEquals() throws Exception { + TestMap a = new TestMap(); + TestMap b = new TestMap(); + + // empty and null map fields are equal. + assertTestMapEqual(a, b); + a.int32ToBytesField = new HashMap(); + assertTestMapEqual(a, b); + + a.int32ToInt32Field = new HashMap(); + b.int32ToInt32Field = new HashMap(); + setMap(a.int32ToInt32Field, deepCopy(int32Values), deepCopy(int32Values)); + setMap(b.int32ToInt32Field, deepCopy(int32Values), deepCopy(int32Values)); + assertTestMapEqual(a, b); + + a.int32ToMessageField = + new HashMap(); + b.int32ToMessageField = + new HashMap(); + setMap(a.int32ToMessageField, + deepCopy(int32Values), deepCopy(messageValues)); + setMap(b.int32ToMessageField, + deepCopy(int32Values), deepCopy(messageValues)); + assertTestMapEqual(a, b); + + a.stringToInt32Field = new HashMap(); + b.stringToInt32Field = new HashMap(); + setMap(a.stringToInt32Field, deepCopy(stringValues), deepCopy(int32Values)); + setMap(b.stringToInt32Field, deepCopy(stringValues), deepCopy(int32Values)); + assertTestMapEqual(a, b); + + a.int32ToBytesField = new HashMap(); + b.int32ToBytesField = new HashMap(); + setMap(a.int32ToBytesField, deepCopy(int32Values), deepCopy(bytesValues)); + setMap(b.int32ToBytesField, deepCopy(int32Values), deepCopy(bytesValues)); + assertTestMapEqual(a, b); + + // Make sure the map implementation does not matter. + a.int32ToStringField = new TreeMap(); + b.int32ToStringField = new HashMap(); + setMap(a.int32ToStringField, deepCopy(int32Values), deepCopy(stringValues)); + setMap(b.int32ToStringField, deepCopy(int32Values), deepCopy(stringValues)); + assertTestMapEqual(a, b); + + a.clear(); + b.clear(); + + // unequal cases: different value + a.int32ToInt32Field = new HashMap(); + b.int32ToInt32Field = new HashMap(); + a.int32ToInt32Field.put(1, 1); + b.int32ToInt32Field.put(1, 2); + assertTestMapUnequal(a, b); + // unequal case: additional entry + b.int32ToInt32Field.put(1, 1); + b.int32ToInt32Field.put(2, 1); + assertTestMapUnequal(a, b); + a.int32ToInt32Field.put(2, 1); + assertTestMapEqual(a, b); + + // unequal case: different message value. + a.int32ToMessageField = + new HashMap(); + b.int32ToMessageField = + new HashMap(); + MessageValue va = new MessageValue(); + va.value = 1; + MessageValue vb = new MessageValue(); + vb.value = 1; + a.int32ToMessageField.put(1, va); + b.int32ToMessageField.put(1, vb); + assertTestMapEqual(a, b); + vb.value = 2; + assertTestMapUnequal(a, b); + } + + private static void assertTestMapEqual(TestMap a, TestMap b) + throws Exception { + assertEquals(a.hashCode(), b.hashCode()); + assertTrue(a.equals(b)); + assertTrue(b.equals(a)); + } + + private static void assertTestMapUnequal(TestMap a, TestMap b) + throws Exception { + assertFalse(a.equals(b)); + assertFalse(b.equals(a)); + } + + private static final Integer[] int32Values = new Integer[] { + 0, 1, -1, Integer.MAX_VALUE, Integer.MIN_VALUE, + }; + + private static final Long[] int64Values = new Long[] { + 0L, 1L, -1L, Long.MAX_VALUE, Long.MIN_VALUE, + }; + + private static final String[] stringValues = new String[] { + "", "hello", "world", "foo", "bar", + }; + + private static final byte[][] bytesValues = new byte[][] { + new byte[] {}, + new byte[] {0}, + new byte[] {1, -1}, + new byte[] {127, -128}, + new byte[] {'a', 'b', '0', '1'}, + }; + + private static final Boolean[] boolValues = new Boolean[] { + false, true, + }; + + private static final Integer[] enumValues = new Integer[] { + TestMap.FOO, TestMap.BAR, TestMap.BAZ, TestMap.QUX, + Integer.MAX_VALUE /* unknown */, + }; + + private static final TestMap.MessageValue[] messageValues = + new TestMap.MessageValue[] { + newMapValueMessage(0), + newMapValueMessage(1), + newMapValueMessage(-1), + newMapValueMessage(Integer.MAX_VALUE), + newMapValueMessage(Integer.MIN_VALUE), + }; + + private static TestMap.MessageValue newMapValueMessage(int value) { + TestMap.MessageValue result = new TestMap.MessageValue(); + result.value = value; + return result; + } + + @SuppressWarnings("unchecked") + private static T[] deepCopy(T[] orig) throws Exception { + if (orig instanceof MessageValue[]) { + MessageValue[] result = new MessageValue[orig.length]; + for (int i = 0; i < orig.length; i++) { + result[i] = new MessageValue(); + MessageNano.mergeFrom( + result[i], MessageNano.toByteArray((MessageValue) orig[i])); + } + return (T[]) result; + } + if (orig instanceof byte[][]) { + byte[][] result = new byte[orig.length][]; + for (int i = 0; i < orig.length; i++) { + byte[] origBytes = (byte[]) orig[i]; + result[i] = Arrays.copyOf(origBytes, origBytes.length); + } + } + return Arrays.copyOf(orig, orig.length); + } + + private void setMap(Map map, K[] keys, V[] values) { + assert(keys.length == values.length); + for (int i = 0; i < keys.length; i++) { + map.put(keys[i], values[i]); + } + } + + private void assertMapSet( + Map map, K[] keys, V[] values) throws Exception { + assert(keys.length == values.length); + for (int i = 0; i < values.length; i++) { + assertEquals(values[i], map.get(keys[i])); + } + assertEquals(keys.length, map.size()); + } + + private void setMapMessage(TestMap testMap) { + testMap.int32ToInt32Field = new HashMap(); + testMap.int32ToBytesField = new HashMap(); + testMap.int32ToEnumField = new HashMap(); + testMap.int32ToMessageField = + new HashMap(); + testMap.int32ToStringField = new HashMap(); + testMap.stringToInt32Field = new HashMap(); + testMap.boolToBoolField = new HashMap(); + testMap.uint32ToUint32Field = new HashMap(); + testMap.sint32ToSint32Field = new HashMap(); + testMap.fixed32ToFixed32Field = new HashMap(); + testMap.sfixed32ToSfixed32Field = new HashMap(); + testMap.int64ToInt64Field = new HashMap(); + testMap.uint64ToUint64Field = new HashMap(); + testMap.sint64ToSint64Field = new HashMap(); + testMap.fixed64ToFixed64Field = new HashMap(); + testMap.sfixed64ToSfixed64Field = new HashMap(); + setMap(testMap.int32ToInt32Field, int32Values, int32Values); + setMap(testMap.int32ToBytesField, int32Values, bytesValues); + setMap(testMap.int32ToEnumField, int32Values, enumValues); + setMap(testMap.int32ToMessageField, int32Values, messageValues); + setMap(testMap.int32ToStringField, int32Values, stringValues); + setMap(testMap.stringToInt32Field, stringValues, int32Values); + setMap(testMap.boolToBoolField, boolValues, boolValues); + setMap(testMap.uint32ToUint32Field, int32Values, int32Values); + setMap(testMap.sint32ToSint32Field, int32Values, int32Values); + setMap(testMap.fixed32ToFixed32Field, int32Values, int32Values); + setMap(testMap.sfixed32ToSfixed32Field, int32Values, int32Values); + setMap(testMap.int64ToInt64Field, int64Values, int64Values); + setMap(testMap.uint64ToUint64Field, int64Values, int64Values); + setMap(testMap.sint64ToSint64Field, int64Values, int64Values); + setMap(testMap.fixed64ToFixed64Field, int64Values, int64Values); + setMap(testMap.sfixed64ToSfixed64Field, int64Values, int64Values); + } + private void assertMapMessageSet(TestMap testMap) throws Exception { + assertMapSet(testMap.int32ToInt32Field, int32Values, int32Values); + assertMapSet(testMap.int32ToBytesField, int32Values, bytesValues); + assertMapSet(testMap.int32ToEnumField, int32Values, enumValues); + assertMapSet(testMap.int32ToMessageField, int32Values, messageValues); + assertMapSet(testMap.int32ToStringField, int32Values, stringValues); + assertMapSet(testMap.stringToInt32Field, stringValues, int32Values); + assertMapSet(testMap.boolToBoolField, boolValues, boolValues); + assertMapSet(testMap.uint32ToUint32Field, int32Values, int32Values); + assertMapSet(testMap.sint32ToSint32Field, int32Values, int32Values); + assertMapSet(testMap.fixed32ToFixed32Field, int32Values, int32Values); + assertMapSet(testMap.sfixed32ToSfixed32Field, int32Values, int32Values); + assertMapSet(testMap.int64ToInt64Field, int64Values, int64Values); + assertMapSet(testMap.uint64ToUint64Field, int64Values, int64Values); + assertMapSet(testMap.sint64ToSint64Field, int64Values, int64Values); + assertMapSet(testMap.fixed64ToFixed64Field, int64Values, int64Values); + assertMapSet(testMap.sfixed64ToSfixed64Field, int64Values, int64Values); + } + private void assertRepeatedPackablesEqual( NanoRepeatedPackables.NonPacked nonPacked, NanoRepeatedPackables.Packed packed) { // Not using MessageNano.equals() -- that belongs to a separate test. diff --git a/javanano/src/test/java/com/google/protobuf/nano/map_test.proto b/javanano/src/test/java/com/google/protobuf/nano/map_test.proto new file mode 100644 index 0000000000..f72833ad63 --- /dev/null +++ b/javanano/src/test/java/com/google/protobuf/nano/map_test.proto @@ -0,0 +1,70 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2008 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +syntax = "proto3"; + +package map_test; + +option java_package = "com.google.protobuf.nano"; +option java_outer_classname = "MapTestProto"; + +message TestMap { + message MessageValue { + int32 value = 1; + int32 value2 = 2; + } + enum EnumValue { + FOO = 0; + BAR = 1; + BAZ = 2; + QUX = 3; + } + + map int32_to_int32_field = 1; + map int32_to_string_field = 2; + map int32_to_bytes_field = 3; + map int32_to_enum_field = 4; + map int32_to_message_field = 5; + map string_to_int32_field = 6; + map bool_to_bool_field = 7; + + // Test all the other primitive types. As the key and value are not coupled in + // the implementation, we do not test all the combinations of key/value pairs, + // so that we can keep the number of test cases manageable + map uint32_to_uint32_field = 11; + map sint32_to_sint32_field = 12; + map fixed32_to_fixed32_field = 13; + map sfixed32_to_sfixed32_field = 14; + map int64_to_int64_field = 15; + map uint64_to_uint64_field = 16; + map sint64_to_sint64_field = 17; + map fixed64_to_fixed64_field = 18; + map sfixed64_to_sfixed64_field = 19; +} diff --git a/python/README.txt b/python/README.txt index da044af44a..7d85245834 100644 --- a/python/README.txt +++ b/python/README.txt @@ -26,7 +26,7 @@ join the Protocol Buffers discussion list and let us know! Installation ============ -1) Make sure you have Python 2.4 or newer. If in doubt, run: +1) Make sure you have Python 2.6 or newer. If in doubt, run: $ python -V @@ -35,7 +35,7 @@ Installation If you would rather install it manually, you may do so by following the instructions on this page: - http://peak.telecommunity.com/DevCenter/EasyInstall#installation-instructions + https://packaging.python.org/en/latest/installing.html#setup-for-installing-packages 3) Build the C++ code, or install a binary distribution of protoc. If you install a binary distribution, make sure that it is the same diff --git a/python/google/protobuf/__init__.py b/python/google/protobuf/__init__.py index e69de29bb2..ec3b09349b 100755 --- a/python/google/protobuf/__init__.py +++ b/python/google/protobuf/__init__.py @@ -0,0 +1,35 @@ +# Protocol Buffers - Google's data interchange format +# Copyright 2008 Google Inc. All rights reserved. +# https://developers.google.com/protocol-buffers/ +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions are +# met: +# +# * Redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer. +# * Redistributions in binary form must reproduce the above +# copyright notice, this list of conditions and the following disclaimer +# in the documentation and/or other materials provided with the +# distribution. +# * Neither the name of Google Inc. nor the names of its +# contributors may be used to endorse or promote products derived from +# this software without specific prior written permission. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +# Needs to stay compatible with Python 2.5 due to GAE. +# +# Copyright 2007 Google Inc. All Rights Reserved. + +__version__ = '3.0.0-pre' diff --git a/python/google/protobuf/internal/import_test_package/BUILD b/python/google/protobuf/internal/import_test_package/BUILD deleted file mode 100644 index 90e59505b7..0000000000 --- a/python/google/protobuf/internal/import_test_package/BUILD +++ /dev/null @@ -1,27 +0,0 @@ -# Description: -# An example package that contains nested protos that are imported from -# __init__.py. See testPackageInitializationImport in reflection_test.py for -# details. - -package( - default_visibility = ["//net/proto2/python/internal:__pkg__"], -) - -proto_library( - name = "inner_proto", - srcs = ["inner.proto"], - py_api_version = 2, -) - -proto_library( - name = "outer_proto", - srcs = ["outer.proto"], - py_api_version = 2, - deps = [":inner_proto"], -) - -py_library( - name = "import_test_package", - srcs = ["__init__.py"], - deps = [":outer_proto"], -) diff --git a/ruby/README.md b/ruby/README.md index c966a1039a..88e9c0e19d 100644 --- a/ruby/README.md +++ b/ruby/README.md @@ -18,13 +18,7 @@ To build this Ruby extension, you will need: * a C compiler * the upb submodule -First, ensure that upb/ is checked out: - - $ cd .. # top level protobuf directory - $ git submodule init - $ git submodule update - -Then install the required Ruby gems: +First, install the required Ruby gems: $ sudo gem install bundler rake rake-compiler rspec rubygems-tasks @@ -32,3 +26,7 @@ Then build the Gem: $ rake gem $ gem install pkg/protobuf-$VERSION.gem + +This gem includes the upb parsing and serialization library as a single-file +amalgamation. It is up-to-date with upb git commit +`535bc2fe2f2b467f59347ffc9449e11e47791257`. diff --git a/ruby/Rakefile b/ruby/Rakefile index b782b403da..ae7d8059e3 100644 --- a/ruby/Rakefile +++ b/ruby/Rakefile @@ -1,35 +1,19 @@ require "rake/extensiontask" require "rake/testtask" -spec = Gem::Specification.new do |s| - s.name = "protobuf" - s.version = "2.6.2" - s.licenses = ["BSD"] - s.summary = "Protocol Buffers" - s.description = "Protocol Buffers are Google's data interchange format." - s.authors = ["Protobuf Authors"] - s.email = "protobuf@googlegroups.com" - - s.files = ["lib/protobuf_c.so", "lib/protobuf.rb"] -end +spec = Gem::Specification.load("google-protobuf.gemspec") Rake::ExtensionTask.new("protobuf_c", spec) do |ext| - ext.lib_dir = "lib" - ext.config_script = "extconf.rb" + ext.ext_dir = "ext/google/protobuf_c" + ext.lib_dir = "lib/google" end Rake::TestTask.new(:test => :build) do |t| t.test_files = FileList["tests/*.rb"] end -task :chmod do - File.chmod(0755, "lib/protobuf_c.so") -end - Gem::PackageTask.new(spec) do |pkg| end -task :package => :chmod -task :gem => :chmod task :build => [:clean, :compile] task :default => [:build] diff --git a/ruby/ext/protobuf_c/defs.c b/ruby/ext/google/protobuf_c/defs.c similarity index 73% rename from ruby/ext/protobuf_c/defs.c rename to ruby/ext/google/protobuf_c/defs.c index bb6f10e1f3..b39c27f493 100644 --- a/ruby/ext/protobuf_c/defs.c +++ b/ruby/ext/google/protobuf_c/defs.c @@ -58,11 +58,15 @@ static upb_def* check_notfrozen(const upb_def* def) { } static upb_msgdef* check_msg_notfrozen(const upb_msgdef* def) { - return (upb_msgdef*)check_notfrozen((const upb_def*)def); + return upb_downcast_msgdef_mutable(check_notfrozen((const upb_def*)def)); } static upb_fielddef* check_field_notfrozen(const upb_fielddef* def) { - return (upb_fielddef*)check_notfrozen((const upb_def*)def); + return upb_downcast_fielddef_mutable(check_notfrozen((const upb_def*)def)); +} + +static upb_oneofdef* check_oneof_notfrozen(const upb_oneofdef* def) { + return (upb_oneofdef*)check_notfrozen((const upb_def*)def); } static upb_enumdef* check_enum_notfrozen(const upb_enumdef* def) { @@ -226,6 +230,7 @@ DEFINE_CLASS(Descriptor, "Google::Protobuf::Descriptor"); void Descriptor_mark(void* _self) { Descriptor* self = _self; rb_gc_mark(self->klass); + rb_gc_mark(self->typeclass_references); } void Descriptor_free(void* _self) { @@ -270,6 +275,7 @@ VALUE Descriptor_alloc(VALUE klass) { self->fill_method = NULL; self->pb_serialize_handlers = NULL; self->json_serialize_handlers = NULL; + self->typeclass_references = rb_ary_new(); return ret; } @@ -280,6 +286,9 @@ void Descriptor_register(VALUE module) { rb_define_method(klass, "each", Descriptor_each, 0); rb_define_method(klass, "lookup", Descriptor_lookup, 1); rb_define_method(klass, "add_field", Descriptor_add_field, 1); + rb_define_method(klass, "add_oneof", Descriptor_add_oneof, 1); + rb_define_method(klass, "each_oneof", Descriptor_each_oneof, 0); + rb_define_method(klass, "lookup_oneof", Descriptor_lookup_oneof, 1); rb_define_method(klass, "msgclass", Descriptor_msgclass, 0); rb_define_method(klass, "name", Descriptor_name, 0); rb_define_method(klass, "name=", Descriptor_name_set, 1); @@ -326,10 +335,10 @@ VALUE Descriptor_name_set(VALUE _self, VALUE str) { VALUE Descriptor_each(VALUE _self) { DEFINE_SELF(Descriptor, self, _self); - upb_msg_iter it; - for (upb_msg_begin(&it, self->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, self->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); VALUE obj = get_def_obj(field); rb_yield(obj); @@ -358,7 +367,7 @@ VALUE Descriptor_lookup(VALUE _self, VALUE name) { * call-seq: * Descriptor.add_field(field) => nil * - * Adds the given FieldDescriptor to this message type. The descriptor must not + * Adds the given FieldDescriptor to this message type. This descriptor must not * have been added to a pool yet. Raises an exception if a field with the same * name or number already exists. Sub-type references (e.g. for fields of type * message) are not resolved at this point. @@ -375,6 +384,67 @@ VALUE Descriptor_add_field(VALUE _self, VALUE obj) { return Qnil; } +/* + * call-seq: + * Descriptor.add_oneof(oneof) => nil + * + * Adds the given OneofDescriptor to this message type. This descriptor must not + * have been added to a pool yet. Raises an exception if a oneof with the same + * name already exists, or if any of the oneof's fields' names or numbers + * conflict with an existing field in this message type. All fields in the oneof + * are added to the message descriptor. Sub-type references (e.g. for fields of + * type message) are not resolved at this point. + */ +VALUE Descriptor_add_oneof(VALUE _self, VALUE obj) { + DEFINE_SELF(Descriptor, self, _self); + upb_msgdef* mut_def = check_msg_notfrozen(self->msgdef); + OneofDescriptor* def = ruby_to_OneofDescriptor(obj); + upb_oneofdef* mut_oneof_def = check_oneof_notfrozen(def->oneofdef); + CHECK_UPB( + upb_msgdef_addoneof(mut_def, mut_oneof_def, NULL, &status), + "Adding oneof to Descriptor failed"); + add_def_obj(def->oneofdef, obj); + return Qnil; +} + +/* + * call-seq: + * Descriptor.each_oneof(&block) => nil + * + * Invokes the given block for each oneof in this message type, passing the + * corresponding OneofDescriptor. + */ +VALUE Descriptor_each_oneof(VALUE _self) { + DEFINE_SELF(Descriptor, self, _self); + + upb_msg_oneof_iter it; + for (upb_msg_oneof_begin(&it, self->msgdef); + !upb_msg_oneof_done(&it); + upb_msg_oneof_next(&it)) { + const upb_oneofdef* oneof = upb_msg_iter_oneof(&it); + VALUE obj = get_def_obj(oneof); + rb_yield(obj); + } + return Qnil; +} + +/* + * call-seq: + * Descriptor.lookup_oneof(name) => OneofDescriptor + * + * Returns the oneof descriptor for the oneof with the given name, if present, + * or nil if none. + */ +VALUE Descriptor_lookup_oneof(VALUE _self, VALUE name) { + DEFINE_SELF(Descriptor, self, _self); + const char* s = get_str(name); + const upb_oneofdef* oneof = upb_msgdef_ntooz(self->msgdef, s); + if (oneof == NULL) { + return Qnil; + } + return get_def_obj(oneof); +} + /* * call-seq: * Descriptor.msgclass => message_klass @@ -741,6 +811,120 @@ VALUE FieldDescriptor_set(VALUE _self, VALUE msg_rb, VALUE value) { return Qnil; } +// ----------------------------------------------------------------------------- +// OneofDescriptor. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(OneofDescriptor, "Google::Protobuf::OneofDescriptor"); + +void OneofDescriptor_mark(void* _self) { +} + +void OneofDescriptor_free(void* _self) { + OneofDescriptor* self = _self; + upb_oneofdef_unref(self->oneofdef, &self->oneofdef); + xfree(self); +} + +/* + * call-seq: + * OneofDescriptor.new => oneof_descriptor + * + * Creates a new, empty, oneof descriptor. The oneof may only be modified prior + * to being added to a message descriptor which is subsequently added to a pool. + */ +VALUE OneofDescriptor_alloc(VALUE klass) { + OneofDescriptor* self = ALLOC(OneofDescriptor); + VALUE ret = TypedData_Wrap_Struct(klass, &_OneofDescriptor_type, self); + self->oneofdef = upb_oneofdef_new(&self->oneofdef); + return ret; +} + +void OneofDescriptor_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "OneofDescriptor", rb_cObject); + rb_define_alloc_func(klass, OneofDescriptor_alloc); + rb_define_method(klass, "name", OneofDescriptor_name, 0); + rb_define_method(klass, "name=", OneofDescriptor_name_set, 1); + rb_define_method(klass, "add_field", OneofDescriptor_add_field, 1); + rb_define_method(klass, "each", OneofDescriptor_each, 0); + rb_include_module(klass, rb_mEnumerable); + cOneofDescriptor = klass; + rb_gc_register_address(&cOneofDescriptor); +} + +/* + * call-seq: + * OneofDescriptor.name => name + * + * Returns the name of this oneof. + */ +VALUE OneofDescriptor_name(VALUE _self) { + DEFINE_SELF(OneofDescriptor, self, _self); + return rb_str_maybe_null(upb_oneofdef_name(self->oneofdef)); +} + +/* + * call-seq: + * OneofDescriptor.name = name + * + * Sets a new name for this oneof. The oneof must not have been added to a + * message descriptor yet. + */ +VALUE OneofDescriptor_name_set(VALUE _self, VALUE value) { + DEFINE_SELF(OneofDescriptor, self, _self); + upb_oneofdef* mut_def = check_oneof_notfrozen(self->oneofdef); + const char* str = get_str(value); + CHECK_UPB(upb_oneofdef_setname(mut_def, str, &status), + "Error setting oneof name"); + return Qnil; +} + +/* + * call-seq: + * OneofDescriptor.add_field(field) => nil + * + * Adds a field to this oneof. The field may have been added to this oneof in + * the past, or the message to which this oneof belongs (if any), but may not + * have already been added to any other oneof or message. Otherwise, an + * exception is raised. + * + * All fields added to the oneof via this method will be automatically added to + * the message to which this oneof belongs, if it belongs to one currently, or + * else will be added to any message to which the oneof is later added at the + * time that it is added. + */ +VALUE OneofDescriptor_add_field(VALUE _self, VALUE obj) { + DEFINE_SELF(OneofDescriptor, self, _self); + upb_oneofdef* mut_def = check_oneof_notfrozen(self->oneofdef); + FieldDescriptor* def = ruby_to_FieldDescriptor(obj); + upb_fielddef* mut_field_def = check_field_notfrozen(def->fielddef); + CHECK_UPB( + upb_oneofdef_addfield(mut_def, mut_field_def, NULL, &status), + "Adding field to OneofDescriptor failed"); + add_def_obj(def->fielddef, obj); + return Qnil; +} + +/* + * call-seq: + * OneofDescriptor.each(&block) => nil + * + * Iterates through fields in this oneof, yielding to the block on each one. + */ +VALUE OneofDescriptor_each(VALUE _self, VALUE field) { + DEFINE_SELF(OneofDescriptor, self, _self); + upb_oneof_iter it; + for (upb_oneof_begin(&it, self->oneofdef); + !upb_oneof_done(&it); + upb_oneof_next(&it)) { + const upb_fielddef* f = upb_oneof_iter_field(&it); + VALUE obj = get_def_obj(f); + rb_yield(obj); + } + return Qnil; +} + // ----------------------------------------------------------------------------- // EnumDescriptor. // ----------------------------------------------------------------------------- @@ -923,6 +1107,7 @@ DEFINE_CLASS(MessageBuilderContext, void MessageBuilderContext_mark(void* _self) { MessageBuilderContext* self = _self; rb_gc_mark(self->descriptor); + rb_gc_mark(self->builder); } void MessageBuilderContext_free(void* _self) { @@ -935,6 +1120,7 @@ VALUE MessageBuilderContext_alloc(VALUE klass) { VALUE ret = TypedData_Wrap_Struct( klass, &_MessageBuilderContext_type, self); self->descriptor = Qnil; + self->builder = Qnil; return ret; } @@ -943,24 +1129,30 @@ void MessageBuilderContext_register(VALUE module) { module, "MessageBuilderContext", rb_cObject); rb_define_alloc_func(klass, MessageBuilderContext_alloc); rb_define_method(klass, "initialize", - MessageBuilderContext_initialize, 1); + MessageBuilderContext_initialize, 2); rb_define_method(klass, "optional", MessageBuilderContext_optional, -1); rb_define_method(klass, "required", MessageBuilderContext_required, -1); rb_define_method(klass, "repeated", MessageBuilderContext_repeated, -1); + rb_define_method(klass, "map", MessageBuilderContext_map, -1); + rb_define_method(klass, "oneof", MessageBuilderContext_oneof, 1); cMessageBuilderContext = klass; rb_gc_register_address(&cMessageBuilderContext); } /* * call-seq: - * MessageBuilderContext.new(desc) => context + * MessageBuilderContext.new(desc, builder) => context * - * Create a new builder context around the given message descriptor. This class - * is intended to serve as a DSL context to be used with #instance_eval. + * Create a new message builder context around the given message descriptor and + * builder context. This class is intended to serve as a DSL context to be used + * with #instance_eval. */ -VALUE MessageBuilderContext_initialize(VALUE _self, VALUE msgdef) { +VALUE MessageBuilderContext_initialize(VALUE _self, + VALUE msgdef, + VALUE builder) { DEFINE_SELF(MessageBuilderContext, self, _self); self->descriptor = msgdef; + self->builder = builder; return Qnil; } @@ -1065,6 +1257,201 @@ VALUE MessageBuilderContext_repeated(int argc, VALUE* argv, VALUE _self) { name, type, number, type_class); } +/* + * call-seq: + * MessageBuilderContext.map(name, key_type, value_type, number, + * value_type_class = nil) + * + * Defines a new map field on this message type with the given key and value + * types, tag number, and type class (for message and enum value types). The key + * type must be :int32/:uint32/:int64/:uint64, :bool, or :string. The value type + * type must be a Ruby symbol (as accepted by FieldDescriptor#type=) and the + * type_class must be a string, if present (as accepted by + * FieldDescriptor#submsg_name=). + */ +VALUE MessageBuilderContext_map(int argc, VALUE* argv, VALUE _self) { + DEFINE_SELF(MessageBuilderContext, self, _self); + + if (argc < 4) { + rb_raise(rb_eArgError, "Expected at least 4 arguments."); + } + VALUE name = argv[0]; + VALUE key_type = argv[1]; + VALUE value_type = argv[2]; + VALUE number = argv[3]; + VALUE type_class = (argc > 4) ? argv[4] : Qnil; + + // Validate the key type. We can't accept enums, messages, or floats/doubles + // as map keys. (We exclude these explicitly, and the field-descriptor setter + // below then ensures that the type is one of the remaining valid options.) + if (SYM2ID(key_type) == rb_intern("float") || + SYM2ID(key_type) == rb_intern("double") || + SYM2ID(key_type) == rb_intern("enum") || + SYM2ID(key_type) == rb_intern("message")) { + rb_raise(rb_eArgError, + "Cannot add a map field with a float, double, enum, or message " + "type."); + } + + // Create a new message descriptor for the map entry message, and create a + // repeated submessage field here with that type. + VALUE mapentry_desc = rb_class_new_instance(0, NULL, cDescriptor); + VALUE mapentry_desc_name = rb_funcall(self->descriptor, rb_intern("name"), 0); + mapentry_desc_name = rb_str_cat2(mapentry_desc_name, "_MapEntry_"); + mapentry_desc_name = rb_str_cat2(mapentry_desc_name, + rb_id2name(SYM2ID(name))); + Descriptor_name_set(mapentry_desc, mapentry_desc_name); + + // The 'mapentry' attribute has no Ruby setter because we do not want the user + // attempting to DIY the setup below; we want to ensure that the fields are + // correct. So we reach into the msgdef here to set the bit manually. + Descriptor* mapentry_desc_self = ruby_to_Descriptor(mapentry_desc); + upb_msgdef_setmapentry((upb_msgdef*)mapentry_desc_self->msgdef, true); + + // optional key = 1; + VALUE key_field = rb_class_new_instance(0, NULL, cFieldDescriptor); + FieldDescriptor_name_set(key_field, rb_str_new2("key")); + FieldDescriptor_label_set(key_field, ID2SYM(rb_intern("optional"))); + FieldDescriptor_number_set(key_field, INT2NUM(1)); + FieldDescriptor_type_set(key_field, key_type); + Descriptor_add_field(mapentry_desc, key_field); + + // optional value = 2; + VALUE value_field = rb_class_new_instance(0, NULL, cFieldDescriptor); + FieldDescriptor_name_set(value_field, rb_str_new2("value")); + FieldDescriptor_label_set(value_field, ID2SYM(rb_intern("optional"))); + FieldDescriptor_number_set(value_field, INT2NUM(2)); + FieldDescriptor_type_set(value_field, value_type); + if (type_class != Qnil) { + VALUE submsg_name = rb_str_new2("."); // prepend '.' to make name absolute. + submsg_name = rb_str_append(submsg_name, type_class); + FieldDescriptor_submsg_name_set(value_field, submsg_name); + } + Descriptor_add_field(mapentry_desc, value_field); + + // Add the map-entry message type to the current builder, and use the type to + // create the map field itself. + Builder* builder_self = ruby_to_Builder(self->builder); + rb_ary_push(builder_self->pending_list, mapentry_desc); + + VALUE map_field = rb_class_new_instance(0, NULL, cFieldDescriptor); + VALUE name_str = rb_str_new2(rb_id2name(SYM2ID(name))); + FieldDescriptor_name_set(map_field, name_str); + FieldDescriptor_number_set(map_field, number); + FieldDescriptor_label_set(map_field, ID2SYM(rb_intern("repeated"))); + FieldDescriptor_type_set(map_field, ID2SYM(rb_intern("message"))); + VALUE submsg_name = rb_str_new2("."); // prepend '.' to make name absolute. + submsg_name = rb_str_append(submsg_name, mapentry_desc_name); + FieldDescriptor_submsg_name_set(map_field, submsg_name); + Descriptor_add_field(self->descriptor, map_field); + + return Qnil; +} + +/* + * call-seq: + * MessageBuilderContext.oneof(name, &block) => nil + * + * Creates a new OneofDescriptor with the given name, creates a + * OneofBuilderContext attached to that OneofDescriptor, evaluates the given + * block in the context of that OneofBuilderContext with #instance_eval, and + * then adds the oneof to the message. + * + * This is the recommended, idiomatic way to build oneof definitions. + */ +VALUE MessageBuilderContext_oneof(VALUE _self, VALUE name) { + DEFINE_SELF(MessageBuilderContext, self, _self); + VALUE oneofdef = rb_class_new_instance(0, NULL, cOneofDescriptor); + VALUE args[2] = { oneofdef, self->builder }; + VALUE ctx = rb_class_new_instance(2, args, cOneofBuilderContext); + VALUE block = rb_block_proc(); + VALUE name_str = rb_str_new2(rb_id2name(SYM2ID(name))); + rb_funcall(oneofdef, rb_intern("name="), 1, name_str); + rb_funcall_with_block(ctx, rb_intern("instance_eval"), 0, NULL, block); + Descriptor_add_oneof(self->descriptor, oneofdef); + + return Qnil; +} + +// ----------------------------------------------------------------------------- +// OneofBuilderContext. +// ----------------------------------------------------------------------------- + +DEFINE_CLASS(OneofBuilderContext, + "Google::Protobuf::Internal::OneofBuilderContext"); + +void OneofBuilderContext_mark(void* _self) { + OneofBuilderContext* self = _self; + rb_gc_mark(self->descriptor); + rb_gc_mark(self->builder); +} + +void OneofBuilderContext_free(void* _self) { + OneofBuilderContext* self = _self; + xfree(self); +} + +VALUE OneofBuilderContext_alloc(VALUE klass) { + OneofBuilderContext* self = ALLOC(OneofBuilderContext); + VALUE ret = TypedData_Wrap_Struct( + klass, &_OneofBuilderContext_type, self); + self->descriptor = Qnil; + self->builder = Qnil; + return ret; +} + +void OneofBuilderContext_register(VALUE module) { + VALUE klass = rb_define_class_under( + module, "OneofBuilderContext", rb_cObject); + rb_define_alloc_func(klass, OneofBuilderContext_alloc); + rb_define_method(klass, "initialize", + OneofBuilderContext_initialize, 2); + rb_define_method(klass, "optional", OneofBuilderContext_optional, -1); + cOneofBuilderContext = klass; + rb_gc_register_address(&cOneofBuilderContext); +} + +/* + * call-seq: + * OneofBuilderContext.new(desc, builder) => context + * + * Create a new oneof builder context around the given oneof descriptor and + * builder context. This class is intended to serve as a DSL context to be used + * with #instance_eval. + */ +VALUE OneofBuilderContext_initialize(VALUE _self, + VALUE oneofdef, + VALUE builder) { + DEFINE_SELF(OneofBuilderContext, self, _self); + self->descriptor = oneofdef; + self->builder = builder; + return Qnil; +} + +/* + * call-seq: + * OneofBuilderContext.optional(name, type, number, type_class = nil) + * + * Defines a new optional field in this oneof with the given type, tag number, + * and type class (for message and enum fields). The type must be a Ruby symbol + * (as accepted by FieldDescriptor#type=) and the type_class must be a string, + * if present (as accepted by FieldDescriptor#submsg_name=). + */ +VALUE OneofBuilderContext_optional(int argc, VALUE* argv, VALUE _self) { + DEFINE_SELF(OneofBuilderContext, self, _self); + + if (argc < 3) { + rb_raise(rb_eArgError, "Expected at least 3 arguments."); + } + VALUE name = argv[0]; + VALUE type = argv[1]; + VALUE number = argv[2]; + VALUE type_class = (argc > 3) ? argv[3] : Qnil; + + return msgdef_add_field(self->descriptor, "optional", + name, type, number, type_class); +} + // ----------------------------------------------------------------------------- // EnumBuilderContext. // ----------------------------------------------------------------------------- @@ -1190,7 +1577,8 @@ void Builder_register(VALUE module) { VALUE Builder_add_message(VALUE _self, VALUE name) { DEFINE_SELF(Builder, self, _self); VALUE msgdef = rb_class_new_instance(0, NULL, cDescriptor); - VALUE ctx = rb_class_new_instance(1, &msgdef, cMessageBuilderContext); + VALUE args[2] = { msgdef, _self }; + VALUE ctx = rb_class_new_instance(2, args, cMessageBuilderContext); VALUE block = rb_block_proc(); rb_funcall(msgdef, rb_intern("name="), 1, name); rb_funcall_with_block(ctx, rb_intern("instance_eval"), 0, NULL, block); @@ -1221,8 +1609,10 @@ VALUE Builder_add_enum(VALUE _self, VALUE name) { static void validate_msgdef(const upb_msgdef* msgdef) { // Verify that no required fields exist. proto3 does not support these. - upb_msg_iter it; - for (upb_msg_begin(&it, msgdef); !upb_msg_done(&it); upb_msg_next(&it)) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); if (upb_fielddef_label(field) == UPB_LABEL_REQUIRED) { rb_raise(rb_eTypeError, "Required fields are unsupported in proto3."); diff --git a/ruby/ext/protobuf_c/encode_decode.c b/ruby/ext/google/protobuf_c/encode_decode.c similarity index 54% rename from ruby/ext/protobuf_c/encode_decode.c rename to ruby/ext/google/protobuf_c/encode_decode.c index 8aba3c9edb..5730504d57 100644 --- a/ruby/ext/protobuf_c/encode_decode.c +++ b/ruby/ext/google/protobuf_c/encode_decode.c @@ -59,16 +59,45 @@ static const void *newsubmsghandlerdata(upb_handlers* h, uint32_t ofs, return hd; } +typedef struct { + size_t ofs; // union data slot + size_t case_ofs; // oneof_case field + uint32_t oneof_case_num; // oneof-case number to place in oneof_case field + const upb_msgdef *md; // msgdef, for oneof submessage handler +} oneof_handlerdata_t; + +static const void *newoneofhandlerdata(upb_handlers *h, + uint32_t ofs, + uint32_t case_ofs, + const upb_fielddef *f) { + oneof_handlerdata_t *hd = ALLOC(oneof_handlerdata_t); + hd->ofs = ofs; + hd->case_ofs = case_ofs; + // We reuse the field tag number as a oneof union discriminant tag. Note that + // we don't expose these numbers to the user, so the only requirement is that + // we have some unique ID for each union case/possibility. The field tag + // numbers are already present and are easy to use so there's no reason to + // create a separate ID space. In addition, using the field tag number here + // lets us easily look up the field in the oneof accessor. + hd->oneof_case_num = upb_fielddef_number(f); + if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE) { + hd->md = upb_fielddef_msgsubdef(f); + } else { + hd->md = NULL; + } + upb_handlers_addcleanup(h, hd, free); + return hd; +} + // A handler that starts a repeated field. Gets the Repeated*Field instance for // this field (such an instance always exists even in an empty message). static void *startseq_handler(void* closure, const void* hd) { MessageHeader* msg = closure; const size_t *ofs = hd; - return (void*)DEREF(Message_data(msg), *ofs, VALUE); + return (void*)DEREF(msg, *ofs, VALUE); } -// Handlers that append primitive values to a repeated field (a regular Ruby -// array for now). +// Handlers that append primitive values to a repeated field. #define DEFINE_APPEND_HANDLER(type, ctype) \ static bool append##type##_handler(void *closure, const void *hd, \ ctype val) { \ @@ -85,7 +114,7 @@ DEFINE_APPEND_HANDLER(int64, int64_t) DEFINE_APPEND_HANDLER(uint64, uint64_t) DEFINE_APPEND_HANDLER(double, double) -// Appends a string to a repeated field (a regular Ruby array for now). +// Appends a string to a repeated field. static void* appendstr_handler(void *closure, const void *hd, size_t size_hint) { @@ -96,7 +125,7 @@ static void* appendstr_handler(void *closure, return (void*)str; } -// Appends a 'bytes' string to a repeated field (a regular Ruby array for now). +// Appends a 'bytes' string to a repeated field. static void* appendbytes_handler(void *closure, const void *hd, size_t size_hint) { @@ -115,7 +144,7 @@ static void* str_handler(void *closure, const size_t *ofs = hd; VALUE str = rb_str_new2(""); rb_enc_associate(str, kRubyStringUtf8Encoding); - DEREF(Message_data(msg), *ofs, VALUE) = str; + DEREF(msg, *ofs, VALUE) = str; return (void*)str; } @@ -127,7 +156,7 @@ static void* bytes_handler(void *closure, const size_t *ofs = hd; VALUE str = rb_str_new2(""); rb_enc_associate(str, kRubyString8bitEncoding); - DEREF(Message_data(msg), *ofs, VALUE) = str; + DEREF(msg, *ofs, VALUE) = str; return (void*)str; } @@ -163,110 +192,391 @@ static void *submsg_handler(void *closure, const void *hd) { get_def_obj((void*)submsgdata->md); VALUE subklass = Descriptor_msgclass(subdesc); - if (DEREF(Message_data(msg), submsgdata->ofs, VALUE) == Qnil) { - DEREF(Message_data(msg), submsgdata->ofs, VALUE) = + if (DEREF(msg, submsgdata->ofs, VALUE) == Qnil) { + DEREF(msg, submsgdata->ofs, VALUE) = rb_class_new_instance(0, NULL, subklass); } - VALUE submsg_rb = DEREF(Message_data(msg), submsgdata->ofs, VALUE); + VALUE submsg_rb = DEREF(msg, submsgdata->ofs, VALUE); MessageHeader* submsg; TypedData_Get_Struct(submsg_rb, MessageHeader, &Message_type, submsg); return submsg; } -static void add_handlers_for_message(const void *closure, upb_handlers *h) { - Descriptor* desc = ruby_to_Descriptor( - get_def_obj((void*)upb_handlers_msgdef(h))); - // Ensure layout exists. We may be invoked to create handlers for a given - // message if we are included as a submsg of another message type before our - // class is actually built, so to work around this, we just create the layout - // (and handlers, in the class-building function) on-demand. - if (desc->layout == NULL) { - desc->layout = create_layout(desc->msgdef); +// Handler data for startmap/endmap handlers. +typedef struct { + size_t ofs; + upb_fieldtype_t key_field_type; + upb_fieldtype_t value_field_type; + + // We know that we can hold this reference because the handlerdata has the + // same lifetime as the upb_handlers struct, and the upb_handlers struct holds + // a reference to the upb_msgdef, which in turn has references to its subdefs. + const upb_def* value_field_subdef; +} map_handlerdata_t; + +// Temporary frame for map parsing: at the beginning of a map entry message, a +// submsg handler allocates a frame to hold (i) a reference to the Map object +// into which this message will be inserted and (ii) storage slots to +// temporarily hold the key and value for this map entry until the end of the +// submessage. When the submessage ends, another handler is called to insert the +// value into the map. +typedef struct { + VALUE map; + char key_storage[NATIVE_SLOT_MAX_SIZE]; + char value_storage[NATIVE_SLOT_MAX_SIZE]; +} map_parse_frame_t; + +// Handler to begin a map entry: allocates a temporary frame. This is the +// 'startsubmsg' handler on the msgdef that contains the map field. +static void *startmapentry_handler(void *closure, const void *hd) { + MessageHeader* msg = closure; + const map_handlerdata_t* mapdata = hd; + VALUE map_rb = DEREF(msg, mapdata->ofs, VALUE); + + map_parse_frame_t* frame = ALLOC(map_parse_frame_t); + frame->map = map_rb; + + native_slot_init(mapdata->key_field_type, &frame->key_storage); + native_slot_init(mapdata->value_field_type, &frame->value_storage); + + return frame; +} + +// Handler to end a map entry: inserts the value defined during the message into +// the map. This is the 'endmsg' handler on the map entry msgdef. +static bool endmap_handler(void *closure, const void *hd, upb_status* s) { + map_parse_frame_t* frame = closure; + const map_handlerdata_t* mapdata = hd; + + VALUE key = native_slot_get( + mapdata->key_field_type, Qnil, + &frame->key_storage); + + VALUE value_field_typeclass = Qnil; + if (mapdata->value_field_type == UPB_TYPE_MESSAGE || + mapdata->value_field_type == UPB_TYPE_ENUM) { + value_field_typeclass = get_def_obj(mapdata->value_field_subdef); } - upb_msg_iter i; + VALUE value = native_slot_get( + mapdata->value_field_type, value_field_typeclass, + &frame->value_storage); - for (upb_msg_begin(&i, desc->msgdef); - !upb_msg_done(&i); - upb_msg_next(&i)) { - const upb_fielddef *f = upb_msg_iter_field(&i); - size_t offset = desc->layout->offsets[upb_fielddef_index(f)]; + Map_index_set(frame->map, key, value); + free(frame); + + return true; +} + +// Allocates a new map_handlerdata_t given the map entry message definition. If +// the offset of the field within the parent message is also given, that is +// added to the handler data as well. Note that this is called *twice* per map +// field: once in the parent message handler setup when setting the startsubmsg +// handler and once in the map entry message handler setup when setting the +// key/value and endmsg handlers. The reason is that there is no easy way to +// pass the handlerdata down to the sub-message handler setup. +static map_handlerdata_t* new_map_handlerdata( + size_t ofs, + const upb_msgdef* mapentry_def, + Descriptor* desc) { + + map_handlerdata_t* hd = ALLOC(map_handlerdata_t); + hd->ofs = ofs; + const upb_fielddef* key_field = upb_msgdef_itof(mapentry_def, + MAP_KEY_FIELD); + assert(key_field != NULL); + hd->key_field_type = upb_fielddef_type(key_field); + const upb_fielddef* value_field = upb_msgdef_itof(mapentry_def, + MAP_VALUE_FIELD); + assert(value_field != NULL); + hd->value_field_type = upb_fielddef_type(value_field); + hd->value_field_subdef = upb_fielddef_subdef(value_field); + + return hd; +} + +// Handlers that set primitive values in oneofs. +#define DEFINE_ONEOF_HANDLER(type, ctype) \ + static bool oneof##type##_handler(void *closure, const void *hd, \ + ctype val) { \ + const oneof_handlerdata_t *oneofdata = hd; \ + DEREF(closure, oneofdata->case_ofs, uint32_t) = \ + oneofdata->oneof_case_num; \ + DEREF(closure, oneofdata->ofs, ctype) = val; \ + return true; \ + } + +DEFINE_ONEOF_HANDLER(bool, bool) +DEFINE_ONEOF_HANDLER(int32, int32_t) +DEFINE_ONEOF_HANDLER(uint32, uint32_t) +DEFINE_ONEOF_HANDLER(float, float) +DEFINE_ONEOF_HANDLER(int64, int64_t) +DEFINE_ONEOF_HANDLER(uint64, uint64_t) +DEFINE_ONEOF_HANDLER(double, double) - if (upb_fielddef_isseq(f)) { +#undef DEFINE_ONEOF_HANDLER + +// Handlers for strings in a oneof. +static void *oneofstr_handler(void *closure, + const void *hd, + size_t size_hint) { + MessageHeader* msg = closure; + const oneof_handlerdata_t *oneofdata = hd; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyStringUtf8Encoding); + DEREF(msg, oneofdata->case_ofs, uint32_t) = + oneofdata->oneof_case_num; + DEREF(msg, oneofdata->ofs, VALUE) = str; + return (void*)str; +} + +static void *oneofbytes_handler(void *closure, + const void *hd, + size_t size_hint) { + MessageHeader* msg = closure; + const oneof_handlerdata_t *oneofdata = hd; + VALUE str = rb_str_new2(""); + rb_enc_associate(str, kRubyString8bitEncoding); + DEREF(msg, oneofdata->case_ofs, uint32_t) = + oneofdata->oneof_case_num; + DEREF(msg, oneofdata->ofs, VALUE) = str; + return (void*)str; +} + +// Handler for a submessage field in a oneof. +static void *oneofsubmsg_handler(void *closure, + const void *hd) { + MessageHeader* msg = closure; + const oneof_handlerdata_t *oneofdata = hd; + uint32_t oldcase = DEREF(msg, oneofdata->case_ofs, uint32_t); + + VALUE subdesc = + get_def_obj((void*)oneofdata->md); + VALUE subklass = Descriptor_msgclass(subdesc); + + if (oldcase != oneofdata->oneof_case_num || + DEREF(msg, oneofdata->ofs, VALUE) == Qnil) { + DEREF(msg, oneofdata->ofs, VALUE) = + rb_class_new_instance(0, NULL, subklass); + } + // Set the oneof case *after* allocating the new class instance -- otherwise, + // if the Ruby GC is invoked as part of a call into the VM, it might invoke + // our mark routines, and our mark routines might see the case value + // indicating a VALUE is present and expect a valid VALUE. See comment in + // layout_set() for more detail: basically, the change to the value and the + // case must be atomic w.r.t. the Ruby VM. + DEREF(msg, oneofdata->case_ofs, uint32_t) = + oneofdata->oneof_case_num; + + VALUE submsg_rb = DEREF(msg, oneofdata->ofs, VALUE); + MessageHeader* submsg; + TypedData_Get_Struct(submsg_rb, MessageHeader, &Message_type, submsg); + return submsg; +} + +// Set up handlers for a repeated field. +static void add_handlers_for_repeated_field(upb_handlers *h, + const upb_fielddef *f, + size_t offset) { + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, newhandlerdata(h, offset)); + upb_handlers_setstartseq(h, f, startseq_handler, &attr); + upb_handlerattr_uninit(&attr); + + switch (upb_fielddef_type(f)) { + +#define SET_HANDLER(utype, ltype) \ + case utype: \ + upb_handlers_set##ltype(h, f, append##ltype##_handler, NULL); \ + break; + + SET_HANDLER(UPB_TYPE_BOOL, bool); + SET_HANDLER(UPB_TYPE_INT32, int32); + SET_HANDLER(UPB_TYPE_UINT32, uint32); + SET_HANDLER(UPB_TYPE_ENUM, int32); + SET_HANDLER(UPB_TYPE_FLOAT, float); + SET_HANDLER(UPB_TYPE_INT64, int64); + SET_HANDLER(UPB_TYPE_UINT64, uint64); + SET_HANDLER(UPB_TYPE_DOUBLE, double); + +#undef SET_HANDLER + + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; + upb_handlers_setstartstr(h, f, is_bytes ? + appendbytes_handler : appendstr_handler, + NULL); + upb_handlers_setstring(h, f, stringdata_handler, NULL); + break; + } + case UPB_TYPE_MESSAGE: { + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, newsubmsghandlerdata(h, 0, f)); + upb_handlers_setstartsubmsg(h, f, appendsubmsg_handler, &attr); + upb_handlerattr_uninit(&attr); + break; + } + } +} + +// Set up handlers for a singular field. +static void add_handlers_for_singular_field(upb_handlers *h, + const upb_fielddef *f, + size_t offset) { + switch (upb_fielddef_type(f)) { + case UPB_TYPE_BOOL: + case UPB_TYPE_INT32: + case UPB_TYPE_UINT32: + case UPB_TYPE_ENUM: + case UPB_TYPE_FLOAT: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT64: + case UPB_TYPE_DOUBLE: + upb_shim_set(h, f, offset, -1); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; upb_handlerattr_sethandlerdata(&attr, newhandlerdata(h, offset)); - upb_handlers_setstartseq(h, f, startseq_handler, &attr); + upb_handlers_setstartstr(h, f, + is_bytes ? bytes_handler : str_handler, + &attr); + upb_handlers_setstring(h, f, stringdata_handler, &attr); + upb_handlerattr_uninit(&attr); + break; + } + case UPB_TYPE_MESSAGE: { + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, newsubmsghandlerdata(h, offset, f)); + upb_handlers_setstartsubmsg(h, f, submsg_handler, &attr); upb_handlerattr_uninit(&attr); + break; + } + } +} - switch (upb_fielddef_type(f)) { +// Adds handlers to a map field. +static void add_handlers_for_mapfield(upb_handlers* h, + const upb_fielddef* fielddef, + size_t offset, + Descriptor* desc) { + const upb_msgdef* map_msgdef = upb_fielddef_msgsubdef(fielddef); + map_handlerdata_t* hd = new_map_handlerdata(offset, map_msgdef, desc); + upb_handlers_addcleanup(h, hd, free); + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, hd); + upb_handlers_setstartsubmsg(h, fielddef, startmapentry_handler, &attr); + upb_handlerattr_uninit(&attr); +} + +// Adds handlers to a map-entry msgdef. +static void add_handlers_for_mapentry(const upb_msgdef* msgdef, + upb_handlers* h, + Descriptor* desc) { + const upb_fielddef* key_field = map_entry_key(msgdef); + const upb_fielddef* value_field = map_entry_value(msgdef); + map_handlerdata_t* hd = new_map_handlerdata(0, msgdef, desc); + upb_handlers_addcleanup(h, hd, free); + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, hd); + upb_handlers_setendmsg(h, endmap_handler, &attr); + + add_handlers_for_singular_field( + h, key_field, + offsetof(map_parse_frame_t, key_storage)); + add_handlers_for_singular_field( + h, value_field, + offsetof(map_parse_frame_t, value_storage)); +} + +// Set up handlers for a oneof field. +static void add_handlers_for_oneof_field(upb_handlers *h, + const upb_fielddef *f, + size_t offset, + size_t oneof_case_offset) { + + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata( + &attr, newoneofhandlerdata(h, offset, oneof_case_offset, f)); + + switch (upb_fielddef_type(f)) { #define SET_HANDLER(utype, ltype) \ case utype: \ - upb_handlers_set##ltype(h, f, append##ltype##_handler, NULL); \ + upb_handlers_set##ltype(h, f, oneof##ltype##_handler, &attr); \ break; - SET_HANDLER(UPB_TYPE_BOOL, bool); - SET_HANDLER(UPB_TYPE_INT32, int32); - SET_HANDLER(UPB_TYPE_UINT32, uint32); - SET_HANDLER(UPB_TYPE_ENUM, int32); - SET_HANDLER(UPB_TYPE_FLOAT, float); - SET_HANDLER(UPB_TYPE_INT64, int64); - SET_HANDLER(UPB_TYPE_UINT64, uint64); - SET_HANDLER(UPB_TYPE_DOUBLE, double); + SET_HANDLER(UPB_TYPE_BOOL, bool); + SET_HANDLER(UPB_TYPE_INT32, int32); + SET_HANDLER(UPB_TYPE_UINT32, uint32); + SET_HANDLER(UPB_TYPE_ENUM, int32); + SET_HANDLER(UPB_TYPE_FLOAT, float); + SET_HANDLER(UPB_TYPE_INT64, int64); + SET_HANDLER(UPB_TYPE_UINT64, uint64); + SET_HANDLER(UPB_TYPE_DOUBLE, double); #undef SET_HANDLER - case UPB_TYPE_STRING: - case UPB_TYPE_BYTES: { - bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; - upb_handlers_setstartstr(h, f, is_bytes ? - appendbytes_handler : appendstr_handler, - NULL); - upb_handlers_setstring(h, f, stringdata_handler, NULL); - } - case UPB_TYPE_MESSAGE: { - upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; - upb_handlerattr_sethandlerdata(&attr, newsubmsghandlerdata(h, 0, f)); - upb_handlers_setstartsubmsg(h, f, appendsubmsg_handler, &attr); - upb_handlerattr_uninit(&attr); - break; - } - } + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; + upb_handlers_setstartstr(h, f, is_bytes ? + oneofbytes_handler : oneofstr_handler, + &attr); + upb_handlers_setstring(h, f, stringdata_handler, NULL); + break; } + case UPB_TYPE_MESSAGE: { + upb_handlers_setstartsubmsg(h, f, oneofsubmsg_handler, &attr); + break; + } + } - switch (upb_fielddef_type(f)) { - case UPB_TYPE_BOOL: - case UPB_TYPE_INT32: - case UPB_TYPE_UINT32: - case UPB_TYPE_ENUM: - case UPB_TYPE_FLOAT: - case UPB_TYPE_INT64: - case UPB_TYPE_UINT64: - case UPB_TYPE_DOUBLE: - // The shim writes directly at the given offset (instead of using - // DEREF()) so we need to add the msg overhead. - upb_shim_set(h, f, offset + sizeof(MessageHeader), -1); - break; - case UPB_TYPE_STRING: - case UPB_TYPE_BYTES: { - bool is_bytes = upb_fielddef_type(f) == UPB_TYPE_BYTES; - upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; - upb_handlerattr_sethandlerdata(&attr, newhandlerdata(h, offset)); - upb_handlers_setstartstr(h, f, - is_bytes ? bytes_handler : str_handler, - &attr); - upb_handlers_setstring(h, f, stringdata_handler, &attr); - upb_handlerattr_uninit(&attr); - break; - } - case UPB_TYPE_MESSAGE: { - upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; - upb_handlerattr_sethandlerdata(&attr, newsubmsghandlerdata(h, offset, f)); - upb_handlers_setstartsubmsg(h, f, submsg_handler, &attr); - upb_handlerattr_uninit(&attr); - break; - } + upb_handlerattr_uninit(&attr); +} + + +static void add_handlers_for_message(const void *closure, upb_handlers *h) { + const upb_msgdef* msgdef = upb_handlers_msgdef(h); + Descriptor* desc = ruby_to_Descriptor(get_def_obj((void*)msgdef)); + + // If this is a mapentry message type, set up a special set of handlers and + // bail out of the normal (user-defined) message type handling. + if (upb_msgdef_mapentry(msgdef)) { + add_handlers_for_mapentry(msgdef, h, desc); + return; + } + + // Ensure layout exists. We may be invoked to create handlers for a given + // message if we are included as a submsg of another message type before our + // class is actually built, so to work around this, we just create the layout + // (and handlers, in the class-building function) on-demand. + if (desc->layout == NULL) { + desc->layout = create_layout(desc->msgdef); + } + + upb_msg_field_iter i; + for (upb_msg_field_begin(&i, desc->msgdef); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + size_t offset = desc->layout->fields[upb_fielddef_index(f)].offset + + sizeof(MessageHeader); + + if (upb_fielddef_containingoneof(f)) { + size_t oneof_case_offset = + desc->layout->fields[upb_fielddef_index(f)].case_offset + + sizeof(MessageHeader); + add_handlers_for_oneof_field(h, f, offset, oneof_case_offset); + } else if (is_map_field(f)) { + add_handlers_for_mapfield(h, f, offset, desc); + } else if (upb_fielddef_isseq(f)) { + add_handlers_for_repeated_field(h, f, offset); + } else { + add_handlers_for_singular_field(h, f, offset); } } } @@ -558,6 +868,88 @@ static void putary(VALUE ary, const upb_fielddef *f, upb_sink *sink, upb_sink_endseq(sink, getsel(f, UPB_HANDLER_ENDSEQ)); } +static void put_ruby_value(VALUE value, + const upb_fielddef *f, + VALUE type_class, + int depth, + upb_sink *sink) { + upb_selector_t sel = 0; + if (upb_fielddef_isprimitive(f)) { + sel = getsel(f, upb_handlers_getprimitivehandlertype(f)); + } + + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: + upb_sink_putint32(sink, sel, NUM2INT(value)); + break; + case UPB_TYPE_INT64: + upb_sink_putint64(sink, sel, NUM2LL(value)); + break; + case UPB_TYPE_UINT32: + upb_sink_putuint32(sink, sel, NUM2UINT(value)); + break; + case UPB_TYPE_UINT64: + upb_sink_putuint64(sink, sel, NUM2ULL(value)); + break; + case UPB_TYPE_FLOAT: + upb_sink_putfloat(sink, sel, NUM2DBL(value)); + break; + case UPB_TYPE_DOUBLE: + upb_sink_putdouble(sink, sel, NUM2DBL(value)); + break; + case UPB_TYPE_ENUM: { + if (TYPE(value) == T_SYMBOL) { + value = rb_funcall(type_class, rb_intern("resolve"), 1, value); + } + upb_sink_putint32(sink, sel, NUM2INT(value)); + break; + } + case UPB_TYPE_BOOL: + upb_sink_putbool(sink, sel, value == Qtrue); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + putstr(value, f, sink); + break; + case UPB_TYPE_MESSAGE: + putsubmsg(value, f, sink, depth); + } +} + +static void putmap(VALUE map, const upb_fielddef *f, upb_sink *sink, + int depth) { + if (map == Qnil) return; + Map* self = ruby_to_Map(map); + + upb_sink subsink; + + upb_sink_startseq(sink, getsel(f, UPB_HANDLER_STARTSEQ), &subsink); + + assert(upb_fielddef_type(f) == UPB_TYPE_MESSAGE); + const upb_fielddef* key_field = map_field_key(f); + const upb_fielddef* value_field = map_field_value(f); + + Map_iter it; + for (Map_begin(map, &it); !Map_done(&it); Map_next(&it)) { + VALUE key = Map_iter_key(&it); + VALUE value = Map_iter_value(&it); + + upb_sink entry_sink; + upb_sink_startsubmsg(&subsink, getsel(f, UPB_HANDLER_STARTSUBMSG), &entry_sink); + upb_sink_startmsg(&entry_sink); + + put_ruby_value(key, key_field, Qnil, depth + 1, &entry_sink); + put_ruby_value(value, value_field, self->value_type_class, depth + 1, + &entry_sink); + + upb_status status; + upb_sink_endmsg(&entry_sink, &status); + upb_sink_endsubmsg(&subsink, getsel(f, UPB_HANDLER_ENDSUBMSG)); + } + + upb_sink_endseq(sink, getsel(f, UPB_HANDLER_ENDSEQ)); +} + static void putmsg(VALUE msg_rb, const Descriptor* desc, upb_sink *sink, int depth) { upb_sink_startmsg(sink); @@ -571,33 +963,53 @@ static void putmsg(VALUE msg_rb, const Descriptor* desc, MessageHeader* msg; TypedData_Get_Struct(msg_rb, MessageHeader, &Message_type, msg); - void* msg_data = Message_data(msg); - upb_msg_iter i; - for (upb_msg_begin(&i, desc->msgdef); - !upb_msg_done(&i); - upb_msg_next(&i)) { + upb_msg_field_iter i; + for (upb_msg_field_begin(&i, desc->msgdef); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { upb_fielddef *f = upb_msg_iter_field(&i); - uint32_t offset = desc->layout->offsets[upb_fielddef_index(f)]; + uint32_t offset = + desc->layout->fields[upb_fielddef_index(f)].offset + + sizeof(MessageHeader); + + if (upb_fielddef_containingoneof(f)) { + uint32_t oneof_case_offset = + desc->layout->fields[upb_fielddef_index(f)].case_offset + + sizeof(MessageHeader); + // For a oneof, check that this field is actually present -- skip all the + // below if not. + if (DEREF(msg, oneof_case_offset, uint32_t) != + upb_fielddef_number(f)) { + continue; + } + // Otherwise, fall through to the appropriate singular-field handler + // below. + } - if (upb_fielddef_isseq(f)) { - VALUE ary = DEREF(msg_data, offset, VALUE); + if (is_map_field(f)) { + VALUE map = DEREF(msg, offset, VALUE); + if (map != Qnil) { + putmap(map, f, sink, depth); + } + } else if (upb_fielddef_isseq(f)) { + VALUE ary = DEREF(msg, offset, VALUE); if (ary != Qnil) { putary(ary, f, sink, depth); } } else if (upb_fielddef_isstring(f)) { - VALUE str = DEREF(msg_data, offset, VALUE); + VALUE str = DEREF(msg, offset, VALUE); if (RSTRING_LEN(str) > 0) { putstr(str, f, sink); } } else if (upb_fielddef_issubmsg(f)) { - putsubmsg(DEREF(msg_data, offset, VALUE), f, sink, depth); + putsubmsg(DEREF(msg, offset, VALUE), f, sink, depth); } else { upb_selector_t sel = getsel(f, upb_handlers_getprimitivehandlertype(f)); #define T(upbtypeconst, upbtype, ctype, default_value) \ case upbtypeconst: { \ - ctype value = DEREF(msg_data, offset, ctype); \ + ctype value = DEREF(msg, offset, ctype); \ if (value != default_value) { \ upb_sink_put##upbtype(sink, sel, value); \ } \ diff --git a/ruby/ext/google/protobuf_c/extconf.rb b/ruby/ext/google/protobuf_c/extconf.rb new file mode 100644 index 0000000000..8d60392c4f --- /dev/null +++ b/ruby/ext/google/protobuf_c/extconf.rb @@ -0,0 +1,10 @@ +#!/usr/bin/ruby + +require 'mkmf' + +$CFLAGS += " -O3 -std=c99 -Wno-unused-function -DNDEBUG " + +$objs = ["protobuf.o", "defs.o", "storage.o", "message.o", + "repeated_field.o", "map.o", "encode_decode.o", "upb.o"] + +create_makefile("google/protobuf_c") diff --git a/ruby/ext/google/protobuf_c/map.c b/ruby/ext/google/protobuf_c/map.c new file mode 100644 index 0000000000..12e7a9d92f --- /dev/null +++ b/ruby/ext/google/protobuf_c/map.c @@ -0,0 +1,805 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include "protobuf.h" + +// ----------------------------------------------------------------------------- +// Basic map operations on top of upb's strtable. +// +// Note that we roll our own `Map` container here because, as for +// `RepeatedField`, we want a strongly-typed container. This is so that any user +// errors due to incorrect map key or value types are raised as close as +// possible to the error site, rather than at some deferred point (e.g., +// serialization). +// +// We build our `Map` on top of upb_strtable so that we're able to take +// advantage of the native_slot storage abstraction, as RepeatedField does. +// (This is not quite a perfect mapping -- see the key conversions below -- but +// gives us full support and error-checking for all value types for free.) +// ----------------------------------------------------------------------------- + +// Map values are stored using the native_slot abstraction (as with repeated +// field values), but keys are a bit special. Since we use a strtable, we need +// to store keys as sequences of bytes such that equality of those bytes maps +// one-to-one to equality of keys. We store strings directly (i.e., they map to +// their own bytes) and integers as native integers (using the native_slot +// abstraction). + +// Note that there is another tradeoff here in keeping string keys as native +// strings rather than Ruby strings: traversing the Map requires conversion to +// Ruby string values on every traversal, potentially creating more garbage. We +// should consider ways to cache a Ruby version of the key if this becomes an +// issue later. + +// Forms a key to use with the underlying strtable from a Ruby key value. |buf| +// must point to TABLE_KEY_BUF_LENGTH bytes of temporary space, used to +// construct a key byte sequence if needed. |out_key| and |out_length| provide +// the resulting key data/length. +#define TABLE_KEY_BUF_LENGTH 8 // sizeof(uint64_t) +static void table_key(Map* self, VALUE key, + char* buf, + const char** out_key, + size_t* out_length) { + switch (self->key_type) { + case UPB_TYPE_BYTES: + case UPB_TYPE_STRING: + // Strings: use string content directly. + Check_Type(key, T_STRING); + native_slot_validate_string_encoding(self->key_type, key); + *out_key = RSTRING_PTR(key); + *out_length = RSTRING_LEN(key); + break; + + case UPB_TYPE_BOOL: + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + native_slot_set(self->key_type, Qnil, buf, key); + *out_key = buf; + *out_length = native_slot_size(self->key_type); + break; + + default: + // Map constructor should not allow a Map with another key type to be + // constructed. + assert(false); + break; + } +} + +static VALUE table_key_to_ruby(Map* self, const char* buf, size_t length) { + switch (self->key_type) { + case UPB_TYPE_BYTES: + case UPB_TYPE_STRING: { + VALUE ret = rb_str_new(buf, length); + rb_enc_associate(ret, + (self->key_type == UPB_TYPE_BYTES) ? + kRubyString8bitEncoding : kRubyStringUtf8Encoding); + return ret; + } + + case UPB_TYPE_BOOL: + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + return native_slot_get(self->key_type, Qnil, buf); + + default: + assert(false); + return Qnil; + } +} + +static void* value_memory(upb_value* v) { + return (void*)(&v->val.uint64); +} + +// ----------------------------------------------------------------------------- +// Map container type. +// ----------------------------------------------------------------------------- + +const rb_data_type_t Map_type = { + "Google::Protobuf::Map", + { Map_mark, Map_free, NULL }, +}; + +VALUE cMap; + +Map* ruby_to_Map(VALUE _self) { + Map* self; + TypedData_Get_Struct(_self, Map, &Map_type, self); + return self; +} + +void Map_mark(void* _self) { + Map* self = _self; + + rb_gc_mark(self->value_type_class); + + if (self->value_type == UPB_TYPE_STRING || + self->value_type == UPB_TYPE_BYTES || + self->value_type == UPB_TYPE_MESSAGE) { + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + native_slot_mark(self->value_type, mem); + } + } +} + +void Map_free(void* _self) { + Map* self = _self; + upb_strtable_uninit(&self->table); + xfree(self); +} + +VALUE Map_alloc(VALUE klass) { + Map* self = ALLOC(Map); + memset(self, 0, sizeof(Map)); + self->value_type_class = Qnil; + VALUE ret = TypedData_Wrap_Struct(klass, &Map_type, self); + return ret; +} + +static bool needs_typeclass(upb_fieldtype_t type) { + switch (type) { + case UPB_TYPE_MESSAGE: + case UPB_TYPE_ENUM: + return true; + default: + return false; + } +} + +/* + * call-seq: + * Map.new(key_type, value_type, value_typeclass = nil, init_hashmap = {}) + * => new map + * + * Allocates a new Map container. This constructor may be called with 2, 3, or 4 + * arguments. The first two arguments are always present and are symbols (taking + * on the same values as field-type symbols in message descriptors) that + * indicate the type of the map key and value fields. + * + * The supported key types are: :int32, :int64, :uint32, :uint64, :bool, + * :string, :bytes. + * + * The supported value types are: :int32, :int64, :uint32, :uint64, :bool, + * :string, :bytes, :enum, :message. + * + * The third argument, value_typeclass, must be present if value_type is :enum + * or :message. As in RepeatedField#new, this argument must be a message class + * (for :message) or enum module (for :enum). + * + * The last argument, if present, provides initial content for map. Note that + * this may be an ordinary Ruby hashmap or another Map instance with identical + * key and value types. Also note that this argument may be present whether or + * not value_typeclass is present (and it is unambiguously separate from + * value_typeclass because value_typeclass's presence is strictly determined by + * value_type). The contents of this initial hashmap or Map instance are + * shallow-copied into the new Map: the original map is unmodified, but + * references to underlying objects will be shared if the value type is a + * message type. + */ +VALUE Map_init(int argc, VALUE* argv, VALUE _self) { + Map* self = ruby_to_Map(_self); + + // We take either two args (:key_type, :value_type), three args (:key_type, + // :value_type, "ValueMessageType"), or four args (the above plus an initial + // hashmap). + if (argc < 2 || argc > 4) { + rb_raise(rb_eArgError, "Map constructor expects 2, 3 or 4 arguments."); + } + + self->key_type = ruby_to_fieldtype(argv[0]); + self->value_type = ruby_to_fieldtype(argv[1]); + + // Check that the key type is an allowed type. + switch (self->key_type) { + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + case UPB_TYPE_BOOL: + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + // These are OK. + break; + default: + rb_raise(rb_eArgError, "Invalid key type for map."); + } + + int init_value_arg = 2; + if (needs_typeclass(self->value_type) && argc > 2) { + self->value_type_class = argv[2]; + validate_type_class(self->value_type, self->value_type_class); + init_value_arg = 3; + } + + // Table value type is always UINT64: this ensures enough space to store the + // native_slot value. + if (!upb_strtable_init(&self->table, UPB_CTYPE_UINT64)) { + rb_raise(rb_eRuntimeError, "Could not allocate table."); + } + + if (argc > init_value_arg) { + Map_merge_into_self(_self, argv[init_value_arg]); + } + + return Qnil; +} + +/* + * call-seq: + * Map.each(&block) + * + * Invokes &block on each |key, value| pair in the map, in unspecified order. + * Note that Map also includes Enumerable; map thus acts like a normal Ruby + * sequence. + */ +VALUE Map_each(VALUE _self) { + Map* self = ruby_to_Map(_self); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + rb_yield_values(2, key, value); + } + + return Qnil; +} + +/* + * call-seq: + * Map.keys => [list_of_keys] + * + * Returns the list of keys contained in the map, in unspecified order. + */ +VALUE Map_keys(VALUE _self) { + Map* self = ruby_to_Map(_self); + + VALUE ret = rb_ary_new(); + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + rb_ary_push(ret, key); + } + + return ret; +} + +/* + * call-seq: + * Map.values => [list_of_values] + * + * Returns the list of values contained in the map, in unspecified order. + */ +VALUE Map_values(VALUE _self) { + Map* self = ruby_to_Map(_self); + + VALUE ret = rb_ary_new(); + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + rb_ary_push(ret, value); + } + + return ret; +} + +/* + * call-seq: + * Map.[](key) => value + * + * Accesses the element at the given key. Throws an exception if the key type is + * incorrect. Returns nil when the key is not present in the map. + */ +VALUE Map_index(VALUE _self, VALUE key) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + table_key(self, key, keybuf, &keyval, &length); + + upb_value v; + if (upb_strtable_lookup2(&self->table, keyval, length, &v)) { + void* mem = value_memory(&v); + return native_slot_get(self->value_type, self->value_type_class, mem); + } else { + return Qnil; + } +} + +/* + * call-seq: + * Map.[]=(key, value) => value + * + * Inserts or overwrites the value at the given key with the given new value. + * Throws an exception if the key type is incorrect. Returns the new value that + * was just inserted. + */ +VALUE Map_index_set(VALUE _self, VALUE key, VALUE value) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + table_key(self, key, keybuf, &keyval, &length); + + upb_value v; + void* mem = value_memory(&v); + native_slot_set(self->value_type, self->value_type_class, mem, value); + + // Replace any existing value by issuing a 'remove' operation first. + upb_strtable_remove2(&self->table, keyval, length, NULL); + if (!upb_strtable_insert2(&self->table, keyval, length, v)) { + rb_raise(rb_eRuntimeError, "Could not insert into table"); + } + + // Ruby hashmap's :[]= method also returns the inserted value. + return value; +} + +/* + * call-seq: + * Map.has_key?(key) => bool + * + * Returns true if the given key is present in the map. Throws an exception if + * the key has the wrong type. + */ +VALUE Map_has_key(VALUE _self, VALUE key) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + table_key(self, key, keybuf, &keyval, &length); + + if (upb_strtable_lookup2(&self->table, keyval, length, NULL)) { + return Qtrue; + } else { + return Qfalse; + } +} + +/* + * call-seq: + * Map.delete(key) => old_value + * + * Deletes the value at the given key, if any, returning either the old value or + * nil if none was present. Throws an exception if the key is of the wrong type. + */ +VALUE Map_delete(VALUE _self, VALUE key) { + Map* self = ruby_to_Map(_self); + + char keybuf[TABLE_KEY_BUF_LENGTH]; + const char* keyval = NULL; + size_t length = 0; + table_key(self, key, keybuf, &keyval, &length); + + upb_value v; + if (upb_strtable_remove2(&self->table, keyval, length, &v)) { + void* mem = value_memory(&v); + return native_slot_get(self->value_type, self->value_type_class, mem); + } else { + return Qnil; + } +} + +/* + * call-seq: + * Map.clear + * + * Removes all entries from the map. + */ +VALUE Map_clear(VALUE _self) { + Map* self = ruby_to_Map(_self); + + // Uninit and reinit the table -- this is faster than iterating and doing a + // delete-lookup on each key. + upb_strtable_uninit(&self->table); + if (!upb_strtable_init(&self->table, UPB_CTYPE_INT64)) { + rb_raise(rb_eRuntimeError, "Unable to re-initialize table"); + } + return Qnil; +} + +/* + * call-seq: + * Map.length + * + * Returns the number of entries (key-value pairs) in the map. + */ +VALUE Map_length(VALUE _self) { + Map* self = ruby_to_Map(_self); + return ULL2NUM(upb_strtable_count(&self->table)); +} + +static VALUE Map_new_this_type(VALUE _self) { + Map* self = ruby_to_Map(_self); + VALUE new_map = Qnil; + VALUE key_type = fieldtype_to_ruby(self->key_type); + VALUE value_type = fieldtype_to_ruby(self->value_type); + if (self->value_type_class != Qnil) { + new_map = rb_funcall(CLASS_OF(_self), rb_intern("new"), 3, + key_type, value_type, self->value_type_class); + } else { + new_map = rb_funcall(CLASS_OF(_self), rb_intern("new"), 2, + key_type, value_type); + } + return new_map; +} + +/* + * call-seq: + * Map.dup => new_map + * + * Duplicates this map with a shallow copy. References to all non-primitive + * element objects (e.g., submessages) are shared. + */ +VALUE Map_dup(VALUE _self) { + Map* self = ruby_to_Map(_self); + VALUE new_map = Map_new_this_type(_self); + Map* new_self = ruby_to_Map(new_map); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + upb_value dup; + void* dup_mem = value_memory(&dup); + native_slot_dup(self->value_type, dup_mem, mem); + + if (!upb_strtable_insert2(&new_self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + dup)) { + rb_raise(rb_eRuntimeError, "Error inserting value into new table"); + } + } + + return new_map; +} + +// Used by Google::Protobuf.deep_copy but not exposed directly. +VALUE Map_deep_copy(VALUE _self) { + Map* self = ruby_to_Map(_self); + VALUE new_map = Map_new_this_type(_self); + Map* new_self = ruby_to_Map(new_map); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + upb_value dup; + void* dup_mem = value_memory(&dup); + native_slot_deep_copy(self->value_type, dup_mem, mem); + + if (!upb_strtable_insert2(&new_self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + dup)) { + rb_raise(rb_eRuntimeError, "Error inserting value into new table"); + } + } + + return new_map; +} + +/* + * call-seq: + * Map.==(other) => boolean + * + * Compares this map to another. Maps are equal if they have identical key sets, + * and for each key, the values in both maps compare equal. Elements are + * compared as per normal Ruby semantics, by calling their :== methods (or + * performing a more efficient comparison for primitive types). + * + * Maps with dissimilar key types or value types/typeclasses are never equal, + * even if value comparison (for example, between integers and floats) would + * have otherwise indicated that every element has equal value. + */ +VALUE Map_eq(VALUE _self, VALUE _other) { + Map* self = ruby_to_Map(_self); + + // Allow comparisons to Ruby hashmaps by converting to a temporary Map + // instance. Slow, but workable. + if (TYPE(_other) == T_HASH) { + VALUE other_map = Map_new_this_type(_self); + Map_merge_into_self(other_map, _other); + _other = other_map; + } + + Map* other = ruby_to_Map(_other); + + if (self == other) { + return Qtrue; + } + if (self->key_type != other->key_type || + self->value_type != other->value_type || + self->value_type_class != other->value_type_class) { + return Qfalse; + } + if (upb_strtable_count(&self->table) != upb_strtable_count(&other->table)) { + return Qfalse; + } + + // For each member of self, check that an equal member exists at the same key + // in other. + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + upb_value other_v; + void* other_mem = value_memory(&other_v); + + if (!upb_strtable_lookup2(&other->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + &other_v)) { + // Not present in other map. + return Qfalse; + } + + if (!native_slot_eq(self->value_type, mem, other_mem)) { + // Present, but value not equal. + return Qfalse; + } + } + + return Qtrue; +} + +/* + * call-seq: + * Map.hash => hash_value + * + * Returns a hash value based on this map's contents. + */ +VALUE Map_hash(VALUE _self) { + Map* self = ruby_to_Map(_self); + + st_index_t h = rb_hash_start(0); + VALUE hash_sym = rb_intern("hash"); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + h = rb_hash_uint(h, NUM2LONG(rb_funcall(key, hash_sym, 0))); + h = rb_hash_uint(h, NUM2LONG(rb_funcall(value, hash_sym, 0))); + } + + return INT2FIX(h); +} + +/* + * call-seq: + * Map.inspect => string + * + * Returns a string representing this map's elements. It will be formatted as + * "{key => value, key => value, ...}", with each key and value string + * representation computed by its own #inspect method. + */ +VALUE Map_inspect(VALUE _self) { + Map* self = ruby_to_Map(_self); + + VALUE str = rb_str_new2("{"); + + bool first = true; + VALUE inspect_sym = rb_intern("inspect"); + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &self->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + VALUE key = table_key_to_ruby( + self, upb_strtable_iter_key(&it), upb_strtable_iter_keylength(&it)); + + upb_value v = upb_strtable_iter_value(&it); + void* mem = value_memory(&v); + VALUE value = native_slot_get(self->value_type, + self->value_type_class, + mem); + + if (!first) { + str = rb_str_cat2(str, ", "); + } else { + first = false; + } + str = rb_str_append(str, rb_funcall(key, inspect_sym, 0)); + str = rb_str_cat2(str, "=>"); + str = rb_str_append(str, rb_funcall(value, inspect_sym, 0)); + } + + str = rb_str_cat2(str, "}"); + return str; +} + +/* + * call-seq: + * Map.merge(other_map) => map + * + * Copies key/value pairs from other_map into a copy of this map. If a key is + * set in other_map and this map, the value from other_map overwrites the value + * in the new copy of this map. Returns the new copy of this map with merged + * contents. + */ +VALUE Map_merge(VALUE _self, VALUE hashmap) { + VALUE dupped = Map_dup(_self); + return Map_merge_into_self(dupped, hashmap); +} + +static int merge_into_self_callback(VALUE key, VALUE value, VALUE self) { + Map_index_set(self, key, value); + return ST_CONTINUE; +} + +// Used only internally -- shared by #merge and #initialize. +VALUE Map_merge_into_self(VALUE _self, VALUE hashmap) { + if (TYPE(hashmap) == T_HASH) { + rb_hash_foreach(hashmap, merge_into_self_callback, _self); + } else if (RB_TYPE_P(hashmap, T_DATA) && RTYPEDDATA_P(hashmap) && + RTYPEDDATA_TYPE(hashmap) == &Map_type) { + + Map* self = ruby_to_Map(_self); + Map* other = ruby_to_Map(hashmap); + + if (self->key_type != other->key_type || + self->value_type != other->value_type || + self->value_type_class != other->value_type_class) { + rb_raise(rb_eArgError, "Attempt to merge Map with mismatching types"); + } + + upb_strtable_iter it; + for (upb_strtable_begin(&it, &other->table); + !upb_strtable_done(&it); + upb_strtable_next(&it)) { + + // Replace any existing value by issuing a 'remove' operation first. + upb_value oldv; + upb_strtable_remove2(&self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + &oldv); + + upb_value v = upb_strtable_iter_value(&it); + upb_strtable_insert2(&self->table, + upb_strtable_iter_key(&it), + upb_strtable_iter_keylength(&it), + v); + } + } else { + rb_raise(rb_eArgError, "Unknown type merging into Map"); + } + return _self; +} + +// Internal method: map iterator initialization (used for serialization). +void Map_begin(VALUE _self, Map_iter* iter) { + Map* self = ruby_to_Map(_self); + iter->self = self; + upb_strtable_begin(&iter->it, &self->table); +} + +void Map_next(Map_iter* iter) { + upb_strtable_next(&iter->it); +} + +bool Map_done(Map_iter* iter) { + return upb_strtable_done(&iter->it); +} + +VALUE Map_iter_key(Map_iter* iter) { + return table_key_to_ruby( + iter->self, + upb_strtable_iter_key(&iter->it), + upb_strtable_iter_keylength(&iter->it)); +} + +VALUE Map_iter_value(Map_iter* iter) { + upb_value v = upb_strtable_iter_value(&iter->it); + void* mem = value_memory(&v); + return native_slot_get(iter->self->value_type, + iter->self->value_type_class, + mem); +} + +void Map_register(VALUE module) { + VALUE klass = rb_define_class_under(module, "Map", rb_cObject); + rb_define_alloc_func(klass, Map_alloc); + cMap = klass; + rb_gc_register_address(&cMap); + + rb_define_method(klass, "initialize", Map_init, -1); + rb_define_method(klass, "each", Map_each, 0); + rb_define_method(klass, "keys", Map_keys, 0); + rb_define_method(klass, "values", Map_values, 0); + rb_define_method(klass, "[]", Map_index, 1); + rb_define_method(klass, "[]=", Map_index_set, 2); + rb_define_method(klass, "has_key?", Map_has_key, 1); + rb_define_method(klass, "delete", Map_delete, 1); + rb_define_method(klass, "clear", Map_clear, 0); + rb_define_method(klass, "length", Map_length, 0); + rb_define_method(klass, "dup", Map_dup, 0); + rb_define_method(klass, "==", Map_eq, 1); + rb_define_method(klass, "hash", Map_hash, 0); + rb_define_method(klass, "inspect", Map_inspect, 0); + rb_define_method(klass, "merge", Map_merge, 1); + rb_include_module(klass, rb_mEnumerable); +} diff --git a/ruby/ext/protobuf_c/message.c b/ruby/ext/google/protobuf_c/message.c similarity index 88% rename from ruby/ext/protobuf_c/message.c rename to ruby/ext/google/protobuf_c/message.c index 105b780739..7e58a6178b 100644 --- a/ruby/ext/protobuf_c/message.c +++ b/ruby/ext/google/protobuf_c/message.c @@ -70,6 +70,35 @@ VALUE Message_alloc(VALUE klass) { return ret; } +static VALUE which_oneof_field(MessageHeader* self, const upb_oneofdef* o) { + // If no fields in the oneof, always nil. + if (upb_oneofdef_numfields(o) == 0) { + return Qnil; + } + // Grab the first field in the oneof so we can get its layout info to find the + // oneof_case field. + upb_oneof_iter it; + upb_oneof_begin(&it, o); + assert(!upb_oneof_done(&it)); + const upb_fielddef* first_field = upb_oneof_iter_field(&it); + assert(upb_fielddef_containingoneof(first_field) != NULL); + + size_t case_ofs = + self->descriptor->layout-> + fields[upb_fielddef_index(first_field)].case_offset; + uint32_t oneof_case = *((uint32_t*)(Message_data(self) + case_ofs)); + + if (oneof_case == ONEOF_CASE_NONE) { + return Qnil; + } + + // oneof_case is a field index, so find that field. + const upb_fielddef* f = upb_oneofdef_itof(o, oneof_case); + assert(f != NULL); + + return ID2SYM(rb_intern(upb_fielddef_name(f))); +} + /* * call-seq: * Message.method_missing(*args) @@ -82,6 +111,10 @@ VALUE Message_alloc(VALUE klass) { * * msg.foo = 42 * puts msg.foo + * + * This method also provides read-only accessors for oneofs. If a oneof exists + * with name 'my_oneof', then msg.my_oneof will return a Ruby symbol equal to + * the name of the field in that oneof that is currently set, or nil if none. */ VALUE Message_method_missing(int argc, VALUE* argv, VALUE _self) { MessageHeader* self; @@ -104,6 +137,17 @@ VALUE Message_method_missing(int argc, VALUE* argv, VALUE _self) { name_len--; } + // Check for a oneof name first. + const upb_oneofdef* o = upb_msgdef_ntoo(self->descriptor->msgdef, + name, name_len); + if (o != NULL) { + if (setter) { + rb_raise(rb_eRuntimeError, "Oneof accessors are read-only."); + } + return which_oneof_field(self, o); + } + + // Otherwise, check for a field with that name. const upb_fielddef* f = upb_msgdef_ntof(self->descriptor->msgdef, name, name_len); @@ -139,7 +183,14 @@ int Message_initialize_kwarg(VALUE key, VALUE val, VALUE _self) { "Unknown field name in initialization map entry."); } - if (upb_fielddef_label(f) == UPB_LABEL_REPEATED) { + if (is_map_field(f)) { + if (TYPE(val) != T_HASH) { + rb_raise(rb_eArgError, + "Expected Hash object as initializer value for map field."); + } + VALUE map = layout_get(self->descriptor->layout, Message_data(self), f); + Map_merge_into_self(map, val); + } else if (upb_fielddef_label(f) == UPB_LABEL_REPEATED) { if (TYPE(val) != T_ARRAY) { rb_raise(rb_eArgError, "Expected array as initializer value for repeated field."); @@ -450,13 +501,15 @@ VALUE build_module_from_enumdesc(EnumDescriptor* enumdesc) { * call-seq: * Google::Protobuf.deep_copy(obj) => copy_of_obj * - * Performs a deep copy of either a RepeatedField instance or a message object, - * recursively copying its members. + * Performs a deep copy of a RepeatedField instance, a Map instance, or a + * message object, recursively copying its members. */ VALUE Google_Protobuf_deep_copy(VALUE self, VALUE obj) { VALUE klass = CLASS_OF(obj); if (klass == cRepeatedField) { return RepeatedField_deep_copy(obj); + } else if (klass == cMap) { + return Map_deep_copy(obj); } else { return Message_deep_copy(obj); } diff --git a/ruby/ext/protobuf_c/protobuf.c b/ruby/ext/google/protobuf_c/protobuf.c similarity index 97% rename from ruby/ext/protobuf_c/protobuf.c rename to ruby/ext/google/protobuf_c/protobuf.c index d5862284d5..d2d3503386 100644 --- a/ruby/ext/protobuf_c/protobuf.c +++ b/ruby/ext/google/protobuf_c/protobuf.c @@ -77,11 +77,14 @@ void Init_protobuf_c() { DescriptorPool_register(protobuf); Descriptor_register(protobuf); FieldDescriptor_register(protobuf); + OneofDescriptor_register(protobuf); EnumDescriptor_register(protobuf); MessageBuilderContext_register(internal); + OneofBuilderContext_register(internal); EnumBuilderContext_register(internal); Builder_register(internal); RepeatedField_register(protobuf); + Map_register(protobuf); rb_define_singleton_method(protobuf, "encode", Google_Protobuf_encode, 1); rb_define_singleton_method(protobuf, "decode", Google_Protobuf_decode, 2); diff --git a/ruby/ext/protobuf_c/protobuf.h b/ruby/ext/google/protobuf_c/protobuf.h similarity index 75% rename from ruby/ext/protobuf_c/protobuf.h rename to ruby/ext/google/protobuf_c/protobuf.h index a7f6f5397d..d8a327aa7d 100644 --- a/ruby/ext/protobuf_c/protobuf.h +++ b/ruby/ext/google/protobuf_c/protobuf.h @@ -35,15 +35,7 @@ #include #include -#include "upb/def.h" -#include "upb/handlers.h" -#include "upb/pb/decoder.h" -#include "upb/pb/encoder.h" -#include "upb/pb/glue.h" -#include "upb/json/parser.h" -#include "upb/json/printer.h" -#include "upb/shim/shim.h" -#include "upb/symtab.h" +#include "upb.h" // Forward decls. struct DescriptorPool; @@ -51,6 +43,7 @@ struct Descriptor; struct FieldDescriptor; struct EnumDescriptor; struct MessageLayout; +struct MessageField; struct MessageHeader; struct MessageBuilderContext; struct EnumBuilderContext; @@ -59,10 +52,13 @@ struct Builder; typedef struct DescriptorPool DescriptorPool; typedef struct Descriptor Descriptor; typedef struct FieldDescriptor FieldDescriptor; +typedef struct OneofDescriptor OneofDescriptor; typedef struct EnumDescriptor EnumDescriptor; typedef struct MessageLayout MessageLayout; +typedef struct MessageField MessageField; typedef struct MessageHeader MessageHeader; typedef struct MessageBuilderContext MessageBuilderContext; +typedef struct OneofBuilderContext OneofBuilderContext; typedef struct EnumBuilderContext EnumBuilderContext; typedef struct Builder Builder; @@ -118,12 +114,20 @@ struct Descriptor { const upb_pbdecodermethod* fill_method; const upb_handlers* pb_serialize_handlers; const upb_handlers* json_serialize_handlers; + // Handlers hold type class references for sub-message fields directly in some + // cases. We need to keep these rooted because they might otherwise be + // collected. + VALUE typeclass_references; }; struct FieldDescriptor { const upb_fielddef* fielddef; }; +struct OneofDescriptor { + const upb_oneofdef* oneofdef; +}; + struct EnumDescriptor { const upb_enumdef* enumdef; VALUE module; // begins as nil @@ -131,6 +135,12 @@ struct EnumDescriptor { struct MessageBuilderContext { VALUE descriptor; + VALUE builder; +}; + +struct OneofBuilderContext { + VALUE descriptor; + VALUE builder; }; struct EnumBuilderContext { @@ -147,6 +157,7 @@ extern VALUE cDescriptor; extern VALUE cFieldDescriptor; extern VALUE cEnumDescriptor; extern VALUE cMessageBuilderContext; +extern VALUE cOneofBuilderContext; extern VALUE cEnumBuilderContext; extern VALUE cBuilder; @@ -178,6 +189,9 @@ VALUE Descriptor_name_set(VALUE _self, VALUE str); VALUE Descriptor_each(VALUE _self); VALUE Descriptor_lookup(VALUE _self, VALUE name); VALUE Descriptor_add_field(VALUE _self, VALUE obj); +VALUE Descriptor_add_oneof(VALUE _self, VALUE obj); +VALUE Descriptor_each_oneof(VALUE _self); +VALUE Descriptor_lookup_oneof(VALUE _self, VALUE name); VALUE Descriptor_msgclass(VALUE _self); extern const rb_data_type_t _Descriptor_type; @@ -202,6 +216,16 @@ VALUE FieldDescriptor_set(VALUE _self, VALUE msg_rb, VALUE value); upb_fieldtype_t ruby_to_fieldtype(VALUE type); VALUE fieldtype_to_ruby(upb_fieldtype_t type); +void OneofDescriptor_mark(void* _self); +void OneofDescriptor_free(void* _self); +VALUE OneofDescriptor_alloc(VALUE klass); +void OneofDescriptor_register(VALUE module); +OneofDescriptor* ruby_to_OneofDescriptor(VALUE value); +VALUE OneofDescriptor_name(VALUE _self); +VALUE OneofDescriptor_name_set(VALUE _self, VALUE value); +VALUE OneofDescriptor_add_field(VALUE _self, VALUE field); +VALUE OneofDescriptor_each(VALUE _self, VALUE field); + void EnumDescriptor_mark(void* _self); void EnumDescriptor_free(void* _self); VALUE EnumDescriptor_alloc(VALUE klass); @@ -221,10 +245,24 @@ void MessageBuilderContext_free(void* _self); VALUE MessageBuilderContext_alloc(VALUE klass); void MessageBuilderContext_register(VALUE module); MessageBuilderContext* ruby_to_MessageBuilderContext(VALUE value); -VALUE MessageBuilderContext_initialize(VALUE _self, VALUE descriptor); +VALUE MessageBuilderContext_initialize(VALUE _self, + VALUE descriptor, + VALUE builder); VALUE MessageBuilderContext_optional(int argc, VALUE* argv, VALUE _self); VALUE MessageBuilderContext_required(int argc, VALUE* argv, VALUE _self); VALUE MessageBuilderContext_repeated(int argc, VALUE* argv, VALUE _self); +VALUE MessageBuilderContext_map(int argc, VALUE* argv, VALUE _self); +VALUE MessageBuilderContext_oneof(VALUE _self, VALUE name); + +void OneofBuilderContext_mark(void* _self); +void OneofBuilderContext_free(void* _self); +VALUE OneofBuilderContext_alloc(VALUE klass); +void OneofBuilderContext_register(VALUE module); +OneofBuilderContext* ruby_to_OneofBuilderContext(VALUE value); +VALUE OneofBuilderContext_initialize(VALUE _self, + VALUE descriptor, + VALUE builder); +VALUE OneofBuilderContext_optional(int argc, VALUE* argv, VALUE _self); void EnumBuilderContext_mark(void* _self); void EnumBuilderContext_free(void* _self); @@ -247,14 +285,25 @@ VALUE Builder_finalize_to_pool(VALUE _self, VALUE pool_rb); // Native slot storage abstraction. // ----------------------------------------------------------------------------- +#define NATIVE_SLOT_MAX_SIZE sizeof(uint64_t) + size_t native_slot_size(upb_fieldtype_t type); void native_slot_set(upb_fieldtype_t type, VALUE type_class, void* memory, VALUE value); +// Atomically (with respect to Ruby VM calls) either update the value and set a +// oneof case, or do neither. If |case_memory| is null, then no case value is +// set. +void native_slot_set_value_and_case(upb_fieldtype_t type, + VALUE type_class, + void* memory, + VALUE value, + uint32_t* case_memory, + uint32_t case_number); VALUE native_slot_get(upb_fieldtype_t type, VALUE type_class, - void* memory); + const void* memory); void native_slot_init(upb_fieldtype_t type, void* memory); void native_slot_mark(upb_fieldtype_t type, void* memory); void native_slot_dup(upb_fieldtype_t type, void* to, void* from); @@ -262,11 +311,32 @@ void native_slot_deep_copy(upb_fieldtype_t type, void* to, void* from); bool native_slot_eq(upb_fieldtype_t type, void* mem1, void* mem2); void native_slot_validate_string_encoding(upb_fieldtype_t type, VALUE value); +void native_slot_check_int_range_precision(upb_fieldtype_t type, VALUE value); extern rb_encoding* kRubyStringUtf8Encoding; extern rb_encoding* kRubyStringASCIIEncoding; extern rb_encoding* kRubyString8bitEncoding; +VALUE field_type_class(const upb_fielddef* field); + +#define MAP_KEY_FIELD 1 +#define MAP_VALUE_FIELD 2 + +// Oneof case slot value to indicate that no oneof case is set. The value `0` is +// safe because field numbers are used as case identifiers, and no field can +// have a number of 0. +#define ONEOF_CASE_NONE 0 + +// These operate on a map field (i.e., a repeated field of submessages whose +// submessage type is a map-entry msgdef). +bool is_map_field(const upb_fielddef* field); +const upb_fielddef* map_field_key(const upb_fielddef* field); +const upb_fielddef* map_field_value(const upb_fielddef* field); + +// These operate on a map-entry msgdef. +const upb_fielddef* map_entry_key(const upb_msgdef* msgdef); +const upb_fielddef* map_entry_value(const upb_msgdef* msgdef); + // ----------------------------------------------------------------------------- // Repeated field container type. // ----------------------------------------------------------------------------- @@ -290,7 +360,6 @@ extern VALUE cRepeatedField; RepeatedField* ruby_to_RepeatedField(VALUE value); -void RepeatedField_register(VALUE module); VALUE RepeatedField_each(VALUE _self); VALUE RepeatedField_index(VALUE _self, VALUE _index); void* RepeatedField_index_native(VALUE _self, int index); @@ -310,20 +379,80 @@ VALUE RepeatedField_hash(VALUE _self); VALUE RepeatedField_inspect(VALUE _self); VALUE RepeatedField_plus(VALUE _self, VALUE list); +// Defined in repeated_field.c; also used by Map. +void validate_type_class(upb_fieldtype_t type, VALUE klass); + +// ----------------------------------------------------------------------------- +// Map container type. +// ----------------------------------------------------------------------------- + +typedef struct { + upb_fieldtype_t key_type; + upb_fieldtype_t value_type; + VALUE value_type_class; + upb_strtable table; +} Map; + +void Map_mark(void* self); +void Map_free(void* self); +VALUE Map_alloc(VALUE klass); +VALUE Map_init(int argc, VALUE* argv, VALUE self); +void Map_register(VALUE module); + +extern const rb_data_type_t Map_type; +extern VALUE cMap; + +Map* ruby_to_Map(VALUE value); + +VALUE Map_each(VALUE _self); +VALUE Map_keys(VALUE _self); +VALUE Map_values(VALUE _self); +VALUE Map_index(VALUE _self, VALUE key); +VALUE Map_index_set(VALUE _self, VALUE key, VALUE value); +VALUE Map_has_key(VALUE _self, VALUE key); +VALUE Map_delete(VALUE _self, VALUE key); +VALUE Map_clear(VALUE _self); +VALUE Map_length(VALUE _self); +VALUE Map_dup(VALUE _self); +VALUE Map_deep_copy(VALUE _self); +VALUE Map_eq(VALUE _self, VALUE _other); +VALUE Map_hash(VALUE _self); +VALUE Map_inspect(VALUE _self); +VALUE Map_merge(VALUE _self, VALUE hashmap); +VALUE Map_merge_into_self(VALUE _self, VALUE hashmap); + +typedef struct { + Map* self; + upb_strtable_iter it; +} Map_iter; + +void Map_begin(VALUE _self, Map_iter* iter); +void Map_next(Map_iter* iter); +bool Map_done(Map_iter* iter); +VALUE Map_iter_key(Map_iter* iter); +VALUE Map_iter_value(Map_iter* iter); + // ----------------------------------------------------------------------------- // Message layout / storage. // ----------------------------------------------------------------------------- +#define MESSAGE_FIELD_NO_CASE ((size_t)-1) + +struct MessageField { + size_t offset; + size_t case_offset; // for oneofs, a uint32. Else, MESSAGE_FIELD_NO_CASE. +}; + struct MessageLayout { const upb_msgdef* msgdef; - size_t* offsets; + MessageField* fields; size_t size; }; MessageLayout* create_layout(const upb_msgdef* msgdef); void free_layout(MessageLayout* layout); VALUE layout_get(MessageLayout* layout, - void* storage, + const void* storage, const upb_fielddef* field); void layout_set(MessageLayout* layout, void* storage, diff --git a/ruby/ext/protobuf_c/repeated_field.c b/ruby/ext/google/protobuf_c/repeated_field.c similarity index 95% rename from ruby/ext/protobuf_c/repeated_field.c rename to ruby/ext/google/protobuf_c/repeated_field.c index 6bd13b0799..8cf2e29b58 100644 --- a/ruby/ext/protobuf_c/repeated_field.c +++ b/ruby/ext/google/protobuf_c/repeated_field.c @@ -316,6 +316,29 @@ VALUE RepeatedField_deep_copy(VALUE _self) { return new_rptfield; } +/* + * call-seq: + * RepeatedField.to_ary => array + * + * Used when converted implicitly into array, e.g. compared to an Array. + * Also called as a fallback of Object#to_a + */ +VALUE RepeatedField_to_ary(VALUE _self) { + RepeatedField* self = ruby_to_RepeatedField(_self); + upb_fieldtype_t field_type = self->field_type; + + size_t elem_size = native_slot_size(field_type); + size_t off = 0; + VALUE ary = rb_ary_new2(self->size); + for (int i = 0; i < self->size; i++, off += elem_size) { + void* mem = ((uint8_t *)self->elements) + off; + VALUE elem = native_slot_get(field_type, self->field_type_class, mem); + + rb_ary_push(ary, elem); + } + return ary; +} + /* * call-seq: * RepeatedField.==(other) => boolean @@ -324,6 +347,10 @@ VALUE RepeatedField_deep_copy(VALUE _self) { * element types are equal, their lengths are equal, and each element is equal. * Elements are compared as per normal Ruby semantics, by calling their :== * methods (or performing a more efficient comparison for primitive types). + * + * Repeated fields with dissimilar element types are never equal, even if value + * comparison (for example, between integers and floats) would have otherwise + * indicated that every element has equal value. */ VALUE RepeatedField_eq(VALUE _self, VALUE _other) { if (_self == _other) { @@ -331,15 +358,9 @@ VALUE RepeatedField_eq(VALUE _self, VALUE _other) { } RepeatedField* self = ruby_to_RepeatedField(_self); - // Inefficient but workable: to support comparison to a generic array, we - // build a temporary RepeatedField of our type. if (TYPE(_other) == T_ARRAY) { - VALUE new_rptfield = RepeatedField_new_this_type(_self); - for (int i = 0; i < RARRAY_LEN(_other); i++) { - VALUE elem = rb_ary_entry(_other, i); - RepeatedField_push(new_rptfield, elem); - } - _other = new_rptfield; + VALUE self_ary = RepeatedField_to_ary(_self); + return rb_equal(self_ary, _other); } RepeatedField* other = ruby_to_RepeatedField(_other); @@ -397,29 +418,8 @@ VALUE RepeatedField_hash(VALUE _self) { * representation computed by its own #inspect method. */ VALUE RepeatedField_inspect(VALUE _self) { - RepeatedField* self = ruby_to_RepeatedField(_self); - - VALUE str = rb_str_new2("["); - - bool first = true; - - upb_fieldtype_t field_type = self->field_type; - VALUE field_type_class = self->field_type_class; - size_t elem_size = native_slot_size(field_type); - size_t off = 0; - for (int i = 0; i < self->size; i++, off += elem_size) { - void* mem = ((uint8_t *)self->elements) + off; - VALUE elem = native_slot_get(field_type, field_type_class, mem); - if (!first) { - str = rb_str_cat2(str, ", "); - } else { - first = false; - } - str = rb_str_append(str, rb_funcall(elem, rb_intern("inspect"), 0)); - } - - str = rb_str_cat2(str, "]"); - return str; + VALUE self_ary = RepeatedField_to_ary(_self); + return rb_funcall(self_ary, rb_intern("inspect"), 0); } /* @@ -458,7 +458,7 @@ VALUE RepeatedField_plus(VALUE _self, VALUE list) { return dupped; } -static void validate_type_class(upb_fieldtype_t type, VALUE klass) { +void validate_type_class(upb_fieldtype_t type, VALUE klass) { if (rb_iv_get(klass, kDescriptorInstanceVar) == Qnil) { rb_raise(rb_eArgError, "Type class has no descriptor. Please pass a " @@ -590,6 +590,7 @@ void RepeatedField_register(VALUE module) { // Also define #clone so that we don't inherit Object#clone. rb_define_method(klass, "clone", RepeatedField_dup, 0); rb_define_method(klass, "==", RepeatedField_eq, 1); + rb_define_method(klass, "to_ary", RepeatedField_to_ary, 0); rb_define_method(klass, "hash", RepeatedField_hash, 0); rb_define_method(klass, "inspect", RepeatedField_inspect, 0); rb_define_method(klass, "+", RepeatedField_plus, 1); diff --git a/ruby/ext/protobuf_c/storage.c b/ruby/ext/google/protobuf_c/storage.c similarity index 50% rename from ruby/ext/protobuf_c/storage.c rename to ruby/ext/google/protobuf_c/storage.c index c4d801af34..5b1549d23a 100644 --- a/ruby/ext/protobuf_c/storage.c +++ b/ruby/ext/google/protobuf_c/storage.c @@ -57,7 +57,17 @@ size_t native_slot_size(upb_fieldtype_t type) { } } -static void check_int_range_precision(upb_fieldtype_t type, VALUE val) { +static bool is_ruby_num(VALUE value) { + return (TYPE(value) == T_FLOAT || + TYPE(value) == T_FIXNUM || + TYPE(value) == T_BIGNUM); +} + +void native_slot_check_int_range_precision(upb_fieldtype_t type, VALUE val) { + if (!is_ruby_num(val)) { + rb_raise(rb_eTypeError, "Expected number type for integral field."); + } + // NUM2{INT,UINT,LL,ULL} macros do the appropriate range checks on upper // bound; we just need to do precision checks (i.e., disallow rounding) and // check for < 0 on unsigned types. @@ -76,12 +86,6 @@ static void check_int_range_precision(upb_fieldtype_t type, VALUE val) { } } -static bool is_ruby_num(VALUE value) { - return (TYPE(value) == T_FLOAT || - TYPE(value) == T_FIXNUM || - TYPE(value) == T_BIGNUM); -} - void native_slot_validate_string_encoding(upb_fieldtype_t type, VALUE value) { bool bad_encoding = false; rb_encoding* string_encoding = rb_enc_from_index(ENCODING_GET(value)); @@ -105,6 +109,17 @@ void native_slot_validate_string_encoding(upb_fieldtype_t type, VALUE value) { void native_slot_set(upb_fieldtype_t type, VALUE type_class, void* memory, VALUE value) { + native_slot_set_value_and_case(type, type_class, memory, value, NULL, 0); +} + +void native_slot_set_value_and_case(upb_fieldtype_t type, VALUE type_class, + void* memory, VALUE value, + uint32_t* case_memory, + uint32_t case_number) { + // Note that in order to atomically change the value in memory and the case + // value (w.r.t. Ruby VM calls), we must set the value at |memory| only after + // all Ruby VM calls are complete. The case is then set at the bottom of this + // function. switch (type) { case UPB_TYPE_FLOAT: if (!is_ruby_num(value)) { @@ -156,14 +171,14 @@ void native_slot_set(upb_fieldtype_t type, VALUE type_class, int32_t int_val = 0; if (TYPE(value) == T_SYMBOL) { // Ensure that the given symbol exists in the enum module. - VALUE lookup = rb_const_get(type_class, SYM2ID(value)); + VALUE lookup = rb_funcall(type_class, rb_intern("resolve"), 1, value); if (lookup == Qnil) { rb_raise(rb_eRangeError, "Unknown symbol value for enum field."); } else { int_val = NUM2INT(lookup); } } else { - check_int_range_precision(UPB_TYPE_INT32, value); + native_slot_check_int_range_precision(UPB_TYPE_INT32, value); int_val = NUM2INT(value); } DEREF(memory, int32_t) = int_val; @@ -173,10 +188,7 @@ void native_slot_set(upb_fieldtype_t type, VALUE type_class, case UPB_TYPE_INT64: case UPB_TYPE_UINT32: case UPB_TYPE_UINT64: - if (!is_ruby_num(value)) { - rb_raise(rb_eTypeError, "Expected number type for integral field."); - } - check_int_range_precision(type, value); + native_slot_check_int_range_precision(type, value); switch (type) { case UPB_TYPE_INT32: DEREF(memory, int32_t) = NUM2INT(value); @@ -197,9 +209,15 @@ void native_slot_set(upb_fieldtype_t type, VALUE type_class, default: break; } + + if (case_memory != NULL) { + *case_memory = case_number; + } } -VALUE native_slot_get(upb_fieldtype_t type, VALUE type_class, void* memory) { +VALUE native_slot_get(upb_fieldtype_t type, + VALUE type_class, + const void* memory) { switch (type) { case UPB_TYPE_FLOAT: return DBL2NUM(DEREF(memory, float)); @@ -210,7 +228,7 @@ VALUE native_slot_get(upb_fieldtype_t type, VALUE type_class, void* memory) { case UPB_TYPE_STRING: case UPB_TYPE_BYTES: case UPB_TYPE_MESSAGE: - return *((VALUE *)memory); + return DEREF(memory, VALUE); case UPB_TYPE_ENUM: { int32_t val = DEREF(memory, int32_t); VALUE symbol = enum_lookup(type_class, INT2NUM(val)); @@ -246,8 +264,9 @@ void native_slot_init(upb_fieldtype_t type, void* memory) { break; case UPB_TYPE_STRING: case UPB_TYPE_BYTES: - // TODO(cfallin): set encoding appropriately DEREF(memory, VALUE) = rb_str_new2(""); + rb_enc_associate(DEREF(memory, VALUE), (type == UPB_TYPE_BYTES) ? + kRubyString8bitEncoding : kRubyStringUtf8Encoding); break; case UPB_TYPE_MESSAGE: DEREF(memory, VALUE) = Qnil; @@ -321,25 +340,135 @@ bool native_slot_eq(upb_fieldtype_t type, void* mem1, void* mem2) { } } +// ----------------------------------------------------------------------------- +// Map field utilities. +// ----------------------------------------------------------------------------- + +bool is_map_field(const upb_fielddef* field) { + if (upb_fielddef_label(field) != UPB_LABEL_REPEATED || + upb_fielddef_type(field) != UPB_TYPE_MESSAGE) { + return false; + } + const upb_msgdef* subdef = upb_fielddef_msgsubdef(field); + return upb_msgdef_mapentry(subdef); +} + +const upb_fielddef* map_field_key(const upb_fielddef* field) { + assert(is_map_field(field)); + const upb_msgdef* subdef = upb_fielddef_msgsubdef(field); + return map_entry_key(subdef); +} + +const upb_fielddef* map_field_value(const upb_fielddef* field) { + assert(is_map_field(field)); + const upb_msgdef* subdef = upb_fielddef_msgsubdef(field); + return map_entry_value(subdef); +} + +const upb_fielddef* map_entry_key(const upb_msgdef* msgdef) { + const upb_fielddef* key_field = upb_msgdef_itof(msgdef, MAP_KEY_FIELD); + assert(key_field != NULL); + return key_field; +} + +const upb_fielddef* map_entry_value(const upb_msgdef* msgdef) { + const upb_fielddef* value_field = upb_msgdef_itof(msgdef, MAP_VALUE_FIELD); + assert(value_field != NULL); + return value_field; +} + // ----------------------------------------------------------------------------- // Memory layout management. // ----------------------------------------------------------------------------- +static size_t align_up_to(size_t offset, size_t granularity) { + // Granularity must be a power of two. + return (offset + granularity - 1) & ~(granularity - 1); +} + MessageLayout* create_layout(const upb_msgdef* msgdef) { MessageLayout* layout = ALLOC(MessageLayout); int nfields = upb_msgdef_numfields(msgdef); - layout->offsets = ALLOC_N(size_t, nfields); + layout->fields = ALLOC_N(MessageField, nfields); - upb_msg_iter it; + upb_msg_field_iter it; size_t off = 0; - for (upb_msg_begin(&it, msgdef); !upb_msg_done(&it); upb_msg_next(&it)) { + for (upb_msg_field_begin(&it, msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); - size_t field_size = - (upb_fielddef_label(field) == UPB_LABEL_REPEATED) ? - sizeof(VALUE) : native_slot_size(upb_fielddef_type(field)); - // align current offset + + if (upb_fielddef_containingoneof(field)) { + // Oneofs are handled separately below. + continue; + } + + // Allocate |field_size| bytes for this field in the layout. + size_t field_size = 0; + if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + field_size = sizeof(VALUE); + } else { + field_size = native_slot_size(upb_fielddef_type(field)); + } + // Align current offset up to |size| granularity. + off = align_up_to(off, field_size); + layout->fields[upb_fielddef_index(field)].offset = off; + layout->fields[upb_fielddef_index(field)].case_offset = MESSAGE_FIELD_NO_CASE; + off += field_size; + } + + // Handle oneofs now -- we iterate over oneofs specifically and allocate only + // one slot per oneof. + // + // We assign all value slots first, then pack the 'case' fields at the end, + // since in the common case (modern 64-bit platform) these are 8 bytes and 4 + // bytes respectively and we want to avoid alignment overhead. + // + // Note that we reserve 4 bytes (a uint32) per 'case' slot because the value + // space for oneof cases is conceptually as wide as field tag numbers. In + // practice, it's unlikely that a oneof would have more than e.g. 256 or 64K + // members (8 or 16 bits respectively), so conceivably we could assign + // consecutive case numbers and then pick a smaller oneof case slot size, but + // the complexity to implement this indirection is probably not worthwhile. + upb_msg_oneof_iter oit; + for (upb_msg_oneof_begin(&oit, msgdef); + !upb_msg_oneof_done(&oit); + upb_msg_oneof_next(&oit)) { + const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit); + + // Always allocate NATIVE_SLOT_MAX_SIZE bytes, but share the slot between + // all fields. + size_t field_size = NATIVE_SLOT_MAX_SIZE; + // Align the offset. + off = align_up_to(off, field_size); + // Assign all fields in the oneof this same offset. + upb_oneof_iter fit; + for (upb_oneof_begin(&fit, oneof); + !upb_oneof_done(&fit); + upb_oneof_next(&fit)) { + const upb_fielddef* field = upb_oneof_iter_field(&fit); + layout->fields[upb_fielddef_index(field)].offset = off; + } + off += field_size; + } + + // Now the case fields. + for (upb_msg_oneof_begin(&oit, msgdef); + !upb_msg_oneof_done(&oit); + upb_msg_oneof_next(&oit)) { + const upb_oneofdef* oneof = upb_msg_iter_oneof(&oit); + + size_t field_size = sizeof(uint32_t); + // Align the offset. off = (off + field_size - 1) & ~(field_size - 1); - layout->offsets[upb_fielddef_index(field)] = off; + // Assign all fields in the oneof this same offset. + upb_oneof_iter fit; + for (upb_oneof_begin(&fit, oneof); + !upb_oneof_done(&fit); + upb_oneof_next(&fit)) { + const upb_fielddef* field = upb_oneof_iter_field(&fit); + layout->fields[upb_fielddef_index(field)].case_offset = off; + } off += field_size; } @@ -352,12 +481,12 @@ MessageLayout* create_layout(const upb_msgdef* msgdef) { } void free_layout(MessageLayout* layout) { - xfree(layout->offsets); + xfree(layout->fields); upb_msgdef_unref(layout->msgdef, &layout->msgdef); xfree(layout); } -static VALUE get_type_class(const upb_fielddef* field) { +VALUE field_type_class(const upb_fielddef* field) { VALUE type_class = Qnil; if (upb_fielddef_type(field) == UPB_TYPE_MESSAGE) { VALUE submsgdesc = @@ -371,16 +500,39 @@ static VALUE get_type_class(const upb_fielddef* field) { return type_class; } +static void* slot_memory(MessageLayout* layout, + const void* storage, + const upb_fielddef* field) { + return ((uint8_t *)storage) + + layout->fields[upb_fielddef_index(field)].offset; +} + +static uint32_t* slot_oneof_case(MessageLayout* layout, + const void* storage, + const upb_fielddef* field) { + return (uint32_t *)(((uint8_t *)storage) + + layout->fields[upb_fielddef_index(field)].case_offset); +} + + VALUE layout_get(MessageLayout* layout, - void* storage, + const void* storage, const upb_fielddef* field) { - void* memory = ((uint8_t *)storage) + - layout->offsets[upb_fielddef_index(field)]; - if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); + + if (upb_fielddef_containingoneof(field)) { + if (*oneof_case != upb_fielddef_number(field)) { + return Qnil; + } + return native_slot_get(upb_fielddef_type(field), + field_type_class(field), + memory); + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { return *((VALUE *)memory); } else { return native_slot_get(upb_fielddef_type(field), - get_type_class(field), + field_type_class(field), memory); } } @@ -398,9 +550,8 @@ static void check_repeated_field_type(VALUE val, const upb_fielddef* field) { rb_raise(rb_eTypeError, "Repeated field array has wrong element type"); } - if (upb_fielddef_type(field) == UPB_TYPE_MESSAGE || - upb_fielddef_type(field) == UPB_TYPE_ENUM) { - RepeatedField* self = ruby_to_RepeatedField(val); + if (self->field_type == UPB_TYPE_MESSAGE || + self->field_type == UPB_TYPE_ENUM) { if (self->field_type_class != get_def_obj(upb_fielddef_subdef(field))) { rb_raise(rb_eTypeError, @@ -409,34 +560,117 @@ static void check_repeated_field_type(VALUE val, const upb_fielddef* field) { } } +static void check_map_field_type(VALUE val, const upb_fielddef* field) { + assert(is_map_field(field)); + const upb_fielddef* key_field = map_field_key(field); + const upb_fielddef* value_field = map_field_value(field); + + if (!RB_TYPE_P(val, T_DATA) || !RTYPEDDATA_P(val) || + RTYPEDDATA_TYPE(val) != &Map_type) { + rb_raise(rb_eTypeError, "Expected Map instance"); + } + + Map* self = ruby_to_Map(val); + if (self->key_type != upb_fielddef_type(key_field)) { + rb_raise(rb_eTypeError, "Map key type does not match field's key type"); + } + if (self->value_type != upb_fielddef_type(value_field)) { + rb_raise(rb_eTypeError, "Map value type does not match field's value type"); + } + if (upb_fielddef_type(value_field) == UPB_TYPE_MESSAGE || + upb_fielddef_type(value_field) == UPB_TYPE_ENUM) { + if (self->value_type_class != + get_def_obj(upb_fielddef_subdef(value_field))) { + rb_raise(rb_eTypeError, + "Map value type has wrong message/enum class"); + } + } +} + + void layout_set(MessageLayout* layout, void* storage, const upb_fielddef* field, VALUE val) { - void* memory = ((uint8_t *)storage) + - layout->offsets[upb_fielddef_index(field)]; - if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); + + if (upb_fielddef_containingoneof(field)) { + if (val == Qnil) { + // Assigning nil to a oneof field clears the oneof completely. + *oneof_case = ONEOF_CASE_NONE; + memset(memory, 0, NATIVE_SLOT_MAX_SIZE); + } else { + // The transition between field types for a single oneof (union) slot is + // somewhat complex because we need to ensure that a GC triggered at any + // point by a call into the Ruby VM sees a valid state for this field and + // does not either go off into the weeds (following what it thinks is a + // VALUE but is actually a different field type) or miss an object (seeing + // what it thinks is a primitive field but is actually a VALUE for the new + // field type). + // + // In order for the transition to be safe, the oneof case slot must be in + // sync with the value slot whenever the Ruby VM has been called. Thus, we + // use native_slot_set_value_and_case(), which ensures that both the value + // and case number are altered atomically (w.r.t. the Ruby VM). + native_slot_set_value_and_case( + upb_fielddef_type(field), field_type_class(field), + memory, val, + oneof_case, upb_fielddef_number(field)); + } + } else if (is_map_field(field)) { + check_map_field_type(val, field); + DEREF(memory, VALUE) = val; + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { check_repeated_field_type(val, field); - *((VALUE *)memory) = val; + DEREF(memory, VALUE) = val; } else { - native_slot_set(upb_fielddef_type(field), get_type_class(field), + native_slot_set(upb_fielddef_type(field), field_type_class(field), memory, val); } } void layout_init(MessageLayout* layout, void* storage) { - upb_msg_iter it; - for (upb_msg_begin(&it, layout->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); - void* memory = ((uint8_t *)storage) + - layout->offsets[upb_fielddef_index(field)]; + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); - if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + if (upb_fielddef_containingoneof(field)) { + memset(memory, 0, NATIVE_SLOT_MAX_SIZE); + *oneof_case = ONEOF_CASE_NONE; + } else if (is_map_field(field)) { + VALUE map = Qnil; + + const upb_fielddef* key_field = map_field_key(field); + const upb_fielddef* value_field = map_field_value(field); + VALUE type_class = field_type_class(value_field); + + if (type_class != Qnil) { + VALUE args[3] = { + fieldtype_to_ruby(upb_fielddef_type(key_field)), + fieldtype_to_ruby(upb_fielddef_type(value_field)), + type_class, + }; + map = rb_class_new_instance(3, args, cMap); + } else { + VALUE args[2] = { + fieldtype_to_ruby(upb_fielddef_type(key_field)), + fieldtype_to_ruby(upb_fielddef_type(value_field)), + }; + map = rb_class_new_instance(2, args, cMap); + } + + DEREF(memory, VALUE) = map; + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { VALUE ary = Qnil; - VALUE type_class = get_type_class(field); + + VALUE type_class = field_type_class(field); + if (type_class != Qnil) { VALUE args[2] = { fieldtype_to_ruby(upb_fielddef_type(field)), @@ -447,7 +681,8 @@ void layout_init(MessageLayout* layout, VALUE args[1] = { fieldtype_to_ruby(upb_fielddef_type(field)) }; ary = rb_class_new_instance(1, args, cRepeatedField); } - *((VALUE *)memory) = ary; + + DEREF(memory, VALUE) = ary; } else { native_slot_init(upb_fielddef_type(field), memory); } @@ -455,16 +690,20 @@ void layout_init(MessageLayout* layout, } void layout_mark(MessageLayout* layout, void* storage) { - upb_msg_iter it; - for (upb_msg_begin(&it, layout->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); - void* memory = ((uint8_t *)storage) + - layout->offsets[upb_fielddef_index(field)]; + void* memory = slot_memory(layout, storage, field); + uint32_t* oneof_case = slot_oneof_case(layout, storage, field); - if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { - rb_gc_mark(*((VALUE *)memory)); + if (upb_fielddef_containingoneof(field)) { + if (*oneof_case == upb_fielddef_number(field)) { + native_slot_mark(upb_fielddef_type(field), memory); + } + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + rb_gc_mark(DEREF(memory, VALUE)); } else { native_slot_mark(upb_fielddef_type(field), memory); } @@ -472,18 +711,26 @@ void layout_mark(MessageLayout* layout, void* storage) { } void layout_dup(MessageLayout* layout, void* to, void* from) { - upb_msg_iter it; - for (upb_msg_begin(&it, layout->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); - void* to_memory = ((uint8_t *)to) + - layout->offsets[upb_fielddef_index(field)]; - void* from_memory = ((uint8_t *)from) + - layout->offsets[upb_fielddef_index(field)]; - if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { - *((VALUE *)to_memory) = RepeatedField_dup(*((VALUE *)from_memory)); + void* to_memory = slot_memory(layout, to, field); + uint32_t* to_oneof_case = slot_oneof_case(layout, to, field); + void* from_memory = slot_memory(layout, from, field); + uint32_t* from_oneof_case = slot_oneof_case(layout, from, field); + + if (upb_fielddef_containingoneof(field)) { + if (*from_oneof_case == upb_fielddef_number(field)) { + *to_oneof_case = *from_oneof_case; + native_slot_dup(upb_fielddef_type(field), to_memory, from_memory); + } + } else if (is_map_field(field)) { + DEREF(to_memory, VALUE) = Map_dup(DEREF(from_memory, VALUE)); + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + DEREF(to_memory, VALUE) = RepeatedField_dup(DEREF(from_memory, VALUE)); } else { native_slot_dup(upb_fielddef_type(field), to_memory, from_memory); } @@ -491,18 +738,28 @@ void layout_dup(MessageLayout* layout, void* to, void* from) { } void layout_deep_copy(MessageLayout* layout, void* to, void* from) { - upb_msg_iter it; - for (upb_msg_begin(&it, layout->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); - void* to_memory = ((uint8_t *)to) + - layout->offsets[upb_fielddef_index(field)]; - void* from_memory = ((uint8_t *)from) + - layout->offsets[upb_fielddef_index(field)]; - if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { - *((VALUE *)to_memory) = RepeatedField_deep_copy(*((VALUE *)from_memory)); + void* to_memory = slot_memory(layout, to, field); + uint32_t* to_oneof_case = slot_oneof_case(layout, to, field); + void* from_memory = slot_memory(layout, from, field); + uint32_t* from_oneof_case = slot_oneof_case(layout, from, field); + + if (upb_fielddef_containingoneof(field)) { + if (*from_oneof_case == upb_fielddef_number(field)) { + *to_oneof_case = *from_oneof_case; + native_slot_deep_copy(upb_fielddef_type(field), to_memory, from_memory); + } + } else if (is_map_field(field)) { + DEREF(to_memory, VALUE) = + Map_deep_copy(DEREF(from_memory, VALUE)); + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + DEREF(to_memory, VALUE) = + RepeatedField_deep_copy(DEREF(from_memory, VALUE)); } else { native_slot_deep_copy(upb_fielddef_type(field), to_memory, from_memory); } @@ -510,19 +767,33 @@ void layout_deep_copy(MessageLayout* layout, void* to, void* from) { } VALUE layout_eq(MessageLayout* layout, void* msg1, void* msg2) { - upb_msg_iter it; - for (upb_msg_begin(&it, layout->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + upb_msg_field_iter it; + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); - void* msg1_memory = ((uint8_t *)msg1) + - layout->offsets[upb_fielddef_index(field)]; - void* msg2_memory = ((uint8_t *)msg2) + - layout->offsets[upb_fielddef_index(field)]; - if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { - if (RepeatedField_eq(*((VALUE *)msg1_memory), - *((VALUE *)msg2_memory)) == Qfalse) { + void* msg1_memory = slot_memory(layout, msg1, field); + uint32_t* msg1_oneof_case = slot_oneof_case(layout, msg1, field); + void* msg2_memory = slot_memory(layout, msg2, field); + uint32_t* msg2_oneof_case = slot_oneof_case(layout, msg2, field); + + if (upb_fielddef_containingoneof(field)) { + if (*msg1_oneof_case != *msg2_oneof_case || + (*msg1_oneof_case == upb_fielddef_number(field) && + !native_slot_eq(upb_fielddef_type(field), + msg1_memory, + msg2_memory))) { + return Qfalse; + } + } else if (is_map_field(field)) { + if (!Map_eq(DEREF(msg1_memory, VALUE), + DEREF(msg2_memory, VALUE))) { + return Qfalse; + } + } else if (upb_fielddef_label(field) == UPB_LABEL_REPEATED) { + if (!RepeatedField_eq(DEREF(msg1_memory, VALUE), + DEREF(msg2_memory, VALUE))) { return Qfalse; } } else { @@ -536,12 +807,12 @@ VALUE layout_eq(MessageLayout* layout, void* msg1, void* msg2) { } VALUE layout_hash(MessageLayout* layout, void* storage) { - upb_msg_iter it; + upb_msg_field_iter it; st_index_t h = rb_hash_start(0); VALUE hash_sym = rb_intern("hash"); - for (upb_msg_begin(&it, layout->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); VALUE field_val = layout_get(layout, storage, field); h = rb_hash_uint(h, NUM2LONG(rb_funcall(field_val, hash_sym, 0))); @@ -554,11 +825,11 @@ VALUE layout_hash(MessageLayout* layout, void* storage) { VALUE layout_inspect(MessageLayout* layout, void* storage) { VALUE str = rb_str_new2(""); - upb_msg_iter it; + upb_msg_field_iter it; bool first = true; - for (upb_msg_begin(&it, layout->msgdef); - !upb_msg_done(&it); - upb_msg_next(&it)) { + for (upb_msg_field_begin(&it, layout->msgdef); + !upb_msg_field_done(&it); + upb_msg_field_next(&it)) { const upb_fielddef* field = upb_msg_iter_field(&it); VALUE field_val = layout_get(layout, storage, field); diff --git a/ruby/ext/google/protobuf_c/upb.c b/ruby/ext/google/protobuf_c/upb.c new file mode 100644 index 0000000000..20bd76bc77 --- /dev/null +++ b/ruby/ext/google/protobuf_c/upb.c @@ -0,0 +1,11220 @@ +// Amalgamated source file +#include "upb.h" +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2008-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + + +#include +#include + +typedef struct { + size_t len; + char str[1]; // Null-terminated string data follows. +} str_t; + +static str_t *newstr(const char *data, size_t len) { + str_t *ret = malloc(sizeof(*ret) + len); + if (!ret) return NULL; + ret->len = len; + memcpy(ret->str, data, len); + ret->str[len] = '\0'; + return ret; +} + +static void freestr(str_t *s) { free(s); } + +// isalpha() etc. from are locale-dependent, which we don't want. +static bool upb_isbetween(char c, char low, char high) { + return c >= low && c <= high; +} + +static bool upb_isletter(char c) { + return upb_isbetween(c, 'A', 'Z') || upb_isbetween(c, 'a', 'z') || c == '_'; +} + +static bool upb_isalphanum(char c) { + return upb_isletter(c) || upb_isbetween(c, '0', '9'); +} + +static bool upb_isident(const char *str, size_t len, bool full, upb_status *s) { + bool start = true; + for (size_t i = 0; i < len; i++) { + char c = str[i]; + if (c == '.') { + if (start || !full) { + upb_status_seterrf(s, "invalid name: unexpected '.' (%s)", str); + return false; + } + start = true; + } else if (start) { + if (!upb_isletter(c)) { + upb_status_seterrf( + s, "invalid name: path components must start with a letter (%s)", + str); + return false; + } + start = false; + } else { + if (!upb_isalphanum(c)) { + upb_status_seterrf(s, "invalid name: non-alphanumeric character (%s)", + str); + return false; + } + } + } + return !start; +} + + +/* upb_def ********************************************************************/ + +upb_deftype_t upb_def_type(const upb_def *d) { return d->type; } + +const char *upb_def_fullname(const upb_def *d) { return d->fullname; } + +bool upb_def_setfullname(upb_def *def, const char *fullname, upb_status *s) { + assert(!upb_def_isfrozen(def)); + if (!upb_isident(fullname, strlen(fullname), true, s)) return false; + free((void*)def->fullname); + def->fullname = upb_strdup(fullname); + return true; +} + +upb_def *upb_def_dup(const upb_def *def, const void *o) { + switch (def->type) { + case UPB_DEF_MSG: + return UPB_UPCAST(upb_msgdef_dup(upb_downcast_msgdef(def), o)); + case UPB_DEF_FIELD: + return UPB_UPCAST(upb_fielddef_dup(upb_downcast_fielddef(def), o)); + case UPB_DEF_ENUM: + return UPB_UPCAST(upb_enumdef_dup(upb_downcast_enumdef(def), o)); + default: assert(false); return NULL; + } +} + +bool upb_def_isfrozen(const upb_def *def) { + return upb_refcounted_isfrozen(UPB_UPCAST(def)); +} + +void upb_def_ref(const upb_def *def, const void *owner) { + upb_refcounted_ref(UPB_UPCAST(def), owner); +} + +void upb_def_unref(const upb_def *def, const void *owner) { + upb_refcounted_unref(UPB_UPCAST(def), owner); +} + +void upb_def_donateref(const upb_def *def, const void *from, const void *to) { + upb_refcounted_donateref(UPB_UPCAST(def), from, to); +} + +void upb_def_checkref(const upb_def *def, const void *owner) { + upb_refcounted_checkref(UPB_UPCAST(def), owner); +} + +static bool upb_def_init(upb_def *def, upb_deftype_t type, + const struct upb_refcounted_vtbl *vtbl, + const void *owner) { + if (!upb_refcounted_init(UPB_UPCAST(def), vtbl, owner)) return false; + def->type = type; + def->fullname = NULL; + def->came_from_user = false; + return true; +} + +static void upb_def_uninit(upb_def *def) { + free((void*)def->fullname); +} + +static const char *msgdef_name(const upb_msgdef *m) { + const char *name = upb_def_fullname(UPB_UPCAST(m)); + return name ? name : "(anonymous)"; +} + +static bool upb_validate_field(upb_fielddef *f, upb_status *s) { + if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) { + upb_status_seterrmsg(s, "fielddef must have name and number set"); + return false; + } + + if (!f->type_is_set_) { + upb_status_seterrmsg(s, "fielddef type was not initialized"); + return false; + } + + if (upb_fielddef_lazy(f) && + upb_fielddef_descriptortype(f) != UPB_DESCRIPTOR_TYPE_MESSAGE) { + upb_status_seterrmsg(s, + "only length-delimited submessage fields may be lazy"); + return false; + } + + if (upb_fielddef_hassubdef(f)) { + if (f->subdef_is_symbolic) { + upb_status_seterrf(s, "field '%s.%s' has not been resolved", + msgdef_name(f->msg.def), upb_fielddef_name(f)); + return false; + } + + const upb_def *subdef = upb_fielddef_subdef(f); + if (subdef == NULL) { + upb_status_seterrf(s, "field %s.%s is missing required subdef", + msgdef_name(f->msg.def), upb_fielddef_name(f)); + return false; + } + + if (!upb_def_isfrozen(subdef) && !subdef->came_from_user) { + upb_status_seterrf(s, + "subdef of field %s.%s is not frozen or being frozen", + msgdef_name(f->msg.def), upb_fielddef_name(f)); + return false; + } + } + + if (upb_fielddef_type(f) == UPB_TYPE_ENUM) { + bool has_default_name = upb_fielddef_enumhasdefaultstr(f); + bool has_default_number = upb_fielddef_enumhasdefaultint32(f); + + // Previously verified by upb_validate_enumdef(). + assert(upb_enumdef_numvals(upb_fielddef_enumsubdef(f)) > 0); + + // We've already validated that we have an associated enumdef and that it + // has at least one member, so at least one of these should be true. + // Because if the user didn't set anything, we'll pick up the enum's + // default, but if the user *did* set something we should at least pick up + // the one they set (int32 or string). + assert(has_default_name || has_default_number); + + if (!has_default_name) { + upb_status_seterrf(s, + "enum default for field %s.%s (%d) is not in the enum", + msgdef_name(f->msg.def), upb_fielddef_name(f), + upb_fielddef_defaultint32(f)); + return false; + } + + if (!has_default_number) { + upb_status_seterrf(s, + "enum default for field %s.%s (%s) is not in the enum", + msgdef_name(f->msg.def), upb_fielddef_name(f), + upb_fielddef_defaultstr(f, NULL)); + return false; + } + + // Lift the effective numeric default into the field's default slot, in case + // we were only getting it "by reference" from the enumdef. + upb_fielddef_setdefaultint32(f, upb_fielddef_defaultint32(f)); + } + + // Ensure that MapEntry submessages only appear as repeated fields, not + // optional/required (singular) fields. + if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE && + upb_fielddef_msgsubdef(f) != NULL) { + const upb_msgdef *subdef = upb_fielddef_msgsubdef(f); + if (upb_msgdef_mapentry(subdef) && !upb_fielddef_isseq(f)) { + upb_status_seterrf(s, + "Field %s refers to mapentry message but is not " + "a repeated field", + upb_fielddef_name(f) ? upb_fielddef_name(f) : + "(unnamed)"); + return false; + } + } + + return true; +} + +static bool upb_validate_enumdef(const upb_enumdef *e, upb_status *s) { + if (upb_enumdef_numvals(e) == 0) { + upb_status_seterrf(s, "enum %s has no members (must have at least one)", + upb_enumdef_fullname(e)); + return false; + } + + return true; +} + +// All submessage fields are lower than all other fields. +// Secondly, fields are increasing in order. +uint32_t field_rank(const upb_fielddef *f) { + uint32_t ret = upb_fielddef_number(f); + const uint32_t high_bit = 1 << 30; + assert(ret < high_bit); + if (!upb_fielddef_issubmsg(f)) + ret |= high_bit; + return ret; +} + +int cmp_fields(const void *p1, const void *p2) { + const upb_fielddef *f1 = *(upb_fielddef*const*)p1; + const upb_fielddef *f2 = *(upb_fielddef*const*)p2; + return field_rank(f1) - field_rank(f2); +} + +static bool assign_msg_indices(upb_msgdef *m, upb_status *s) { + // Sort fields. upb internally relies on UPB_TYPE_MESSAGE fields having the + // lowest indexes, but we do not publicly guarantee this. + int n = upb_msgdef_numfields(m); + upb_fielddef **fields = malloc(n * sizeof(*fields)); + if (!fields) return false; + + upb_msg_field_iter j; + int i; + m->submsg_field_count = 0; + for(i = 0, upb_msg_field_begin(&j, m); + !upb_msg_field_done(&j); + upb_msg_field_next(&j), i++) { + upb_fielddef *f = upb_msg_iter_field(&j); + assert(f->msg.def == m); + if (!upb_validate_field(f, s)) { + free(fields); + return false; + } + if (upb_fielddef_issubmsg(f)) { + m->submsg_field_count++; + } + fields[i] = f; + } + + qsort(fields, n, sizeof(*fields), cmp_fields); + + uint32_t selector = UPB_STATIC_SELECTOR_COUNT + m->submsg_field_count; + for (i = 0; i < n; i++) { + upb_fielddef *f = fields[i]; + f->index_ = i; + f->selector_base = selector + upb_handlers_selectorbaseoffset(f); + selector += upb_handlers_selectorcount(f); + } + m->selector_count = selector; + +#ifndef NDEBUG + // Verify that all selectors for the message are distinct. + // +#define TRY(type) \ + if (upb_handlers_getselector(f, type, &sel)) upb_inttable_insert(&t, sel, v); + + upb_inttable t; + upb_inttable_init(&t, UPB_CTYPE_BOOL); + upb_value v = upb_value_bool(true); + upb_selector_t sel; + upb_inttable_insert(&t, UPB_STARTMSG_SELECTOR, v); + upb_inttable_insert(&t, UPB_ENDMSG_SELECTOR, v); + for(upb_msg_field_begin(&j, m); + !upb_msg_field_done(&j); + upb_msg_field_next(&j)) { + upb_fielddef *f = upb_msg_iter_field(&j); + // These calls will assert-fail in upb_table if the value already exists. + TRY(UPB_HANDLER_INT32); + TRY(UPB_HANDLER_INT64) + TRY(UPB_HANDLER_UINT32) + TRY(UPB_HANDLER_UINT64) + TRY(UPB_HANDLER_FLOAT) + TRY(UPB_HANDLER_DOUBLE) + TRY(UPB_HANDLER_BOOL) + TRY(UPB_HANDLER_STARTSTR) + TRY(UPB_HANDLER_STRING) + TRY(UPB_HANDLER_ENDSTR) + TRY(UPB_HANDLER_STARTSUBMSG) + TRY(UPB_HANDLER_ENDSUBMSG) + TRY(UPB_HANDLER_STARTSEQ) + TRY(UPB_HANDLER_ENDSEQ) + } + upb_inttable_uninit(&t); +#undef TRY +#endif + + free(fields); + return true; +} + +bool upb_def_freeze(upb_def *const* defs, int n, upb_status *s) { + upb_status_clear(s); + + // First perform validation, in two passes so we can check that we have a + // transitive closure without needing to search. + for (int i = 0; i < n; i++) { + upb_def *def = defs[i]; + if (upb_def_isfrozen(def)) { + // Could relax this requirement if it's annoying. + upb_status_seterrmsg(s, "def is already frozen"); + goto err; + } else if (def->type == UPB_DEF_FIELD) { + upb_status_seterrmsg(s, "standalone fielddefs can not be frozen"); + goto err; + } else if (def->type == UPB_DEF_ENUM) { + if (!upb_validate_enumdef(upb_dyncast_enumdef(def), s)) { + goto err; + } + } else { + // Set now to detect transitive closure in the second pass. + def->came_from_user = true; + } + } + + // Second pass of validation. Also assign selector bases and indexes, and + // compact tables. + for (int i = 0; i < n; i++) { + upb_msgdef *m = upb_dyncast_msgdef_mutable(defs[i]); + upb_enumdef *e = upb_dyncast_enumdef_mutable(defs[i]); + if (m) { + upb_inttable_compact(&m->itof); + if (!assign_msg_indices(m, s)) { + goto err; + } + } else if (e) { + upb_inttable_compact(&e->iton); + } + } + + // Def graph contains FieldDefs between each MessageDef, so double the limit. + int maxdepth = UPB_MAX_MESSAGE_DEPTH * 2; + + // Validation all passed; freeze the defs. + bool ret = + upb_refcounted_freeze((upb_refcounted * const *)defs, n, s, maxdepth); + assert(!(s && ret != upb_ok(s))); + return ret; + +err: + for (int i = 0; i < n; i++) { + defs[i]->came_from_user = false; + } + assert(!(s && upb_ok(s))); + return false; +} + + +/* upb_enumdef ****************************************************************/ + +static void upb_enumdef_free(upb_refcounted *r) { + upb_enumdef *e = (upb_enumdef*)r; + upb_inttable_iter i; + upb_inttable_begin(&i, &e->iton); + for( ; !upb_inttable_done(&i); upb_inttable_next(&i)) { + // To clean up the upb_strdup() from upb_enumdef_addval(). + free(upb_value_getcstr(upb_inttable_iter_value(&i))); + } + upb_strtable_uninit(&e->ntoi); + upb_inttable_uninit(&e->iton); + upb_def_uninit(UPB_UPCAST(e)); + free(e); +} + +upb_enumdef *upb_enumdef_new(const void *owner) { + static const struct upb_refcounted_vtbl vtbl = {NULL, &upb_enumdef_free}; + upb_enumdef *e = malloc(sizeof(*e)); + if (!e) return NULL; + if (!upb_def_init(UPB_UPCAST(e), UPB_DEF_ENUM, &vtbl, owner)) goto err2; + if (!upb_strtable_init(&e->ntoi, UPB_CTYPE_INT32)) goto err2; + if (!upb_inttable_init(&e->iton, UPB_CTYPE_CSTR)) goto err1; + return e; + +err1: + upb_strtable_uninit(&e->ntoi); +err2: + free(e); + return NULL; +} + +upb_enumdef *upb_enumdef_dup(const upb_enumdef *e, const void *owner) { + upb_enumdef *new_e = upb_enumdef_new(owner); + if (!new_e) return NULL; + upb_enum_iter i; + for(upb_enum_begin(&i, e); !upb_enum_done(&i); upb_enum_next(&i)) { + bool success = upb_enumdef_addval( + new_e, upb_enum_iter_name(&i),upb_enum_iter_number(&i), NULL); + if (!success) { + upb_enumdef_unref(new_e, owner); + return NULL; + } + } + return new_e; +} + +bool upb_enumdef_isfrozen(const upb_enumdef *e) { + return upb_def_isfrozen(UPB_UPCAST(e)); +} + +void upb_enumdef_ref(const upb_enumdef *e, const void *owner) { + upb_def_ref(UPB_UPCAST(e), owner); +} + +void upb_enumdef_unref(const upb_enumdef *e, const void *owner) { + upb_def_unref(UPB_UPCAST(e), owner); +} + +void upb_enumdef_donateref( + const upb_enumdef *e, const void *from, const void *to) { + upb_def_donateref(UPB_UPCAST(e), from, to); +} + +void upb_enumdef_checkref(const upb_enumdef *e, const void *owner) { + upb_def_checkref(UPB_UPCAST(e), owner); +} + +bool upb_enumdef_freeze(upb_enumdef *e, upb_status *status) { + upb_def *d = UPB_UPCAST(e); + return upb_def_freeze(&d, 1, status); +} + +const char *upb_enumdef_fullname(const upb_enumdef *e) { + return upb_def_fullname(UPB_UPCAST(e)); +} + +bool upb_enumdef_setfullname(upb_enumdef *e, const char *fullname, + upb_status *s) { + return upb_def_setfullname(UPB_UPCAST(e), fullname, s); +} + +bool upb_enumdef_addval(upb_enumdef *e, const char *name, int32_t num, + upb_status *status) { + if (!upb_isident(name, strlen(name), false, status)) { + return false; + } + if (upb_enumdef_ntoiz(e, name, NULL)) { + upb_status_seterrf(status, "name '%s' is already defined", name); + return false; + } + if (!upb_strtable_insert(&e->ntoi, name, upb_value_int32(num))) { + upb_status_seterrmsg(status, "out of memory"); + return false; + } + if (!upb_inttable_lookup(&e->iton, num, NULL) && + !upb_inttable_insert(&e->iton, num, upb_value_cstr(upb_strdup(name)))) { + upb_status_seterrmsg(status, "out of memory"); + upb_strtable_remove(&e->ntoi, name, NULL); + return false; + } + if (upb_enumdef_numvals(e) == 1) { + bool ok = upb_enumdef_setdefault(e, num, NULL); + UPB_ASSERT_VAR(ok, ok); + } + return true; +} + +int32_t upb_enumdef_default(const upb_enumdef *e) { + assert(upb_enumdef_iton(e, e->defaultval)); + return e->defaultval; +} + +bool upb_enumdef_setdefault(upb_enumdef *e, int32_t val, upb_status *s) { + assert(!upb_enumdef_isfrozen(e)); + if (!upb_enumdef_iton(e, val)) { + upb_status_seterrf(s, "number '%d' is not in the enum.", val); + return false; + } + e->defaultval = val; + return true; +} + +int upb_enumdef_numvals(const upb_enumdef *e) { + return upb_strtable_count(&e->ntoi); +} + +void upb_enum_begin(upb_enum_iter *i, const upb_enumdef *e) { + // We iterate over the ntoi table, to account for duplicate numbers. + upb_strtable_begin(i, &e->ntoi); +} + +void upb_enum_next(upb_enum_iter *iter) { upb_strtable_next(iter); } +bool upb_enum_done(upb_enum_iter *iter) { return upb_strtable_done(iter); } + +bool upb_enumdef_ntoi(const upb_enumdef *def, const char *name, + size_t len, int32_t *num) { + upb_value v; + if (!upb_strtable_lookup2(&def->ntoi, name, len, &v)) { + return false; + } + if (num) *num = upb_value_getint32(v); + return true; +} + +const char *upb_enumdef_iton(const upb_enumdef *def, int32_t num) { + upb_value v; + return upb_inttable_lookup32(&def->iton, num, &v) ? + upb_value_getcstr(v) : NULL; +} + +const char *upb_enum_iter_name(upb_enum_iter *iter) { + return upb_strtable_iter_key(iter); +} + +int32_t upb_enum_iter_number(upb_enum_iter *iter) { + return upb_value_getint32(upb_strtable_iter_value(iter)); +} + + +/* upb_fielddef ***************************************************************/ + +static void upb_fielddef_init_default(upb_fielddef *f); + +static void upb_fielddef_uninit_default(upb_fielddef *f) { + if (f->type_is_set_ && f->default_is_string && f->defaultval.bytes) + freestr(f->defaultval.bytes); +} + +static void visitfield(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_fielddef *f = (const upb_fielddef*)r; + if (upb_fielddef_containingtype(f)) { + visit(r, UPB_UPCAST2(upb_fielddef_containingtype(f)), closure); + } + if (upb_fielddef_containingoneof(f)) { + visit(r, UPB_UPCAST2(upb_fielddef_containingoneof(f)), closure); + } + if (upb_fielddef_subdef(f)) { + visit(r, UPB_UPCAST(upb_fielddef_subdef(f)), closure); + } +} + +static void freefield(upb_refcounted *r) { + upb_fielddef *f = (upb_fielddef*)r; + upb_fielddef_uninit_default(f); + if (f->subdef_is_symbolic) + free(f->sub.name); + upb_def_uninit(UPB_UPCAST(f)); + free(f); +} + +static const char *enumdefaultstr(const upb_fielddef *f) { + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + const upb_enumdef *e = upb_fielddef_enumsubdef(f); + if (f->default_is_string && f->defaultval.bytes) { + // Default was explicitly set as a string. + str_t *s = f->defaultval.bytes; + return s->str; + } else if (e) { + if (!f->default_is_string) { + // Default was explicitly set as an integer; look it up in enumdef. + const char *name = upb_enumdef_iton(e, f->defaultval.sint); + if (name) { + return name; + } + } else { + // Default is completely unset; pull enumdef default. + if (upb_enumdef_numvals(e) > 0) { + const char *name = upb_enumdef_iton(e, upb_enumdef_default(e)); + assert(name); + return name; + } + } + } + return NULL; +} + +static bool enumdefaultint32(const upb_fielddef *f, int32_t *val) { + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + const upb_enumdef *e = upb_fielddef_enumsubdef(f); + if (!f->default_is_string) { + // Default was explicitly set as an integer. + *val = f->defaultval.sint; + return true; + } else if (e) { + if (f->defaultval.bytes) { + // Default was explicitly set as a str; try to lookup corresponding int. + str_t *s = f->defaultval.bytes; + if (upb_enumdef_ntoiz(e, s->str, val)) { + return true; + } + } else { + // Default is unset; try to pull in enumdef default. + if (upb_enumdef_numvals(e) > 0) { + *val = upb_enumdef_default(e); + return true; + } + } + } + return false; +} + +upb_fielddef *upb_fielddef_new(const void *owner) { + static const struct upb_refcounted_vtbl vtbl = {visitfield, freefield}; + upb_fielddef *f = malloc(sizeof(*f)); + if (!f) return NULL; + if (!upb_def_init(UPB_UPCAST(f), UPB_DEF_FIELD, &vtbl, owner)) { + free(f); + return NULL; + } + f->msg.def = NULL; + f->sub.def = NULL; + f->oneof = NULL; + f->subdef_is_symbolic = false; + f->msg_is_symbolic = false; + f->label_ = UPB_LABEL_OPTIONAL; + f->type_ = UPB_TYPE_INT32; + f->number_ = 0; + f->type_is_set_ = false; + f->tagdelim = false; + f->is_extension_ = false; + f->lazy_ = false; + f->packed_ = true; + + // For the moment we default this to UPB_INTFMT_VARIABLE, since it will work + // with all integer types and is in some since more "default" since the most + // normal-looking proto2 types int32/int64/uint32/uint64 use variable. + // + // Other options to consider: + // - there is no default; users must set this manually (like type). + // - default signed integers to UPB_INTFMT_ZIGZAG, since it's more likely to + // be an optimal default for signed integers. + f->intfmt = UPB_INTFMT_VARIABLE; + return f; +} + +upb_fielddef *upb_fielddef_dup(const upb_fielddef *f, const void *owner) { + upb_fielddef *newf = upb_fielddef_new(owner); + if (!newf) return NULL; + upb_fielddef_settype(newf, upb_fielddef_type(f)); + upb_fielddef_setlabel(newf, upb_fielddef_label(f)); + upb_fielddef_setnumber(newf, upb_fielddef_number(f), NULL); + upb_fielddef_setname(newf, upb_fielddef_name(f), NULL); + if (f->default_is_string && f->defaultval.bytes) { + str_t *s = f->defaultval.bytes; + upb_fielddef_setdefaultstr(newf, s->str, s->len, NULL); + } else { + newf->default_is_string = f->default_is_string; + newf->defaultval = f->defaultval; + } + + const char *srcname; + if (f->subdef_is_symbolic) { + srcname = f->sub.name; // Might be NULL. + } else { + srcname = f->sub.def ? upb_def_fullname(f->sub.def) : NULL; + } + if (srcname) { + char *newname = malloc(strlen(f->sub.def->fullname) + 2); + if (!newname) { + upb_fielddef_unref(newf, owner); + return NULL; + } + strcpy(newname, "."); + strcat(newname, f->sub.def->fullname); + upb_fielddef_setsubdefname(newf, newname, NULL); + free(newname); + } + + return newf; +} + +bool upb_fielddef_isfrozen(const upb_fielddef *f) { + return upb_def_isfrozen(UPB_UPCAST(f)); +} + +void upb_fielddef_ref(const upb_fielddef *f, const void *owner) { + upb_def_ref(UPB_UPCAST(f), owner); +} + +void upb_fielddef_unref(const upb_fielddef *f, const void *owner) { + upb_def_unref(UPB_UPCAST(f), owner); +} + +void upb_fielddef_donateref( + const upb_fielddef *f, const void *from, const void *to) { + upb_def_donateref(UPB_UPCAST(f), from, to); +} + +void upb_fielddef_checkref(const upb_fielddef *f, const void *owner) { + upb_def_checkref(UPB_UPCAST(f), owner); +} + +bool upb_fielddef_typeisset(const upb_fielddef *f) { + return f->type_is_set_; +} + +upb_fieldtype_t upb_fielddef_type(const upb_fielddef *f) { + assert(f->type_is_set_); + return f->type_; +} + +uint32_t upb_fielddef_index(const upb_fielddef *f) { + return f->index_; +} + +upb_label_t upb_fielddef_label(const upb_fielddef *f) { + return f->label_; +} + +upb_intfmt_t upb_fielddef_intfmt(const upb_fielddef *f) { + return f->intfmt; +} + +bool upb_fielddef_istagdelim(const upb_fielddef *f) { + return f->tagdelim; +} + +uint32_t upb_fielddef_number(const upb_fielddef *f) { + return f->number_; +} + +bool upb_fielddef_isextension(const upb_fielddef *f) { + return f->is_extension_; +} + +bool upb_fielddef_lazy(const upb_fielddef *f) { + return f->lazy_; +} + +bool upb_fielddef_packed(const upb_fielddef *f) { + return f->packed_; +} + +const char *upb_fielddef_name(const upb_fielddef *f) { + return upb_def_fullname(UPB_UPCAST(f)); +} + +const upb_msgdef *upb_fielddef_containingtype(const upb_fielddef *f) { + return f->msg_is_symbolic ? NULL : f->msg.def; +} + +const upb_oneofdef *upb_fielddef_containingoneof(const upb_fielddef *f) { + return f->oneof; +} + +upb_msgdef *upb_fielddef_containingtype_mutable(upb_fielddef *f) { + return (upb_msgdef*)upb_fielddef_containingtype(f); +} + +const char *upb_fielddef_containingtypename(upb_fielddef *f) { + return f->msg_is_symbolic ? f->msg.name : NULL; +} + +static void release_containingtype(upb_fielddef *f) { + if (f->msg_is_symbolic) free(f->msg.name); +} + +bool upb_fielddef_setcontainingtypename(upb_fielddef *f, const char *name, + upb_status *s) { + assert(!upb_fielddef_isfrozen(f)); + if (upb_fielddef_containingtype(f)) { + upb_status_seterrmsg(s, "field has already been added to a message."); + return false; + } + // TODO: validate name (upb_isident() doesn't quite work atm because this name + // may have a leading "."). + release_containingtype(f); + f->msg.name = upb_strdup(name); + f->msg_is_symbolic = true; + return true; +} + +bool upb_fielddef_setname(upb_fielddef *f, const char *name, upb_status *s) { + if (upb_fielddef_containingtype(f) || upb_fielddef_containingoneof(f)) { + upb_status_seterrmsg(s, "Already added to message or oneof"); + return false; + } + return upb_def_setfullname(UPB_UPCAST(f), name, s); +} + +static void chkdefaulttype(const upb_fielddef *f, upb_fieldtype_t type) { + UPB_UNUSED(f); + UPB_UNUSED(type); + assert(f->type_is_set_ && upb_fielddef_type(f) == type); +} + +int64_t upb_fielddef_defaultint64(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_INT64); + return f->defaultval.sint; +} + +int32_t upb_fielddef_defaultint32(const upb_fielddef *f) { + if (f->type_is_set_ && upb_fielddef_type(f) == UPB_TYPE_ENUM) { + int32_t val; + bool ok = enumdefaultint32(f, &val); + UPB_ASSERT_VAR(ok, ok); + return val; + } else { + chkdefaulttype(f, UPB_TYPE_INT32); + return f->defaultval.sint; + } +} + +uint64_t upb_fielddef_defaultuint64(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_UINT64); + return f->defaultval.uint; +} + +uint32_t upb_fielddef_defaultuint32(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_UINT32); + return f->defaultval.uint; +} + +bool upb_fielddef_defaultbool(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_BOOL); + return f->defaultval.uint; +} + +float upb_fielddef_defaultfloat(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_FLOAT); + return f->defaultval.flt; +} + +double upb_fielddef_defaultdouble(const upb_fielddef *f) { + chkdefaulttype(f, UPB_TYPE_DOUBLE); + return f->defaultval.dbl; +} + +const char *upb_fielddef_defaultstr(const upb_fielddef *f, size_t *len) { + assert(f->type_is_set_); + assert(upb_fielddef_type(f) == UPB_TYPE_STRING || + upb_fielddef_type(f) == UPB_TYPE_BYTES || + upb_fielddef_type(f) == UPB_TYPE_ENUM); + + if (upb_fielddef_type(f) == UPB_TYPE_ENUM) { + const char *ret = enumdefaultstr(f); + assert(ret); + // Enum defaults can't have embedded NULLs. + if (len) *len = strlen(ret); + return ret; + } + + if (f->default_is_string) { + str_t *str = f->defaultval.bytes; + if (len) *len = str->len; + return str->str; + } + + return NULL; +} + +static void upb_fielddef_init_default(upb_fielddef *f) { + f->default_is_string = false; + switch (upb_fielddef_type(f)) { + case UPB_TYPE_DOUBLE: f->defaultval.dbl = 0; break; + case UPB_TYPE_FLOAT: f->defaultval.flt = 0; break; + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: f->defaultval.sint = 0; break; + case UPB_TYPE_UINT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_BOOL: f->defaultval.uint = 0; break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + f->defaultval.bytes = newstr("", 0); + f->default_is_string = true; + break; + case UPB_TYPE_MESSAGE: break; + case UPB_TYPE_ENUM: + // This is our special sentinel that indicates "not set" for an enum. + f->default_is_string = true; + f->defaultval.bytes = NULL; + break; + } +} + +const upb_def *upb_fielddef_subdef(const upb_fielddef *f) { + return f->subdef_is_symbolic ? NULL : f->sub.def; +} + +const upb_msgdef *upb_fielddef_msgsubdef(const upb_fielddef *f) { + const upb_def *def = upb_fielddef_subdef(f); + return def ? upb_dyncast_msgdef(def) : NULL; +} + +const upb_enumdef *upb_fielddef_enumsubdef(const upb_fielddef *f) { + const upb_def *def = upb_fielddef_subdef(f); + return def ? upb_dyncast_enumdef(def) : NULL; +} + +upb_def *upb_fielddef_subdef_mutable(upb_fielddef *f) { + return (upb_def*)upb_fielddef_subdef(f); +} + +const char *upb_fielddef_subdefname(const upb_fielddef *f) { + if (f->subdef_is_symbolic) { + return f->sub.name; + } else if (f->sub.def) { + return upb_def_fullname(f->sub.def); + } else { + return NULL; + } +} + +bool upb_fielddef_setnumber(upb_fielddef *f, uint32_t number, upb_status *s) { + if (upb_fielddef_containingtype(f)) { + upb_status_seterrmsg( + s, "cannot change field number after adding to a message"); + return false; + } + if (number == 0 || number > UPB_MAX_FIELDNUMBER) { + upb_status_seterrf(s, "invalid field number (%u)", number); + return false; + } + f->number_ = number; + return true; +} + +void upb_fielddef_settype(upb_fielddef *f, upb_fieldtype_t type) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_checktype(type)); + upb_fielddef_uninit_default(f); + f->type_ = type; + f->type_is_set_ = true; + upb_fielddef_init_default(f); +} + +void upb_fielddef_setdescriptortype(upb_fielddef *f, int type) { + assert(!upb_fielddef_isfrozen(f)); + switch (type) { + case UPB_DESCRIPTOR_TYPE_DOUBLE: + upb_fielddef_settype(f, UPB_TYPE_DOUBLE); + break; + case UPB_DESCRIPTOR_TYPE_FLOAT: + upb_fielddef_settype(f, UPB_TYPE_FLOAT); + break; + case UPB_DESCRIPTOR_TYPE_INT64: + case UPB_DESCRIPTOR_TYPE_SFIXED64: + case UPB_DESCRIPTOR_TYPE_SINT64: + upb_fielddef_settype(f, UPB_TYPE_INT64); + break; + case UPB_DESCRIPTOR_TYPE_UINT64: + case UPB_DESCRIPTOR_TYPE_FIXED64: + upb_fielddef_settype(f, UPB_TYPE_UINT64); + break; + case UPB_DESCRIPTOR_TYPE_INT32: + case UPB_DESCRIPTOR_TYPE_SFIXED32: + case UPB_DESCRIPTOR_TYPE_SINT32: + upb_fielddef_settype(f, UPB_TYPE_INT32); + break; + case UPB_DESCRIPTOR_TYPE_UINT32: + case UPB_DESCRIPTOR_TYPE_FIXED32: + upb_fielddef_settype(f, UPB_TYPE_UINT32); + break; + case UPB_DESCRIPTOR_TYPE_BOOL: + upb_fielddef_settype(f, UPB_TYPE_BOOL); + break; + case UPB_DESCRIPTOR_TYPE_STRING: + upb_fielddef_settype(f, UPB_TYPE_STRING); + break; + case UPB_DESCRIPTOR_TYPE_BYTES: + upb_fielddef_settype(f, UPB_TYPE_BYTES); + break; + case UPB_DESCRIPTOR_TYPE_GROUP: + case UPB_DESCRIPTOR_TYPE_MESSAGE: + upb_fielddef_settype(f, UPB_TYPE_MESSAGE); + break; + case UPB_DESCRIPTOR_TYPE_ENUM: + upb_fielddef_settype(f, UPB_TYPE_ENUM); + break; + default: assert(false); + } + + if (type == UPB_DESCRIPTOR_TYPE_FIXED64 || + type == UPB_DESCRIPTOR_TYPE_FIXED32 || + type == UPB_DESCRIPTOR_TYPE_SFIXED64 || + type == UPB_DESCRIPTOR_TYPE_SFIXED32) { + upb_fielddef_setintfmt(f, UPB_INTFMT_FIXED); + } else if (type == UPB_DESCRIPTOR_TYPE_SINT64 || + type == UPB_DESCRIPTOR_TYPE_SINT32) { + upb_fielddef_setintfmt(f, UPB_INTFMT_ZIGZAG); + } else { + upb_fielddef_setintfmt(f, UPB_INTFMT_VARIABLE); + } + + upb_fielddef_settagdelim(f, type == UPB_DESCRIPTOR_TYPE_GROUP); +} + +upb_descriptortype_t upb_fielddef_descriptortype(const upb_fielddef *f) { + switch (upb_fielddef_type(f)) { + case UPB_TYPE_FLOAT: return UPB_DESCRIPTOR_TYPE_FLOAT; + case UPB_TYPE_DOUBLE: return UPB_DESCRIPTOR_TYPE_DOUBLE; + case UPB_TYPE_BOOL: return UPB_DESCRIPTOR_TYPE_BOOL; + case UPB_TYPE_STRING: return UPB_DESCRIPTOR_TYPE_STRING; + case UPB_TYPE_BYTES: return UPB_DESCRIPTOR_TYPE_BYTES; + case UPB_TYPE_ENUM: return UPB_DESCRIPTOR_TYPE_ENUM; + case UPB_TYPE_INT32: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_INT32; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_SFIXED32; + case UPB_INTFMT_ZIGZAG: return UPB_DESCRIPTOR_TYPE_SINT32; + } + case UPB_TYPE_INT64: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_INT64; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_SFIXED64; + case UPB_INTFMT_ZIGZAG: return UPB_DESCRIPTOR_TYPE_SINT64; + } + case UPB_TYPE_UINT32: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_UINT32; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_FIXED32; + case UPB_INTFMT_ZIGZAG: return -1; + } + case UPB_TYPE_UINT64: + switch (upb_fielddef_intfmt(f)) { + case UPB_INTFMT_VARIABLE: return UPB_DESCRIPTOR_TYPE_UINT64; + case UPB_INTFMT_FIXED: return UPB_DESCRIPTOR_TYPE_FIXED64; + case UPB_INTFMT_ZIGZAG: return -1; + } + case UPB_TYPE_MESSAGE: + return upb_fielddef_istagdelim(f) ? + UPB_DESCRIPTOR_TYPE_GROUP : UPB_DESCRIPTOR_TYPE_MESSAGE; + } + return 0; +} + +void upb_fielddef_setisextension(upb_fielddef *f, bool is_extension) { + assert(!upb_fielddef_isfrozen(f)); + f->is_extension_ = is_extension; +} + +void upb_fielddef_setlazy(upb_fielddef *f, bool lazy) { + assert(!upb_fielddef_isfrozen(f)); + f->lazy_ = lazy; +} + +void upb_fielddef_setpacked(upb_fielddef *f, bool packed) { + assert(!upb_fielddef_isfrozen(f)); + f->packed_ = packed; +} + +void upb_fielddef_setlabel(upb_fielddef *f, upb_label_t label) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_checklabel(label)); + f->label_ = label; +} + +void upb_fielddef_setintfmt(upb_fielddef *f, upb_intfmt_t fmt) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_checkintfmt(fmt)); + f->intfmt = fmt; +} + +void upb_fielddef_settagdelim(upb_fielddef *f, bool tag_delim) { + assert(!upb_fielddef_isfrozen(f)); + f->tagdelim = tag_delim; + f->tagdelim = tag_delim; +} + +static bool checksetdefault(upb_fielddef *f, upb_fieldtype_t type) { + if (!f->type_is_set_ || upb_fielddef_isfrozen(f) || + upb_fielddef_type(f) != type) { + assert(false); + return false; + } + if (f->default_is_string) { + str_t *s = f->defaultval.bytes; + assert(s || type == UPB_TYPE_ENUM); + if (s) freestr(s); + } + f->default_is_string = false; + return true; +} + +void upb_fielddef_setdefaultint64(upb_fielddef *f, int64_t value) { + if (checksetdefault(f, UPB_TYPE_INT64)) + f->defaultval.sint = value; +} + +void upb_fielddef_setdefaultint32(upb_fielddef *f, int32_t value) { + if ((upb_fielddef_type(f) == UPB_TYPE_ENUM && + checksetdefault(f, UPB_TYPE_ENUM)) || + checksetdefault(f, UPB_TYPE_INT32)) { + f->defaultval.sint = value; + } +} + +void upb_fielddef_setdefaultuint64(upb_fielddef *f, uint64_t value) { + if (checksetdefault(f, UPB_TYPE_UINT64)) + f->defaultval.uint = value; +} + +void upb_fielddef_setdefaultuint32(upb_fielddef *f, uint32_t value) { + if (checksetdefault(f, UPB_TYPE_UINT32)) + f->defaultval.uint = value; +} + +void upb_fielddef_setdefaultbool(upb_fielddef *f, bool value) { + if (checksetdefault(f, UPB_TYPE_BOOL)) + f->defaultval.uint = value; +} + +void upb_fielddef_setdefaultfloat(upb_fielddef *f, float value) { + if (checksetdefault(f, UPB_TYPE_FLOAT)) + f->defaultval.flt = value; +} + +void upb_fielddef_setdefaultdouble(upb_fielddef *f, double value) { + if (checksetdefault(f, UPB_TYPE_DOUBLE)) + f->defaultval.dbl = value; +} + +bool upb_fielddef_setdefaultstr(upb_fielddef *f, const void *str, size_t len, + upb_status *s) { + assert(upb_fielddef_isstring(f) || f->type_ == UPB_TYPE_ENUM); + if (f->type_ == UPB_TYPE_ENUM && !upb_isident(str, len, false, s)) + return false; + + if (f->default_is_string) { + str_t *s = f->defaultval.bytes; + assert(s || f->type_ == UPB_TYPE_ENUM); + if (s) freestr(s); + } else { + assert(f->type_ == UPB_TYPE_ENUM); + } + + str_t *str2 = newstr(str, len); + f->defaultval.bytes = str2; + f->default_is_string = true; + return true; +} + +void upb_fielddef_setdefaultcstr(upb_fielddef *f, const char *str, + upb_status *s) { + assert(f->type_is_set_); + upb_fielddef_setdefaultstr(f, str, str ? strlen(str) : 0, s); +} + +bool upb_fielddef_enumhasdefaultint32(const upb_fielddef *f) { + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + int32_t val; + return enumdefaultint32(f, &val); +} + +bool upb_fielddef_enumhasdefaultstr(const upb_fielddef *f) { + assert(f->type_is_set_ && f->type_ == UPB_TYPE_ENUM); + return enumdefaultstr(f) != NULL; +} + +static bool upb_subdef_typecheck(upb_fielddef *f, const upb_def *subdef, + upb_status *s) { + if (f->type_ == UPB_TYPE_MESSAGE) { + if (upb_dyncast_msgdef(subdef)) return true; + upb_status_seterrmsg(s, "invalid subdef type for this submessage field"); + return false; + } else if (f->type_ == UPB_TYPE_ENUM) { + if (upb_dyncast_enumdef(subdef)) return true; + upb_status_seterrmsg(s, "invalid subdef type for this enum field"); + return false; + } else { + upb_status_seterrmsg(s, "only message and enum fields can have a subdef"); + return false; + } +} + +static void release_subdef(upb_fielddef *f) { + if (f->subdef_is_symbolic) { + free(f->sub.name); + } else if (f->sub.def) { + upb_unref2(f->sub.def, f); + } +} + +bool upb_fielddef_setsubdef(upb_fielddef *f, const upb_def *subdef, + upb_status *s) { + assert(!upb_fielddef_isfrozen(f)); + assert(upb_fielddef_hassubdef(f)); + if (subdef && !upb_subdef_typecheck(f, subdef, s)) return false; + release_subdef(f); + f->sub.def = subdef; + f->subdef_is_symbolic = false; + if (f->sub.def) upb_ref2(f->sub.def, f); + return true; +} + +bool upb_fielddef_setmsgsubdef(upb_fielddef *f, const upb_msgdef *subdef, + upb_status *s) { + return upb_fielddef_setsubdef(f, UPB_UPCAST(subdef), s); +} + +bool upb_fielddef_setenumsubdef(upb_fielddef *f, const upb_enumdef *subdef, + upb_status *s) { + return upb_fielddef_setsubdef(f, UPB_UPCAST(subdef), s); +} + +bool upb_fielddef_setsubdefname(upb_fielddef *f, const char *name, + upb_status *s) { + assert(!upb_fielddef_isfrozen(f)); + if (!upb_fielddef_hassubdef(f)) { + upb_status_seterrmsg(s, "field type does not accept a subdef"); + return false; + } + // TODO: validate name (upb_isident() doesn't quite work atm because this name + // may have a leading "."). + release_subdef(f); + f->sub.name = upb_strdup(name); + f->subdef_is_symbolic = true; + return true; +} + +bool upb_fielddef_issubmsg(const upb_fielddef *f) { + return upb_fielddef_type(f) == UPB_TYPE_MESSAGE; +} + +bool upb_fielddef_isstring(const upb_fielddef *f) { + return upb_fielddef_type(f) == UPB_TYPE_STRING || + upb_fielddef_type(f) == UPB_TYPE_BYTES; +} + +bool upb_fielddef_isseq(const upb_fielddef *f) { + return upb_fielddef_label(f) == UPB_LABEL_REPEATED; +} + +bool upb_fielddef_isprimitive(const upb_fielddef *f) { + return !upb_fielddef_isstring(f) && !upb_fielddef_issubmsg(f); +} + +bool upb_fielddef_ismap(const upb_fielddef *f) { + return upb_fielddef_isseq(f) && upb_fielddef_issubmsg(f) && + upb_msgdef_mapentry(upb_fielddef_msgsubdef(f)); +} + +bool upb_fielddef_hassubdef(const upb_fielddef *f) { + return upb_fielddef_issubmsg(f) || upb_fielddef_type(f) == UPB_TYPE_ENUM; +} + +static bool between(int32_t x, int32_t low, int32_t high) { + return x >= low && x <= high; +} + +bool upb_fielddef_checklabel(int32_t label) { return between(label, 1, 3); } +bool upb_fielddef_checktype(int32_t type) { return between(type, 1, 11); } +bool upb_fielddef_checkintfmt(int32_t fmt) { return between(fmt, 1, 3); } + +bool upb_fielddef_checkdescriptortype(int32_t type) { + return between(type, 1, 18); +} + +/* upb_msgdef *****************************************************************/ + +static void visitmsg(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_msgdef *m = (const upb_msgdef*)r; + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + visit(r, UPB_UPCAST2(f), closure); + } + upb_msg_oneof_iter o; + for(upb_msg_oneof_begin(&o, m); + !upb_msg_oneof_done(&o); + upb_msg_oneof_next(&o)) { + upb_oneofdef *f = upb_msg_iter_oneof(&o); + visit(r, UPB_UPCAST2(f), closure); + } +} + +static void freemsg(upb_refcounted *r) { + upb_msgdef *m = (upb_msgdef*)r; + upb_strtable_uninit(&m->ntoo); + upb_strtable_uninit(&m->ntof); + upb_inttable_uninit(&m->itof); + upb_def_uninit(UPB_UPCAST(m)); + free(m); +} + +upb_msgdef *upb_msgdef_new(const void *owner) { + static const struct upb_refcounted_vtbl vtbl = {visitmsg, freemsg}; + upb_msgdef *m = malloc(sizeof(*m)); + if (!m) return NULL; + if (!upb_def_init(UPB_UPCAST(m), UPB_DEF_MSG, &vtbl, owner)) goto err2; + if (!upb_inttable_init(&m->itof, UPB_CTYPE_PTR)) goto err3; + if (!upb_strtable_init(&m->ntof, UPB_CTYPE_PTR)) goto err2; + if (!upb_strtable_init(&m->ntoo, UPB_CTYPE_PTR)) goto err1; + m->map_entry = false; + return m; + +err1: + upb_strtable_uninit(&m->ntof); +err2: + upb_inttable_uninit(&m->itof); +err3: + free(m); + return NULL; +} + +upb_msgdef *upb_msgdef_dup(const upb_msgdef *m, const void *owner) { + upb_msgdef *newm = upb_msgdef_new(owner); + if (!newm) return NULL; + bool ok = upb_def_setfullname(UPB_UPCAST(newm), + upb_def_fullname(UPB_UPCAST(m)), NULL); + newm->map_entry = m->map_entry; + UPB_ASSERT_VAR(ok, ok); + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_fielddef_dup(upb_msg_iter_field(&i), &f); + // Fields in oneofs are dup'd below. + if (upb_fielddef_containingoneof(f)) continue; + if (!f || !upb_msgdef_addfield(newm, f, &f, NULL)) { + upb_msgdef_unref(newm, owner); + return NULL; + } + } + upb_msg_oneof_iter o; + for(upb_msg_oneof_begin(&o, m); + !upb_msg_oneof_done(&o); + upb_msg_oneof_next(&o)) { + upb_oneofdef *f = upb_oneofdef_dup(upb_msg_iter_oneof(&o), &f); + if (!f || !upb_msgdef_addoneof(newm, f, &f, NULL)) { + upb_msgdef_unref(newm, owner); + return NULL; + } + } + return newm; +} + +bool upb_msgdef_isfrozen(const upb_msgdef *m) { + return upb_def_isfrozen(UPB_UPCAST(m)); +} + +void upb_msgdef_ref(const upb_msgdef *m, const void *owner) { + upb_def_ref(UPB_UPCAST(m), owner); +} + +void upb_msgdef_unref(const upb_msgdef *m, const void *owner) { + upb_def_unref(UPB_UPCAST(m), owner); +} + +void upb_msgdef_donateref( + const upb_msgdef *m, const void *from, const void *to) { + upb_def_donateref(UPB_UPCAST(m), from, to); +} + +void upb_msgdef_checkref(const upb_msgdef *m, const void *owner) { + upb_def_checkref(UPB_UPCAST(m), owner); +} + +bool upb_msgdef_freeze(upb_msgdef *m, upb_status *status) { + upb_def *d = UPB_UPCAST(m); + return upb_def_freeze(&d, 1, status); +} + +const char *upb_msgdef_fullname(const upb_msgdef *m) { + return upb_def_fullname(UPB_UPCAST(m)); +} + +bool upb_msgdef_setfullname(upb_msgdef *m, const char *fullname, + upb_status *s) { + return upb_def_setfullname(UPB_UPCAST(m), fullname, s); +} + +// Helper: check that the field |f| is safe to add to msgdef |m|. Set an error +// on status |s| and return false if not. +static bool check_field_add(const upb_msgdef *m, const upb_fielddef *f, + upb_status *s) { + if (upb_fielddef_containingtype(f) != NULL) { + upb_status_seterrmsg(s, "fielddef already belongs to a message"); + return false; + } else if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) { + upb_status_seterrmsg(s, "field name or number were not set"); + return false; + } else if (upb_msgdef_ntofz(m, upb_fielddef_name(f)) || + upb_msgdef_itof(m, upb_fielddef_number(f))) { + upb_status_seterrmsg(s, "duplicate field name or number for field"); + return false; + } + return true; +} + +static void add_field(upb_msgdef *m, upb_fielddef *f, const void *ref_donor) { + release_containingtype(f); + f->msg.def = m; + f->msg_is_symbolic = false; + upb_inttable_insert(&m->itof, upb_fielddef_number(f), upb_value_ptr(f)); + upb_strtable_insert(&m->ntof, upb_fielddef_name(f), upb_value_ptr(f)); + upb_ref2(f, m); + upb_ref2(m, f); + if (ref_donor) upb_fielddef_unref(f, ref_donor); +} + +bool upb_msgdef_addfield(upb_msgdef *m, upb_fielddef *f, const void *ref_donor, + upb_status *s) { + // TODO: extensions need to have a separate namespace, because proto2 allows a + // top-level extension (ie. one not in any package) to have the same name as a + // field from the message. + // + // This also implies that there needs to be a separate lookup-by-name method + // for extensions. It seems desirable for iteration to return both extensions + // and non-extensions though. + // + // We also need to validate that the field number is in an extension range iff + // it is an extension. + + // This method is idempotent. Check if |f| is already part of this msgdef and + // return immediately if so. + if (upb_fielddef_containingtype(f) == m) { + return true; + } + + // Check constraints for all fields before performing any action. + if (!check_field_add(m, f, s)) { + return false; + } else if (upb_fielddef_containingoneof(f) != NULL) { + // Fields in a oneof can only be added by adding the oneof to the msgdef. + upb_status_seterrmsg(s, "fielddef is part of a oneof"); + return false; + } + + // Constraint checks ok, perform the action. + add_field(m, f, ref_donor); + return true; +} + +bool upb_msgdef_addoneof(upb_msgdef *m, upb_oneofdef *o, const void *ref_donor, + upb_status *s) { + // Check various conditions that would prevent this oneof from being added. + if (upb_oneofdef_containingtype(o)) { + upb_status_seterrmsg(s, "oneofdef already belongs to a message"); + return false; + } else if (upb_oneofdef_name(o) == NULL) { + upb_status_seterrmsg(s, "oneofdef name was not set"); + return false; + } else if (upb_msgdef_ntooz(m, upb_oneofdef_name(o))) { + upb_status_seterrmsg(s, "duplicate oneof name"); + return false; + } + + // Check that all of the oneof's fields do not conflict with names or numbers + // of fields already in the message. + upb_oneof_iter it; + for (upb_oneof_begin(&it, o); !upb_oneof_done(&it); upb_oneof_next(&it)) { + const upb_fielddef *f = upb_oneof_iter_field(&it); + if (!check_field_add(m, f, s)) { + return false; + } + } + + // Everything checks out -- commit now. + + // Add oneof itself first. + o->parent = m; + upb_strtable_insert(&m->ntoo, upb_oneofdef_name(o), upb_value_ptr(o)); + upb_ref2(o, m); + upb_ref2(m, o); + + // Add each field of the oneof directly to the msgdef. + for (upb_oneof_begin(&it, o); !upb_oneof_done(&it); upb_oneof_next(&it)) { + upb_fielddef *f = upb_oneof_iter_field(&it); + add_field(m, f, NULL); + } + + if (ref_donor) upb_oneofdef_unref(o, ref_donor); + + return true; +} + +const upb_fielddef *upb_msgdef_itof(const upb_msgdef *m, uint32_t i) { + upb_value val; + return upb_inttable_lookup32(&m->itof, i, &val) ? + upb_value_getptr(val) : NULL; +} + +const upb_fielddef *upb_msgdef_ntof(const upb_msgdef *m, const char *name, + size_t len) { + upb_value val; + return upb_strtable_lookup2(&m->ntof, name, len, &val) ? + upb_value_getptr(val) : NULL; +} + +const upb_oneofdef *upb_msgdef_ntoo(const upb_msgdef *m, const char *name, + size_t len) { + upb_value val; + return upb_strtable_lookup2(&m->ntoo, name, len, &val) ? + upb_value_getptr(val) : NULL; +} + +int upb_msgdef_numfields(const upb_msgdef *m) { + return upb_strtable_count(&m->ntof); +} + +int upb_msgdef_numoneofs(const upb_msgdef *m) { + return upb_strtable_count(&m->ntoo); +} + +void upb_msgdef_setmapentry(upb_msgdef *m, bool map_entry) { + assert(!upb_msgdef_isfrozen(m)); + m->map_entry = map_entry; +} + +bool upb_msgdef_mapentry(const upb_msgdef *m) { + return m->map_entry; +} + +void upb_msg_field_begin(upb_msg_field_iter *iter, const upb_msgdef *m) { + upb_inttable_begin(iter, &m->itof); +} + +void upb_msg_field_next(upb_msg_field_iter *iter) { upb_inttable_next(iter); } + +bool upb_msg_field_done(const upb_msg_field_iter *iter) { + return upb_inttable_done(iter); +} + +upb_fielddef *upb_msg_iter_field(const upb_msg_field_iter *iter) { + return (upb_fielddef*)upb_value_getptr(upb_inttable_iter_value(iter)); +} + +void upb_msg_field_iter_setdone(upb_msg_field_iter *iter) { + upb_inttable_iter_setdone(iter); +} + +void upb_msg_oneof_begin(upb_msg_oneof_iter *iter, const upb_msgdef *m) { + upb_strtable_begin(iter, &m->ntoo); +} + +void upb_msg_oneof_next(upb_msg_oneof_iter *iter) { upb_strtable_next(iter); } + +bool upb_msg_oneof_done(const upb_msg_oneof_iter *iter) { + return upb_strtable_done(iter); +} + +upb_oneofdef *upb_msg_iter_oneof(const upb_msg_oneof_iter *iter) { + return (upb_oneofdef*)upb_value_getptr(upb_strtable_iter_value(iter)); +} + +void upb_msg_oneof_iter_setdone(upb_msg_oneof_iter *iter) { + upb_strtable_iter_setdone(iter); +} + +/* upb_oneofdef ***************************************************************/ + +static void visitoneof(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_oneofdef *o = (const upb_oneofdef*)r; + upb_oneof_iter i; + for (upb_oneof_begin(&i, o); !upb_oneof_done(&i); upb_oneof_next(&i)) { + const upb_fielddef *f = upb_oneof_iter_field(&i); + visit(r, UPB_UPCAST2(f), closure); + } + if (o->parent) { + visit(r, UPB_UPCAST2(o->parent), closure); + } +} + +static void freeoneof(upb_refcounted *r) { + upb_oneofdef *o = (upb_oneofdef*)r; + upb_strtable_uninit(&o->ntof); + upb_inttable_uninit(&o->itof); + upb_def_uninit(UPB_UPCAST(o)); + free(o); +} + +upb_oneofdef *upb_oneofdef_new(const void *owner) { + static const struct upb_refcounted_vtbl vtbl = {visitoneof, freeoneof}; + upb_oneofdef *o = malloc(sizeof(*o)); + o->parent = NULL; + if (!o) return NULL; + if (!upb_def_init(UPB_UPCAST(o), UPB_DEF_ONEOF, &vtbl, owner)) goto err2; + if (!upb_inttable_init(&o->itof, UPB_CTYPE_PTR)) goto err2; + if (!upb_strtable_init(&o->ntof, UPB_CTYPE_PTR)) goto err1; + return o; + +err1: + upb_inttable_uninit(&o->itof); +err2: + free(o); + return NULL; +} + +upb_oneofdef *upb_oneofdef_dup(const upb_oneofdef *o, const void *owner) { + upb_oneofdef *newo = upb_oneofdef_new(owner); + if (!newo) return NULL; + bool ok = upb_def_setfullname(UPB_UPCAST(newo), + upb_def_fullname(UPB_UPCAST(o)), NULL); + UPB_ASSERT_VAR(ok, ok); + upb_oneof_iter i; + for (upb_oneof_begin(&i, o); !upb_oneof_done(&i); upb_oneof_next(&i)) { + upb_fielddef *f = upb_fielddef_dup(upb_oneof_iter_field(&i), &f); + if (!f || !upb_oneofdef_addfield(newo, f, &f, NULL)) { + upb_oneofdef_unref(newo, owner); + return NULL; + } + } + return newo; +} + +bool upb_oneofdef_isfrozen(const upb_oneofdef *o) { + return upb_def_isfrozen(UPB_UPCAST(o)); +} + +void upb_oneofdef_ref(const upb_oneofdef *o, const void *owner) { + upb_def_ref(UPB_UPCAST(o), owner); +} + +void upb_oneofdef_unref(const upb_oneofdef *o, const void *owner) { + upb_def_unref(UPB_UPCAST(o), owner); +} + +void upb_oneofdef_donateref(const upb_oneofdef *o, const void *from, + const void *to) { + upb_def_donateref(UPB_UPCAST(o), from, to); +} + +void upb_oneofdef_checkref(const upb_oneofdef *o, const void *owner) { + upb_def_checkref(UPB_UPCAST(o), owner); +} + +const char *upb_oneofdef_name(const upb_oneofdef *o) { + return upb_def_fullname(UPB_UPCAST(o)); +} + +bool upb_oneofdef_setname(upb_oneofdef *o, const char *fullname, + upb_status *s) { + if (upb_oneofdef_containingtype(o)) { + upb_status_seterrmsg(s, "oneof already added to a message"); + return false; + } + return upb_def_setfullname(UPB_UPCAST(o), fullname, s); +} + +const upb_msgdef *upb_oneofdef_containingtype(const upb_oneofdef *o) { + return o->parent; +} + +int upb_oneofdef_numfields(const upb_oneofdef *o) { + return upb_strtable_count(&o->ntof); +} + +bool upb_oneofdef_addfield(upb_oneofdef *o, upb_fielddef *f, + const void *ref_donor, + upb_status *s) { + assert(!upb_oneofdef_isfrozen(o)); + assert(!o->parent || !upb_msgdef_isfrozen(o->parent)); + + // This method is idempotent. Check if |f| is already part of this oneofdef + // and return immediately if so. + if (upb_fielddef_containingoneof(f) == o) { + return true; + } + + // The field must have an OPTIONAL label. + if (upb_fielddef_label(f) != UPB_LABEL_OPTIONAL) { + upb_status_seterrmsg(s, "fields in oneof must have OPTIONAL label"); + return false; + } + + // Check that no field with this name or number exists already in the oneof. + // Also check that the field is not already part of a oneof. + if (upb_fielddef_name(f) == NULL || upb_fielddef_number(f) == 0) { + upb_status_seterrmsg(s, "field name or number were not set"); + return false; + } else if (upb_oneofdef_itof(o, upb_fielddef_number(f)) || + upb_oneofdef_ntofz(o, upb_fielddef_name(f))) { + upb_status_seterrmsg(s, "duplicate field name or number"); + return false; + } else if (upb_fielddef_containingoneof(f) != NULL) { + upb_status_seterrmsg(s, "fielddef already belongs to a oneof"); + return false; + } + + // We allow adding a field to the oneof either if the field is not part of a + // msgdef, or if it is and we are also part of the same msgdef. + if (o->parent == NULL) { + // If we're not in a msgdef, the field cannot be either. Otherwise we would + // need to magically add this oneof to a msgdef to remain consistent, which + // is surprising behavior. + if (upb_fielddef_containingtype(f) != NULL) { + upb_status_seterrmsg(s, "fielddef already belongs to a message, but " + "oneof does not"); + return false; + } + } else { + // If we're in a msgdef, the user can add fields that either aren't in any + // msgdef (in which case they're added to our msgdef) or already a part of + // our msgdef. + if (upb_fielddef_containingtype(f) != NULL && + upb_fielddef_containingtype(f) != o->parent) { + upb_status_seterrmsg(s, "fielddef belongs to a different message " + "than oneof"); + return false; + } + } + + // Commit phase. First add the field to our parent msgdef, if any, because + // that may fail; then add the field to our own tables. + + if (o->parent != NULL && upb_fielddef_containingtype(f) == NULL) { + if (!upb_msgdef_addfield((upb_msgdef*)o->parent, f, NULL, s)) { + return false; + } + } + + release_containingtype(f); + f->oneof = o; + upb_inttable_insert(&o->itof, upb_fielddef_number(f), upb_value_ptr(f)); + upb_strtable_insert(&o->ntof, upb_fielddef_name(f), upb_value_ptr(f)); + upb_ref2(f, o); + upb_ref2(o, f); + if (ref_donor) upb_fielddef_unref(f, ref_donor); + + return true; +} + +const upb_fielddef *upb_oneofdef_ntof(const upb_oneofdef *o, + const char *name, size_t length) { + upb_value val; + return upb_strtable_lookup2(&o->ntof, name, length, &val) ? + upb_value_getptr(val) : NULL; +} + +const upb_fielddef *upb_oneofdef_itof(const upb_oneofdef *o, uint32_t num) { + upb_value val; + return upb_inttable_lookup32(&o->itof, num, &val) ? + upb_value_getptr(val) : NULL; +} + +void upb_oneof_begin(upb_oneof_iter *iter, const upb_oneofdef *o) { + upb_inttable_begin(iter, &o->itof); +} + +void upb_oneof_next(upb_oneof_iter *iter) { + upb_inttable_next(iter); +} + +bool upb_oneof_done(upb_oneof_iter *iter) { + return upb_inttable_done(iter); +} + +upb_fielddef *upb_oneof_iter_field(const upb_oneof_iter *iter) { + return (upb_fielddef*)upb_value_getptr(upb_inttable_iter_value(iter)); +} + +void upb_oneof_iter_setdone(upb_oneof_iter *iter) { + upb_inttable_iter_setdone(iter); +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2011-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * TODO(haberman): it's unclear whether a lot of the consistency checks should + * assert() or return false. + */ + + +#include +#include + + +// Defined for the sole purpose of having a unique pointer value for +// UPB_NO_CLOSURE. +char _upb_noclosure; + +static void freehandlers(upb_refcounted *r) { + upb_handlers *h = (upb_handlers*)r; + + upb_inttable_iter i; + upb_inttable_begin(&i, &h->cleanup_); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + void *val = (void*)upb_inttable_iter_key(&i); + upb_value func_val = upb_inttable_iter_value(&i); + upb_handlerfree *func = upb_value_getfptr(func_val); + func(val); + } + + upb_inttable_uninit(&h->cleanup_); + upb_msgdef_unref(h->msg, h); + free(h->sub); + free(h); +} + +static void visithandlers(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_handlers *h = (const upb_handlers*)r; + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, h->msg); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + if (!upb_fielddef_issubmsg(f)) continue; + const upb_handlers *sub = upb_handlers_getsubhandlers(h, f); + if (sub) visit(r, UPB_UPCAST(sub), closure); + } +} + +static const struct upb_refcounted_vtbl vtbl = {visithandlers, freehandlers}; + +typedef struct { + upb_inttable tab; // maps upb_msgdef* -> upb_handlers*. + upb_handlers_callback *callback; + const void *closure; +} dfs_state; + +// TODO(haberman): discard upb_handlers* objects that do not actually have any +// handlers set and cannot reach any upb_handlers* object that does. This is +// slightly tricky to do correctly. +static upb_handlers *newformsg(const upb_msgdef *m, const void *owner, + dfs_state *s) { + upb_handlers *h = upb_handlers_new(m, owner); + if (!h) return NULL; + if (!upb_inttable_insertptr(&s->tab, m, upb_value_ptr(h))) goto oom; + + s->callback(s->closure, h); + + // For each submessage field, get or create a handlers object and set it as + // the subhandlers. + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + if (!upb_fielddef_issubmsg(f)) continue; + + const upb_msgdef *subdef = upb_downcast_msgdef(upb_fielddef_subdef(f)); + upb_value subm_ent; + if (upb_inttable_lookupptr(&s->tab, subdef, &subm_ent)) { + upb_handlers_setsubhandlers(h, f, upb_value_getptr(subm_ent)); + } else { + upb_handlers *sub_mh = newformsg(subdef, &sub_mh, s); + if (!sub_mh) goto oom; + upb_handlers_setsubhandlers(h, f, sub_mh); + upb_handlers_unref(sub_mh, &sub_mh); + } + } + return h; + +oom: + upb_handlers_unref(h, owner); + return NULL; +} + +// Given a selector for a STARTSUBMSG handler, resolves to a pointer to the +// subhandlers for this submessage field. +#define SUBH(h, selector) (h->sub[selector]) + +// The selector for a submessage field is the field index. +#define SUBH_F(h, f) SUBH(h, f->index_) + +static int32_t trygetsel(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + upb_selector_t sel; + assert(!upb_handlers_isfrozen(h)); + if (upb_handlers_msgdef(h) != upb_fielddef_containingtype(f)) { + upb_status_seterrf( + &h->status_, "type mismatch: field %s does not belong to message %s", + upb_fielddef_name(f), upb_msgdef_fullname(upb_handlers_msgdef(h))); + return -1; + } + if (!upb_handlers_getselector(f, type, &sel)) { + upb_status_seterrf( + &h->status_, + "type mismatch: cannot register handler type %d for field %s", + type, upb_fielddef_name(f)); + return -1; + } + return sel; +} + +static upb_selector_t handlers_getsel(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + int32_t sel = trygetsel(h, f, type); + assert(sel >= 0); + return sel; +} + +static const void **returntype(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + return &h->table[handlers_getsel(h, f, type)].attr.return_closure_type_; +} + +static bool doset(upb_handlers *h, int32_t sel, const upb_fielddef *f, + upb_handlertype_t type, upb_func *func, + upb_handlerattr *attr) { + assert(!upb_handlers_isfrozen(h)); + + if (sel < 0) { + upb_status_seterrmsg(&h->status_, + "incorrect handler type for this field."); + return false; + } + + if (h->table[sel].func) { + upb_status_seterrmsg(&h->status_, + "cannot change handler once it has been set."); + return false; + } + + upb_handlerattr set_attr = UPB_HANDLERATTR_INITIALIZER; + if (attr) { + set_attr = *attr; + } + + // Check that the given closure type matches the closure type that has been + // established for this context (if any). + const void *closure_type = upb_handlerattr_closuretype(&set_attr); + const void **context_closure_type; + + if (type == UPB_HANDLER_STRING) { + context_closure_type = returntype(h, f, UPB_HANDLER_STARTSTR); + } else if (f && upb_fielddef_isseq(f) && + type != UPB_HANDLER_STARTSEQ && + type != UPB_HANDLER_ENDSEQ) { + context_closure_type = returntype(h, f, UPB_HANDLER_STARTSEQ); + } else { + context_closure_type = &h->top_closure_type; + } + + if (closure_type && *context_closure_type && + closure_type != *context_closure_type) { + // TODO(haberman): better message for debugging. + upb_status_seterrmsg(&h->status_, "closure type does not match"); + return false; + } + + if (closure_type) + *context_closure_type = closure_type; + + // If this is a STARTSEQ or STARTSTR handler, check that the returned pointer + // matches any pre-existing expectations about what type is expected. + if (type == UPB_HANDLER_STARTSEQ || type == UPB_HANDLER_STARTSTR) { + const void *return_type = upb_handlerattr_returnclosuretype(&set_attr); + const void *table_return_type = + upb_handlerattr_returnclosuretype(&h->table[sel].attr); + if (return_type && table_return_type && return_type != table_return_type) { + upb_status_seterrmsg(&h->status_, "closure return type does not match"); + return false; + } + + if (table_return_type && !return_type) + upb_handlerattr_setreturnclosuretype(&set_attr, table_return_type); + } + + h->table[sel].func = (upb_func*)func; + h->table[sel].attr = set_attr; + return true; +} + +// Returns the effective closure type for this handler (which will propagate +// from outer frames if this frame has no START* handler). Not implemented for +// UPB_HANDLER_STRING at the moment since this is not needed. Returns NULL is +// the effective closure type is unspecified (either no handler was registered +// to specify it or the handler that was registered did not specify the closure +// type). +const void *effective_closure_type(upb_handlers *h, const upb_fielddef *f, + upb_handlertype_t type) { + assert(type != UPB_HANDLER_STRING); + const void *ret = h->top_closure_type; + upb_selector_t sel; + if (upb_fielddef_isseq(f) && + type != UPB_HANDLER_STARTSEQ && + type != UPB_HANDLER_ENDSEQ && + h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSEQ)].func) { + ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr); + } + + if (type == UPB_HANDLER_STRING && + h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSTR)].func) { + ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr); + } + + // The effective type of the submessage; not used yet. + // if (type == SUBMESSAGE && + // h->table[sel = handlers_getsel(h, f, UPB_HANDLER_STARTSUBMSG)].func) { + // ret = upb_handlerattr_returnclosuretype(&h->table[sel].attr); + // } + + return ret; +} + +// Checks whether the START* handler specified by f & type is missing even +// though it is required to convert the established type of an outer frame +// ("closure_type") into the established type of an inner frame (represented in +// the return closure type of this handler's attr. +bool checkstart(upb_handlers *h, const upb_fielddef *f, upb_handlertype_t type, + upb_status *status) { + upb_selector_t sel = handlers_getsel(h, f, type); + if (h->table[sel].func) return true; + const void *closure_type = effective_closure_type(h, f, type); + const upb_handlerattr *attr = &h->table[sel].attr; + const void *return_closure_type = upb_handlerattr_returnclosuretype(attr); + if (closure_type && return_closure_type && + closure_type != return_closure_type) { + upb_status_seterrf(status, + "expected start handler to return sub type for field %f", + upb_fielddef_name(f)); + return false; + } + return true; +} + +/* Public interface ***********************************************************/ + +bool upb_handlers_isfrozen(const upb_handlers *h) { + return upb_refcounted_isfrozen(UPB_UPCAST(h)); +} + +void upb_handlers_ref(const upb_handlers *h, const void *owner) { + upb_refcounted_ref(UPB_UPCAST(h), owner); +} + +void upb_handlers_unref(const upb_handlers *h, const void *owner) { + upb_refcounted_unref(UPB_UPCAST(h), owner); +} + +void upb_handlers_donateref( + const upb_handlers *h, const void *from, const void *to) { + upb_refcounted_donateref(UPB_UPCAST(h), from, to); +} + +void upb_handlers_checkref(const upb_handlers *h, const void *owner) { + upb_refcounted_checkref(UPB_UPCAST(h), owner); +} + +upb_handlers *upb_handlers_new(const upb_msgdef *md, const void *owner) { + assert(upb_msgdef_isfrozen(md)); + + int extra = sizeof(upb_handlers_tabent) * (md->selector_count - 1); + upb_handlers *h = calloc(sizeof(*h) + extra, 1); + if (!h) return NULL; + + h->msg = md; + upb_msgdef_ref(h->msg, h); + upb_status_clear(&h->status_); + h->sub = calloc(md->submsg_field_count, sizeof(*h->sub)); + if (!h->sub) goto oom; + if (!upb_refcounted_init(UPB_UPCAST(h), &vtbl, owner)) goto oom; + if (!upb_inttable_init(&h->cleanup_, UPB_CTYPE_FPTR)) goto oom; + + // calloc() above initialized all handlers to NULL. + return h; + +oom: + freehandlers(UPB_UPCAST(h)); + return NULL; +} + +const upb_handlers *upb_handlers_newfrozen(const upb_msgdef *m, + const void *owner, + upb_handlers_callback *callback, + const void *closure) { + dfs_state state; + state.callback = callback; + state.closure = closure; + if (!upb_inttable_init(&state.tab, UPB_CTYPE_PTR)) return NULL; + + upb_handlers *ret = newformsg(m, owner, &state); + + upb_inttable_uninit(&state.tab); + if (!ret) return NULL; + + upb_refcounted *r = UPB_UPCAST(ret); + bool ok = upb_refcounted_freeze(&r, 1, NULL, UPB_MAX_HANDLER_DEPTH); + UPB_ASSERT_VAR(ok, ok); + + return ret; +} + +const upb_status *upb_handlers_status(upb_handlers *h) { + assert(!upb_handlers_isfrozen(h)); + return &h->status_; +} + +void upb_handlers_clearerr(upb_handlers *h) { + assert(!upb_handlers_isfrozen(h)); + upb_status_clear(&h->status_); +} + +#define SETTER(name, handlerctype, handlertype) \ + bool upb_handlers_set ## name(upb_handlers *h, const upb_fielddef *f, \ + handlerctype func, upb_handlerattr *attr) { \ + int32_t sel = trygetsel(h, f, handlertype); \ + return doset(h, sel, f, handlertype, (upb_func*)func, attr); \ + } + +SETTER(int32, upb_int32_handlerfunc*, UPB_HANDLER_INT32); +SETTER(int64, upb_int64_handlerfunc*, UPB_HANDLER_INT64); +SETTER(uint32, upb_uint32_handlerfunc*, UPB_HANDLER_UINT32); +SETTER(uint64, upb_uint64_handlerfunc*, UPB_HANDLER_UINT64); +SETTER(float, upb_float_handlerfunc*, UPB_HANDLER_FLOAT); +SETTER(double, upb_double_handlerfunc*, UPB_HANDLER_DOUBLE); +SETTER(bool, upb_bool_handlerfunc*, UPB_HANDLER_BOOL); +SETTER(startstr, upb_startstr_handlerfunc*, UPB_HANDLER_STARTSTR); +SETTER(string, upb_string_handlerfunc*, UPB_HANDLER_STRING); +SETTER(endstr, upb_endfield_handlerfunc*, UPB_HANDLER_ENDSTR); +SETTER(startseq, upb_startfield_handlerfunc*, UPB_HANDLER_STARTSEQ); +SETTER(startsubmsg, upb_startfield_handlerfunc*, UPB_HANDLER_STARTSUBMSG); +SETTER(endsubmsg, upb_endfield_handlerfunc*, UPB_HANDLER_ENDSUBMSG); +SETTER(endseq, upb_endfield_handlerfunc*, UPB_HANDLER_ENDSEQ); + +#undef SETTER + +bool upb_handlers_setstartmsg(upb_handlers *h, upb_startmsg_handlerfunc *func, + upb_handlerattr *attr) { + return doset(h, UPB_STARTMSG_SELECTOR, NULL, UPB_HANDLER_INT32, + (upb_func *)func, attr); +} + +bool upb_handlers_setendmsg(upb_handlers *h, upb_endmsg_handlerfunc *func, + upb_handlerattr *attr) { + assert(!upb_handlers_isfrozen(h)); + return doset(h, UPB_ENDMSG_SELECTOR, NULL, UPB_HANDLER_INT32, + (upb_func *)func, attr); +} + +bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f, + const upb_handlers *sub) { + assert(sub); + assert(!upb_handlers_isfrozen(h)); + assert(upb_fielddef_issubmsg(f)); + if (SUBH_F(h, f)) return false; // Can't reset. + if (UPB_UPCAST(upb_handlers_msgdef(sub)) != upb_fielddef_subdef(f)) { + return false; + } + SUBH_F(h, f) = sub; + upb_ref2(sub, h); + return true; +} + +const upb_handlers *upb_handlers_getsubhandlers(const upb_handlers *h, + const upb_fielddef *f) { + assert(upb_fielddef_issubmsg(f)); + return SUBH_F(h, f); +} + +bool upb_handlers_getattr(const upb_handlers *h, upb_selector_t sel, + upb_handlerattr *attr) { + if (!upb_handlers_gethandler(h, sel)) + return false; + *attr = h->table[sel].attr; + return true; +} + +const upb_handlers *upb_handlers_getsubhandlers_sel(const upb_handlers *h, + upb_selector_t sel) { + // STARTSUBMSG selector in sel is the field's selector base. + return SUBH(h, sel - UPB_STATIC_SELECTOR_COUNT); +} + +const upb_msgdef *upb_handlers_msgdef(const upb_handlers *h) { return h->msg; } + +bool upb_handlers_addcleanup(upb_handlers *h, void *p, upb_handlerfree *func) { + if (upb_inttable_lookupptr(&h->cleanup_, p, NULL)) { + return false; + } + bool ok = upb_inttable_insertptr(&h->cleanup_, p, upb_value_fptr(func)); + UPB_ASSERT_VAR(ok, ok); + return true; +} + + +/* "Static" methods ***********************************************************/ + +bool upb_handlers_freeze(upb_handlers *const*handlers, int n, upb_status *s) { + // TODO: verify we have a transitive closure. + for (int i = 0; i < n; i++) { + upb_handlers *h = handlers[i]; + + if (!upb_ok(&h->status_)) { + upb_status_seterrf(s, "handlers for message %s had error status: %s", + upb_msgdef_fullname(upb_handlers_msgdef(h)), + upb_status_errmsg(&h->status_)); + return false; + } + + // Check that there are no closure mismatches due to missing Start* handlers + // or subhandlers with different type-level types. + upb_msg_field_iter j; + for(upb_msg_field_begin(&j, h->msg); + !upb_msg_field_done(&j); + upb_msg_field_next(&j)) { + + const upb_fielddef *f = upb_msg_iter_field(&j); + if (upb_fielddef_isseq(f)) { + if (!checkstart(h, f, UPB_HANDLER_STARTSEQ, s)) + return false; + } + + if (upb_fielddef_isstring(f)) { + if (!checkstart(h, f, UPB_HANDLER_STARTSTR, s)) + return false; + } + + if (upb_fielddef_issubmsg(f)) { + bool hashandler = false; + if (upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_STARTSUBMSG)) || + upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_ENDSUBMSG))) { + hashandler = true; + } + + if (upb_fielddef_isseq(f) && + (upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_STARTSEQ)) || + upb_handlers_gethandler( + h, handlers_getsel(h, f, UPB_HANDLER_ENDSEQ)))) { + hashandler = true; + } + + if (hashandler && !upb_handlers_getsubhandlers(h, f)) { + // For now we add an empty subhandlers in this case. It makes the + // decoder code generator simpler, because it only has to handle two + // cases (submessage has handlers or not) as opposed to three + // (submessage has handlers in enclosing message but no subhandlers). + // + // This makes parsing less efficient in the case that we want to + // notice a submessage but skip its contents (like if we're testing + // for submessage presence or counting the number of repeated + // submessages). In this case we will end up parsing the submessage + // field by field and throwing away the results for each, instead of + // skipping the whole delimited thing at once. If this is an issue we + // can revisit it, but do remember that this only arises when you have + // handlers (startseq/startsubmsg/endsubmsg/endseq) set for the + // submessage but no subhandlers. The uses cases for this are + // limited. + upb_handlers *sub = upb_handlers_new(upb_fielddef_msgsubdef(f), &sub); + upb_handlers_setsubhandlers(h, f, sub); + upb_handlers_unref(sub, &sub); + } + + // TODO(haberman): check type of submessage. + // This is slightly tricky; also consider whether we should check that + // they match at setsubhandlers time. + } + } + } + + if (!upb_refcounted_freeze((upb_refcounted*const*)handlers, n, s, + UPB_MAX_HANDLER_DEPTH)) { + return false; + } + + return true; +} + +upb_handlertype_t upb_handlers_getprimitivehandlertype(const upb_fielddef *f) { + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: + case UPB_TYPE_ENUM: return UPB_HANDLER_INT32; + case UPB_TYPE_INT64: return UPB_HANDLER_INT64; + case UPB_TYPE_UINT32: return UPB_HANDLER_UINT32; + case UPB_TYPE_UINT64: return UPB_HANDLER_UINT64; + case UPB_TYPE_FLOAT: return UPB_HANDLER_FLOAT; + case UPB_TYPE_DOUBLE: return UPB_HANDLER_DOUBLE; + case UPB_TYPE_BOOL: return UPB_HANDLER_BOOL; + default: assert(false); return -1; // Invalid input. + } +} + +bool upb_handlers_getselector(const upb_fielddef *f, upb_handlertype_t type, + upb_selector_t *s) { + switch (type) { + case UPB_HANDLER_INT32: + case UPB_HANDLER_INT64: + case UPB_HANDLER_UINT32: + case UPB_HANDLER_UINT64: + case UPB_HANDLER_FLOAT: + case UPB_HANDLER_DOUBLE: + case UPB_HANDLER_BOOL: + if (!upb_fielddef_isprimitive(f) || + upb_handlers_getprimitivehandlertype(f) != type) + return false; + *s = f->selector_base; + break; + case UPB_HANDLER_STRING: + if (upb_fielddef_isstring(f)) { + *s = f->selector_base; + } else if (upb_fielddef_lazy(f)) { + *s = f->selector_base + 3; + } else { + return false; + } + break; + case UPB_HANDLER_STARTSTR: + if (upb_fielddef_isstring(f) || upb_fielddef_lazy(f)) { + *s = f->selector_base + 1; + } else { + return false; + } + break; + case UPB_HANDLER_ENDSTR: + if (upb_fielddef_isstring(f) || upb_fielddef_lazy(f)) { + *s = f->selector_base + 2; + } else { + return false; + } + break; + case UPB_HANDLER_STARTSEQ: + if (!upb_fielddef_isseq(f)) return false; + *s = f->selector_base - 2; + break; + case UPB_HANDLER_ENDSEQ: + if (!upb_fielddef_isseq(f)) return false; + *s = f->selector_base - 1; + break; + case UPB_HANDLER_STARTSUBMSG: + if (!upb_fielddef_issubmsg(f)) return false; + // Selectors for STARTSUBMSG are at the beginning of the table so that the + // selector can also be used as an index into the "sub" array of + // subhandlers. The indexes for the two into these two tables are the + // same, except that in the handler table the static selectors come first. + *s = f->index_ + UPB_STATIC_SELECTOR_COUNT; + break; + case UPB_HANDLER_ENDSUBMSG: + if (!upb_fielddef_issubmsg(f)) return false; + *s = f->selector_base; + break; + } + assert(*s < upb_fielddef_containingtype(f)->selector_count); + return true; +} + +uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f) { + return upb_fielddef_isseq(f) ? 2 : 0; +} + +uint32_t upb_handlers_selectorcount(const upb_fielddef *f) { + uint32_t ret = 1; + if (upb_fielddef_isseq(f)) ret += 2; // STARTSEQ/ENDSEQ + if (upb_fielddef_isstring(f)) ret += 2; // [STRING]/STARTSTR/ENDSTR + if (upb_fielddef_issubmsg(f)) { + // ENDSUBMSG (STARTSUBMSG is at table beginning) + ret += 0; + if (upb_fielddef_lazy(f)) { + // STARTSTR/ENDSTR/STRING (for lazy) + ret += 3; + } + } + return ret; +} + + +/* upb_handlerattr ************************************************************/ + +void upb_handlerattr_init(upb_handlerattr *attr) { + upb_handlerattr from = UPB_HANDLERATTR_INITIALIZER; + memcpy(attr, &from, sizeof(*attr)); +} + +void upb_handlerattr_uninit(upb_handlerattr *attr) { + UPB_UNUSED(attr); +} + +bool upb_handlerattr_sethandlerdata(upb_handlerattr *attr, const void *hd) { + attr->handler_data_ = hd; + return true; +} + +bool upb_handlerattr_setclosuretype(upb_handlerattr *attr, const void *type) { + attr->closure_type_ = type; + return true; +} + +const void *upb_handlerattr_closuretype(const upb_handlerattr *attr) { + return attr->closure_type_; +} + +bool upb_handlerattr_setreturnclosuretype(upb_handlerattr *attr, + const void *type) { + attr->return_closure_type_ = type; + return true; +} + +const void *upb_handlerattr_returnclosuretype(const upb_handlerattr *attr) { + return attr->return_closure_type_; +} + +bool upb_handlerattr_setalwaysok(upb_handlerattr *attr, bool alwaysok) { + attr->alwaysok_ = alwaysok; + return true; +} + +bool upb_handlerattr_alwaysok(const upb_handlerattr *attr) { + return attr->alwaysok_; +} + +/* upb_bufhandle **************************************************************/ + +size_t upb_bufhandle_objofs(const upb_bufhandle *h) { + return h->objofs_; +} + +/* upb_byteshandler ***********************************************************/ + +void upb_byteshandler_init(upb_byteshandler* h) { + memset(h, 0, sizeof(*h)); +} + +// For when we support handlerfree callbacks. +void upb_byteshandler_uninit(upb_byteshandler* h) { + UPB_UNUSED(h); +} + +bool upb_byteshandler_setstartstr(upb_byteshandler *h, + upb_startstr_handlerfunc *func, void *d) { + h->table[UPB_STARTSTR_SELECTOR].func = (upb_func*)func; + h->table[UPB_STARTSTR_SELECTOR].attr.handler_data_ = d; + return true; +} + +bool upb_byteshandler_setstring(upb_byteshandler *h, + upb_string_handlerfunc *func, void *d) { + h->table[UPB_STRING_SELECTOR].func = (upb_func*)func; + h->table[UPB_STRING_SELECTOR].attr.handler_data_ = d; + return true; +} + +bool upb_byteshandler_setendstr(upb_byteshandler *h, + upb_endfield_handlerfunc *func, void *d) { + h->table[UPB_ENDSTR_SELECTOR].func = (upb_func*)func; + h->table[UPB_ENDSTR_SELECTOR].attr.handler_data_ = d; + return true; +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Our key invariants are: + * 1. reference cycles never span groups + * 2. for ref2(to, from), we increment to's count iff group(from) != group(to) + * + * The previous two are how we avoid leaking cycles. Other important + * invariants are: + * 3. for mutable objects "from" and "to", if there exists a ref2(to, from) + * this implies group(from) == group(to). (In practice, what we implement + * is even stronger; "from" and "to" will share a group if there has *ever* + * been a ref2(to, from), but all that is necessary for correctness is the + * weaker one). + * 4. mutable and immutable objects are never in the same group. + */ + + +#include +#include + +static void freeobj(upb_refcounted *o); + +const char untracked_val; +const void *UPB_UNTRACKED_REF = &untracked_val; + +/* arch-specific atomic primitives *******************************************/ + +#ifdef UPB_THREAD_UNSAFE ////////////////////////////////////////////////////// + +static void atomic_inc(uint32_t *a) { (*a)++; } +static bool atomic_dec(uint32_t *a) { return --(*a) == 0; } + +#elif defined(__GNUC__) || defined(__clang__) ////////////////////////////////// + +static void atomic_inc(uint32_t *a) { __sync_fetch_and_add(a, 1); } +static bool atomic_dec(uint32_t *a) { return __sync_sub_and_fetch(a, 1) == 0; } + +#elif defined(WIN32) /////////////////////////////////////////////////////////// + +#include + +static void atomic_inc(upb_atomic_t *a) { InterlockedIncrement(&a->val); } +static bool atomic_dec(upb_atomic_t *a) { + return InterlockedDecrement(&a->val) == 0; +} + +#else +#error Atomic primitives not defined for your platform/CPU. \ + Implement them or compile with UPB_THREAD_UNSAFE. +#endif + +// All static objects point to this refcount. +// It is special-cased in ref/unref below. +uint32_t static_refcount = -1; + +// We can avoid atomic ops for statically-declared objects. +// This is a minor optimization but nice since we can avoid degrading under +// contention in this case. + +static void refgroup(uint32_t *group) { + if (group != &static_refcount) + atomic_inc(group); +} + +static bool unrefgroup(uint32_t *group) { + if (group == &static_refcount) { + return false; + } else { + return atomic_dec(group); + } +} + + +/* Reference tracking (debug only) ********************************************/ + +#ifdef UPB_DEBUG_REFS + +#ifdef UPB_THREAD_UNSAFE + +static void upb_lock() {} +static void upb_unlock() {} + +#else + +// User must define functions that lock/unlock a global mutex and link this +// file against them. +void upb_lock(); +void upb_unlock(); + +#endif + +// UPB_DEBUG_REFS mode counts on being able to malloc() memory in some +// code-paths that can normally never fail, like upb_refcounted_ref(). Since +// we have no way to propagage out-of-memory errors back to the user, and since +// these errors can only occur in UPB_DEBUG_REFS mode, we immediately fail. +#define CHECK_OOM(predicate) if (!(predicate)) { assert(predicate); exit(1); } + +typedef struct { + int count; // How many refs there are (duplicates only allowed for ref2). + bool is_ref2; +} trackedref; + +static trackedref *trackedref_new(bool is_ref2) { + trackedref *ret = malloc(sizeof(*ret)); + CHECK_OOM(ret); + ret->count = 1; + ret->is_ref2 = is_ref2; + return ret; +} + +static void track(const upb_refcounted *r, const void *owner, bool ref2) { + assert(owner); + if (owner == UPB_UNTRACKED_REF) return; + + upb_lock(); + upb_value v; + if (upb_inttable_lookupptr(r->refs, owner, &v)) { + trackedref *ref = upb_value_getptr(v); + // Since we allow multiple ref2's for the same to/from pair without + // allocating separate memory for each one, we lose the fine-grained + // tracking behavior we get with regular refs. Since ref2s only happen + // inside upb, we'll accept this limitation until/unless there is a really + // difficult upb-internal bug that can't be figured out without it. + assert(ref2); + assert(ref->is_ref2); + ref->count++; + } else { + trackedref *ref = trackedref_new(ref2); + bool ok = upb_inttable_insertptr(r->refs, owner, upb_value_ptr(ref)); + CHECK_OOM(ok); + if (ref2) { + // We know this cast is safe when it is a ref2, because it's coming from + // another refcounted object. + const upb_refcounted *from = owner; + assert(!upb_inttable_lookupptr(from->ref2s, r, NULL)); + ok = upb_inttable_insertptr(from->ref2s, r, upb_value_ptr(NULL)); + CHECK_OOM(ok); + } + } + upb_unlock(); +} + +static void untrack(const upb_refcounted *r, const void *owner, bool ref2) { + assert(owner); + if (owner == UPB_UNTRACKED_REF) return; + + upb_lock(); + upb_value v; + bool found = upb_inttable_lookupptr(r->refs, owner, &v); + // This assert will fail if an owner attempts to release a ref it didn't have. + UPB_ASSERT_VAR(found, found); + trackedref *ref = upb_value_getptr(v); + assert(ref->is_ref2 == ref2); + if (--ref->count == 0) { + free(ref); + upb_inttable_removeptr(r->refs, owner, NULL); + if (ref2) { + // We know this cast is safe when it is a ref2, because it's coming from + // another refcounted object. + const upb_refcounted *from = owner; + bool removed = upb_inttable_removeptr(from->ref2s, r, NULL); + assert(removed); + } + } + upb_unlock(); +} + +static void checkref(const upb_refcounted *r, const void *owner, bool ref2) { + upb_lock(); + upb_value v; + bool found = upb_inttable_lookupptr(r->refs, owner, &v); + UPB_ASSERT_VAR(found, found); + trackedref *ref = upb_value_getptr(v); + assert(ref->is_ref2 == ref2); + upb_unlock(); +} + +// Populates the given UPB_CTYPE_INT32 inttable with counts of ref2's that +// originate from the given owner. +static void getref2s(const upb_refcounted *owner, upb_inttable *tab) { + upb_lock(); + upb_inttable_iter i; + upb_inttable_begin(&i, owner->ref2s); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_refcounted *to = (upb_refcounted*)upb_inttable_iter_key(&i); + + // To get the count we need to look in the target's table. + upb_value v; + bool found = upb_inttable_lookupptr(to->refs, owner, &v); + assert(found); + trackedref *ref = upb_value_getptr(v); + upb_value count = upb_value_int32(ref->count); + + bool ok = upb_inttable_insertptr(tab, to, count); + CHECK_OOM(ok); + } + upb_unlock(); +} + +typedef struct { + upb_inttable ref2; + const upb_refcounted *obj; +} check_state; + +static void visit_check(const upb_refcounted *obj, const upb_refcounted *subobj, + void *closure) { + check_state *s = closure; + assert(obj == s->obj); + assert(subobj); + upb_inttable *ref2 = &s->ref2; + upb_value v; + bool removed = upb_inttable_removeptr(ref2, subobj, &v); + // The following assertion will fail if the visit() function visits a subobj + // that it did not have a ref2 on, or visits the same subobj too many times. + assert(removed); + int32_t newcount = upb_value_getint32(v) - 1; + if (newcount > 0) { + upb_inttable_insert(ref2, (uintptr_t)subobj, upb_value_int32(newcount)); + } +} + +static void visit(const upb_refcounted *r, upb_refcounted_visit *v, + void *closure) { + // In DEBUG_REFS mode we know what existing ref2 refs there are, so we know + // exactly the set of nodes that visit() should visit. So we verify visit()'s + // correctness here. + check_state state; + state.obj = r; + bool ok = upb_inttable_init(&state.ref2, UPB_CTYPE_INT32); + CHECK_OOM(ok); + getref2s(r, &state.ref2); + + // This should visit any children in the ref2 table. + if (r->vtbl->visit) r->vtbl->visit(r, visit_check, &state); + + // This assertion will fail if the visit() function missed any children. + assert(upb_inttable_count(&state.ref2) == 0); + upb_inttable_uninit(&state.ref2); + if (r->vtbl->visit) r->vtbl->visit(r, v, closure); +} + +static bool trackinit(upb_refcounted *r) { + r->refs = malloc(sizeof(*r->refs)); + r->ref2s = malloc(sizeof(*r->ref2s)); + if (!r->refs || !r->ref2s) goto err1; + + if (!upb_inttable_init(r->refs, UPB_CTYPE_PTR)) goto err1; + if (!upb_inttable_init(r->ref2s, UPB_CTYPE_PTR)) goto err2; + return true; + +err2: + upb_inttable_uninit(r->refs); +err1: + free(r->refs); + free(r->ref2s); + return false; +} + +static void trackfree(const upb_refcounted *r) { + upb_inttable_uninit(r->refs); + upb_inttable_uninit(r->ref2s); + free(r->refs); + free(r->ref2s); +} + +#else + +static void track(const upb_refcounted *r, const void *owner, bool ref2) { + UPB_UNUSED(r); + UPB_UNUSED(owner); + UPB_UNUSED(ref2); +} + +static void untrack(const upb_refcounted *r, const void *owner, bool ref2) { + UPB_UNUSED(r); + UPB_UNUSED(owner); + UPB_UNUSED(ref2); +} + +static void checkref(const upb_refcounted *r, const void *owner, bool ref2) { + UPB_UNUSED(r); + UPB_UNUSED(owner); + UPB_UNUSED(ref2); +} + +static bool trackinit(upb_refcounted *r) { + UPB_UNUSED(r); + return true; +} + +static void trackfree(const upb_refcounted *r) { + UPB_UNUSED(r); +} + +static void visit(const upb_refcounted *r, upb_refcounted_visit *v, + void *closure) { + if (r->vtbl->visit) r->vtbl->visit(r, v, closure); +} + +#endif // UPB_DEBUG_REFS + + +/* freeze() *******************************************************************/ + +// The freeze() operation is by far the most complicated part of this scheme. +// We compute strongly-connected components and then mutate the graph such that +// we preserve the invariants documented at the top of this file. And we must +// handle out-of-memory errors gracefully (without leaving the graph +// inconsistent), which adds to the fun. + +// The state used by the freeze operation (shared across many functions). +typedef struct { + int depth; + int maxdepth; + uint64_t index; + // Maps upb_refcounted* -> attributes (color, etc). attr layout varies by + // color. + upb_inttable objattr; + upb_inttable stack; // stack of upb_refcounted* for Tarjan's algorithm. + upb_inttable groups; // array of uint32_t*, malloc'd refcounts for new groups + upb_status *status; + jmp_buf err; +} tarjan; + +static void release_ref2(const upb_refcounted *obj, + const upb_refcounted *subobj, + void *closure); + +// Node attributes ///////////////////////////////////////////////////////////// + +// After our analysis phase all nodes will be either GRAY or WHITE. + +typedef enum { + BLACK = 0, // Object has not been seen. + GRAY, // Object has been found via a refgroup but may not be reachable. + GREEN, // Object is reachable and is currently on the Tarjan stack. + WHITE, // Object is reachable and has been assigned a group (SCC). +} color_t; + +UPB_NORETURN static void err(tarjan *t) { longjmp(t->err, 1); } +UPB_NORETURN static void oom(tarjan *t) { + upb_status_seterrmsg(t->status, "out of memory"); + err(t); +} + +static uint64_t trygetattr(const tarjan *t, const upb_refcounted *r) { + upb_value v; + return upb_inttable_lookupptr(&t->objattr, r, &v) ? + upb_value_getuint64(v) : 0; +} + +static uint64_t getattr(const tarjan *t, const upb_refcounted *r) { + upb_value v; + bool found = upb_inttable_lookupptr(&t->objattr, r, &v); + UPB_ASSERT_VAR(found, found); + return upb_value_getuint64(v); +} + +static void setattr(tarjan *t, const upb_refcounted *r, uint64_t attr) { + upb_inttable_removeptr(&t->objattr, r, NULL); + upb_inttable_insertptr(&t->objattr, r, upb_value_uint64(attr)); +} + +static color_t color(tarjan *t, const upb_refcounted *r) { + return trygetattr(t, r) & 0x3; // Color is always stored in the low 2 bits. +} + +static void set_gray(tarjan *t, const upb_refcounted *r) { + assert(color(t, r) == BLACK); + setattr(t, r, GRAY); +} + +// Pushes an obj onto the Tarjan stack and sets it to GREEN. +static void push(tarjan *t, const upb_refcounted *r) { + assert(color(t, r) == BLACK || color(t, r) == GRAY); + // This defines the attr layout for the GREEN state. "index" and "lowlink" + // get 31 bits, which is plenty (limit of 2B objects frozen at a time). + setattr(t, r, GREEN | (t->index << 2) | (t->index << 33)); + if (++t->index == 0x80000000) { + upb_status_seterrmsg(t->status, "too many objects to freeze"); + err(t); + } + upb_inttable_push(&t->stack, upb_value_ptr((void*)r)); +} + +// Pops an obj from the Tarjan stack and sets it to WHITE, with a ptr to its +// SCC group. +static upb_refcounted *pop(tarjan *t) { + upb_refcounted *r = upb_value_getptr(upb_inttable_pop(&t->stack)); + assert(color(t, r) == GREEN); + // This defines the attr layout for nodes in the WHITE state. + // Top of group stack is [group, NULL]; we point at group. + setattr(t, r, WHITE | (upb_inttable_count(&t->groups) - 2) << 8); + return r; +} + +static void tarjan_newgroup(tarjan *t) { + uint32_t *group = malloc(sizeof(*group)); + if (!group) oom(t); + // Push group and empty group leader (we'll fill in leader later). + if (!upb_inttable_push(&t->groups, upb_value_ptr(group)) || + !upb_inttable_push(&t->groups, upb_value_ptr(NULL))) { + free(group); + oom(t); + } + *group = 0; +} + +static uint32_t idx(tarjan *t, const upb_refcounted *r) { + assert(color(t, r) == GREEN); + return (getattr(t, r) >> 2) & 0x7FFFFFFF; +} + +static uint32_t lowlink(tarjan *t, const upb_refcounted *r) { + if (color(t, r) == GREEN) { + return getattr(t, r) >> 33; + } else { + return UINT32_MAX; + } +} + +static void set_lowlink(tarjan *t, const upb_refcounted *r, uint32_t lowlink) { + assert(color(t, r) == GREEN); + setattr(t, r, ((uint64_t)lowlink << 33) | (getattr(t, r) & 0x1FFFFFFFF)); +} + +static uint32_t *group(tarjan *t, upb_refcounted *r) { + assert(color(t, r) == WHITE); + uint64_t groupnum = getattr(t, r) >> 8; + upb_value v; + bool found = upb_inttable_lookup(&t->groups, groupnum, &v); + UPB_ASSERT_VAR(found, found); + return upb_value_getptr(v); +} + +// If the group leader for this object's group has not previously been set, +// the given object is assigned to be its leader. +static upb_refcounted *groupleader(tarjan *t, upb_refcounted *r) { + assert(color(t, r) == WHITE); + uint64_t leader_slot = (getattr(t, r) >> 8) + 1; + upb_value v; + bool found = upb_inttable_lookup(&t->groups, leader_slot, &v); + UPB_ASSERT_VAR(found, found); + if (upb_value_getptr(v)) { + return upb_value_getptr(v); + } else { + upb_inttable_remove(&t->groups, leader_slot, NULL); + upb_inttable_insert(&t->groups, leader_slot, upb_value_ptr(r)); + return r; + } +} + + +// Tarjan's algorithm ////////////////////////////////////////////////////////// + +// See: +// http://en.wikipedia.org/wiki/Tarjan%27s_strongly_connected_components_algorithm +static void do_tarjan(const upb_refcounted *obj, tarjan *t); + +static void tarjan_visit(const upb_refcounted *obj, + const upb_refcounted *subobj, + void *closure) { + tarjan *t = closure; + if (++t->depth > t->maxdepth) { + upb_status_seterrf(t->status, "graph too deep to freeze (%d)", t->maxdepth); + err(t); + } else if (subobj->is_frozen || color(t, subobj) == WHITE) { + // Do nothing: we don't want to visit or color already-frozen nodes, + // and WHITE nodes have already been assigned a SCC. + } else if (color(t, subobj) < GREEN) { + // Subdef has not yet been visited; recurse on it. + do_tarjan(subobj, t); + set_lowlink(t, obj, UPB_MIN(lowlink(t, obj), lowlink(t, subobj))); + } else if (color(t, subobj) == GREEN) { + // Subdef is in the stack and hence in the current SCC. + set_lowlink(t, obj, UPB_MIN(lowlink(t, obj), idx(t, subobj))); + } + --t->depth; +} + +static void do_tarjan(const upb_refcounted *obj, tarjan *t) { + if (color(t, obj) == BLACK) { + // We haven't seen this object's group; mark the whole group GRAY. + const upb_refcounted *o = obj; + do { set_gray(t, o); } while ((o = o->next) != obj); + } + + push(t, obj); + visit(obj, tarjan_visit, t); + if (lowlink(t, obj) == idx(t, obj)) { + tarjan_newgroup(t); + while (pop(t) != obj) + ; + } +} + + +// freeze() //////////////////////////////////////////////////////////////////// + +static void crossref(const upb_refcounted *r, const upb_refcounted *subobj, + void *_t) { + tarjan *t = _t; + assert(color(t, r) > BLACK); + if (color(t, subobj) > BLACK && r->group != subobj->group) { + // Previously this ref was not reflected in subobj->group because they + // were in the same group; now that they are split a ref must be taken. + refgroup(subobj->group); + } +} + +static bool freeze(upb_refcounted *const*roots, int n, upb_status *s, + int maxdepth) { + volatile bool ret = false; + + // We run in two passes so that we can allocate all memory before performing + // any mutation of the input -- this allows us to leave the input unchanged + // in the case of memory allocation failure. + tarjan t; + t.index = 0; + t.depth = 0; + t.maxdepth = maxdepth; + t.status = s; + if (!upb_inttable_init(&t.objattr, UPB_CTYPE_UINT64)) goto err1; + if (!upb_inttable_init(&t.stack, UPB_CTYPE_PTR)) goto err2; + if (!upb_inttable_init(&t.groups, UPB_CTYPE_PTR)) goto err3; + if (setjmp(t.err) != 0) goto err4; + + + for (int i = 0; i < n; i++) { + if (color(&t, roots[i]) < GREEN) { + do_tarjan(roots[i], &t); + } + } + + // If we've made it this far, no further errors are possible so it's safe to + // mutate the objects without risk of leaving them in an inconsistent state. + ret = true; + + // The transformation that follows requires care. The preconditions are: + // - all objects in attr map are WHITE or GRAY, and are in mutable groups + // (groups of all mutable objs) + // - no ref2(to, from) refs have incremented count(to) if both "to" and + // "from" are in our attr map (this follows from invariants (2) and (3)) + + // Pass 1: we remove WHITE objects from their mutable groups, and add them to + // new groups according to the SCC's we computed. These new groups will + // consist of only frozen objects. None will be immediately collectible, + // because WHITE objects are by definition reachable from one of "roots", + // which the caller must own refs on. + upb_inttable_iter i; + upb_inttable_begin(&i, &t.objattr); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&i); + // Since removal from a singly-linked list requires access to the object's + // predecessor, we consider obj->next instead of obj for moving. With the + // while() loop we guarantee that we will visit every node's predecessor. + // Proof: + // 1. every node's predecessor is in our attr map. + // 2. though the loop body may change a node's predecessor, it will only + // change it to be the node we are currently operating on, so with a + // while() loop we guarantee ourselves the chance to remove each node. + while (color(&t, obj->next) == WHITE && + group(&t, obj->next) != obj->next->group) { + // Remove from old group. + upb_refcounted *move = obj->next; + if (obj == move) { + // Removing the last object from a group. + assert(*obj->group == obj->individual_count); + free(obj->group); + } else { + obj->next = move->next; + // This may decrease to zero; we'll collect GRAY objects (if any) that + // remain in the group in the third pass. + assert(*move->group >= move->individual_count); + *move->group -= move->individual_count; + } + + // Add to new group. + upb_refcounted *leader = groupleader(&t, move); + if (move == leader) { + // First object added to new group is its leader. + move->group = group(&t, move); + move->next = move; + *move->group = move->individual_count; + } else { + // Group already has at least one object in it. + assert(leader->group == group(&t, move)); + move->group = group(&t, move); + move->next = leader->next; + leader->next = move; + *move->group += move->individual_count; + } + + move->is_frozen = true; + } + } + + // Pass 2: GRAY and WHITE objects "obj" with ref2(to, obj) references must + // increment count(to) if group(obj) != group(to) (which could now be the + // case if "to" was just frozen). + upb_inttable_begin(&i, &t.objattr); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&i); + visit(obj, crossref, &t); + } + + // Pass 3: GRAY objects are collected if their group's refcount dropped to + // zero when we removed its white nodes. This can happen if they had only + // been kept alive by virtue of sharing a group with an object that was just + // frozen. + // + // It is important that we do this last, since the GRAY object's free() + // function could call unref2() on just-frozen objects, which will decrement + // refs that were added in pass 2. + upb_inttable_begin(&i, &t.objattr); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_refcounted *obj = (upb_refcounted*)upb_inttable_iter_key(&i); + if (obj->group == NULL || *obj->group == 0) { + if (obj->group) { + // We eagerly free() the group's count (since we can't easily determine + // the group's remaining size it's the easiest way to ensure it gets + // done). + free(obj->group); + + // Visit to release ref2's (done in a separate pass since release_ref2 + // depends on o->group being unmodified so it can test merged()). + upb_refcounted *o = obj; + do { visit(o, release_ref2, NULL); } while ((o = o->next) != obj); + + // Mark "group" fields as NULL so we know to free the objects later in + // this loop, but also don't try to delete the group twice. + o = obj; + do { o->group = NULL; } while ((o = o->next) != obj); + } + freeobj(obj); + } + } + +err4: + if (!ret) { + upb_inttable_begin(&i, &t.groups); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) + free(upb_value_getptr(upb_inttable_iter_value(&i))); + } + upb_inttable_uninit(&t.groups); +err3: + upb_inttable_uninit(&t.stack); +err2: + upb_inttable_uninit(&t.objattr); +err1: + return ret; +} + + +/* Misc internal functions ***************************************************/ + +static bool merged(const upb_refcounted *r, const upb_refcounted *r2) { + return r->group == r2->group; +} + +static void merge(upb_refcounted *r, upb_refcounted *from) { + if (merged(r, from)) return; + *r->group += *from->group; + free(from->group); + upb_refcounted *base = from; + + // Set all refcount pointers in the "from" chain to the merged refcount. + // + // TODO(haberman): this linear algorithm can result in an overall O(n^2) bound + // if the user continuously extends a group by one object. Prevent this by + // using one of the techniques in this paper: + // ftp://www.ncedc.org/outgoing/geomorph/dino/orals/p245-tarjan.pdf + do { from->group = r->group; } while ((from = from->next) != base); + + // Merge the two circularly linked lists by swapping their next pointers. + upb_refcounted *tmp = r->next; + r->next = base->next; + base->next = tmp; +} + +static void unref(const upb_refcounted *r); + +static void release_ref2(const upb_refcounted *obj, + const upb_refcounted *subobj, + void *closure) { + UPB_UNUSED(closure); + untrack(subobj, obj, true); + if (!merged(obj, subobj)) { + assert(subobj->is_frozen); + unref(subobj); + } +} + +static void unref(const upb_refcounted *r) { + if (unrefgroup(r->group)) { + free(r->group); + + // In two passes, since release_ref2 needs a guarantee that any subobjs + // are alive. + const upb_refcounted *o = r; + do { visit(o, release_ref2, NULL); } while((o = o->next) != r); + + o = r; + do { + const upb_refcounted *next = o->next; + assert(o->is_frozen || o->individual_count == 0); + freeobj((upb_refcounted*)o); + o = next; + } while(o != r); + } +} + +static void freeobj(upb_refcounted *o) { + trackfree(o); + o->vtbl->free((upb_refcounted*)o); +} + + +/* Public interface ***********************************************************/ + +bool upb_refcounted_init(upb_refcounted *r, + const struct upb_refcounted_vtbl *vtbl, + const void *owner) { + r->next = r; + r->vtbl = vtbl; + r->individual_count = 0; + r->is_frozen = false; + r->group = malloc(sizeof(*r->group)); + if (!r->group) return false; + *r->group = 0; + if (!trackinit(r)) { + free(r->group); + return false; + } + upb_refcounted_ref(r, owner); + return true; +} + +bool upb_refcounted_isfrozen(const upb_refcounted *r) { + return r->is_frozen; +} + +void upb_refcounted_ref(const upb_refcounted *r, const void *owner) { + track(r, owner, false); + if (!r->is_frozen) + ((upb_refcounted*)r)->individual_count++; + refgroup(r->group); +} + +void upb_refcounted_unref(const upb_refcounted *r, const void *owner) { + untrack(r, owner, false); + if (!r->is_frozen) + ((upb_refcounted*)r)->individual_count--; + unref(r); +} + +void upb_refcounted_ref2(const upb_refcounted *r, upb_refcounted *from) { + assert(!from->is_frozen); // Non-const pointer implies this. + track(r, from, true); + if (r->is_frozen) { + refgroup(r->group); + } else { + merge((upb_refcounted*)r, from); + } +} + +void upb_refcounted_unref2(const upb_refcounted *r, upb_refcounted *from) { + assert(!from->is_frozen); // Non-const pointer implies this. + untrack(r, from, true); + if (r->is_frozen) { + unref(r); + } else { + assert(merged(r, from)); + } +} + +void upb_refcounted_donateref( + const upb_refcounted *r, const void *from, const void *to) { + assert(from != to); + if (to != NULL) + upb_refcounted_ref(r, to); + if (from != NULL) + upb_refcounted_unref(r, from); +} + +void upb_refcounted_checkref(const upb_refcounted *r, const void *owner) { + checkref(r, owner, false); +} + +bool upb_refcounted_freeze(upb_refcounted *const*roots, int n, upb_status *s, + int maxdepth) { + for (int i = 0; i < n; i++) { + assert(!roots[i]->is_frozen); + } + return freeze(roots, n, s, maxdepth); +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2013 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + + +#include + +// Fallback implementation if the shim is not specialized by the JIT. +#define SHIM_WRITER(type, ctype) \ + bool upb_shim_set ## type (void *c, const void *hd, ctype val) { \ + uint8_t *m = c; \ + const upb_shim_data *d = hd; \ + if (d->hasbit > 0) \ + *(uint8_t*)&m[d->hasbit / 8] |= 1 << (d->hasbit % 8); \ + *(ctype*)&m[d->offset] = val; \ + return true; \ + } \ + +SHIM_WRITER(double, double) +SHIM_WRITER(float, float) +SHIM_WRITER(int32, int32_t) +SHIM_WRITER(int64, int64_t) +SHIM_WRITER(uint32, uint32_t) +SHIM_WRITER(uint64, uint64_t) +SHIM_WRITER(bool, bool) +#undef SHIM_WRITER + +bool upb_shim_set(upb_handlers *h, const upb_fielddef *f, size_t offset, + int32_t hasbit) { + upb_shim_data *d = malloc(sizeof(*d)); + if (!d) return false; + d->offset = offset; + d->hasbit = hasbit; + + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, d); + upb_handlerattr_setalwaysok(&attr, true); + upb_handlers_addcleanup(h, d, free); + +#define TYPE(u, l) \ + case UPB_TYPE_##u: \ + ok = upb_handlers_set##l(h, f, upb_shim_set##l, &attr); break; + + bool ok = false; + + switch (upb_fielddef_type(f)) { + TYPE(INT64, int64); + TYPE(INT32, int32); + TYPE(ENUM, int32); + TYPE(UINT64, uint64); + TYPE(UINT32, uint32); + TYPE(DOUBLE, double); + TYPE(FLOAT, float); + TYPE(BOOL, bool); + default: assert(false); break; + } +#undef TYPE + + upb_handlerattr_uninit(&attr); + return ok; +} + +const upb_shim_data *upb_shim_getdata(const upb_handlers *h, upb_selector_t s, + upb_fieldtype_t *type) { + upb_func *f = upb_handlers_gethandler(h, s); + + if ((upb_int64_handlerfunc*)f == upb_shim_setint64) { + *type = UPB_TYPE_INT64; + } else if ((upb_int32_handlerfunc*)f == upb_shim_setint32) { + *type = UPB_TYPE_INT32; + } else if ((upb_uint64_handlerfunc*)f == upb_shim_setuint64) { + *type = UPB_TYPE_UINT64; + } else if ((upb_uint32_handlerfunc*)f == upb_shim_setuint32) { + *type = UPB_TYPE_UINT32; + } else if ((upb_double_handlerfunc*)f == upb_shim_setdouble) { + *type = UPB_TYPE_DOUBLE; + } else if ((upb_float_handlerfunc*)f == upb_shim_setfloat) { + *type = UPB_TYPE_FLOAT; + } else if ((upb_bool_handlerfunc*)f == upb_shim_setbool) { + *type = UPB_TYPE_BOOL; + } else { + return NULL; + } + + return (const upb_shim_data*)upb_handlers_gethandlerdata(h, s); +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2008-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + + +#include +#include + +bool upb_symtab_isfrozen(const upb_symtab *s) { + return upb_refcounted_isfrozen(UPB_UPCAST(s)); +} + +void upb_symtab_ref(const upb_symtab *s, const void *owner) { + upb_refcounted_ref(UPB_UPCAST(s), owner); +} + +void upb_symtab_unref(const upb_symtab *s, const void *owner) { + upb_refcounted_unref(UPB_UPCAST(s), owner); +} + +void upb_symtab_donateref( + const upb_symtab *s, const void *from, const void *to) { + upb_refcounted_donateref(UPB_UPCAST(s), from, to); +} + +void upb_symtab_checkref(const upb_symtab *s, const void *owner) { + upb_refcounted_checkref(UPB_UPCAST(s), owner); +} + +static void upb_symtab_free(upb_refcounted *r) { + upb_symtab *s = (upb_symtab*)r; + upb_strtable_iter i; + upb_strtable_begin(&i, &s->symtab); + for (; !upb_strtable_done(&i); upb_strtable_next(&i)) { + const upb_def *def = upb_value_getptr(upb_strtable_iter_value(&i)); + upb_def_unref(def, s); + } + upb_strtable_uninit(&s->symtab); + free(s); +} + + +upb_symtab *upb_symtab_new(const void *owner) { + static const struct upb_refcounted_vtbl vtbl = {NULL, &upb_symtab_free}; + upb_symtab *s = malloc(sizeof(*s)); + upb_refcounted_init(UPB_UPCAST(s), &vtbl, owner); + upb_strtable_init(&s->symtab, UPB_CTYPE_PTR); + return s; +} + +void upb_symtab_freeze(upb_symtab *s) { + assert(!upb_symtab_isfrozen(s)); + upb_refcounted *r = UPB_UPCAST(s); + // The symtab does not take ref2's (see refcounted.h) on the defs, because + // defs cannot refer back to the table and therefore cannot create cycles. So + // 0 will suffice for maxdepth here. + bool ok = upb_refcounted_freeze(&r, 1, NULL, 0); + UPB_ASSERT_VAR(ok, ok); +} + +const upb_def *upb_symtab_lookup(const upb_symtab *s, const char *sym) { + upb_value v; + upb_def *ret = upb_strtable_lookup(&s->symtab, sym, &v) ? + upb_value_getptr(v) : NULL; + return ret; +} + +const upb_msgdef *upb_symtab_lookupmsg(const upb_symtab *s, const char *sym) { + upb_value v; + upb_def *def = upb_strtable_lookup(&s->symtab, sym, &v) ? + upb_value_getptr(v) : NULL; + return def ? upb_dyncast_msgdef(def) : NULL; +} + +const upb_enumdef *upb_symtab_lookupenum(const upb_symtab *s, const char *sym) { + upb_value v; + upb_def *def = upb_strtable_lookup(&s->symtab, sym, &v) ? + upb_value_getptr(v) : NULL; + return def ? upb_dyncast_enumdef(def) : NULL; +} + +// Given a symbol and the base symbol inside which it is defined, find the +// symbol's definition in t. +static upb_def *upb_resolvename(const upb_strtable *t, + const char *base, const char *sym) { + if(strlen(sym) == 0) return NULL; + if(sym[0] == '.') { + // Symbols starting with '.' are absolute, so we do a single lookup. + // Slice to omit the leading '.' + upb_value v; + return upb_strtable_lookup(t, sym + 1, &v) ? upb_value_getptr(v) : NULL; + } else { + // Remove components from base until we find an entry or run out. + // TODO: This branch is totally broken, but currently not used. + (void)base; + assert(false); + return NULL; + } +} + +const upb_def *upb_symtab_resolve(const upb_symtab *s, const char *base, + const char *sym) { + upb_def *ret = upb_resolvename(&s->symtab, base, sym); + return ret; +} + +// Searches def and its children to find defs that have the same name as any +// def in "addtab." Returns true if any where found, and as a side-effect adds +// duplicates of these defs into addtab. +// +// We use a modified depth-first traversal that traverses each SCC (which we +// already computed) as if it were a single node. This allows us to traverse +// the possibly-cyclic graph as if it were a DAG and to dup the correct set of +// nodes with O(n) time. +static bool upb_resolve_dfs(const upb_def *def, upb_strtable *addtab, + const void *new_owner, upb_inttable *seen, + upb_status *s) { + // Memoize results of this function for efficiency (since we're traversing a + // DAG this is not needed to limit the depth of the search). + upb_value v; + if (upb_inttable_lookup(seen, (uintptr_t)def, &v)) + return upb_value_getbool(v); + + // Visit submessages for all messages in the SCC. + bool need_dup = false; + const upb_def *base = def; + do { + assert(upb_def_isfrozen(def)); + if (def->type == UPB_DEF_FIELD) continue; + upb_value v; + if (upb_strtable_lookup(addtab, upb_def_fullname(def), &v)) { + need_dup = true; + } + + // For messages, continue the recursion by visiting all subdefs. + const upb_msgdef *m = upb_dyncast_msgdef(def); + if (m) { + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + if (!upb_fielddef_hassubdef(f)) continue; + // |= to avoid short-circuit; we need its side-effects. + need_dup |= upb_resolve_dfs( + upb_fielddef_subdef(f), addtab, new_owner, seen, s); + if (!upb_ok(s)) return false; + } + } + } while ((def = (upb_def*)def->base.next) != base); + + if (need_dup) { + // Dup any defs that don't already have entries in addtab. + def = base; + do { + if (def->type == UPB_DEF_FIELD) continue; + const char *name = upb_def_fullname(def); + if (!upb_strtable_lookup(addtab, name, NULL)) { + upb_def *newdef = upb_def_dup(def, new_owner); + if (!newdef) goto oom; + newdef->came_from_user = false; + if (!upb_strtable_insert(addtab, name, upb_value_ptr(newdef))) + goto oom; + } + } while ((def = (upb_def*)def->base.next) != base); + } + + upb_inttable_insert(seen, (uintptr_t)def, upb_value_bool(need_dup)); + return need_dup; + +oom: + upb_status_seterrmsg(s, "out of memory"); + return false; +} + +// TODO(haberman): we need a lot more testing of error conditions. +// The came_from_user stuff in particular is not tested. +bool upb_symtab_add(upb_symtab *s, upb_def *const*defs, int n, void *ref_donor, + upb_status *status) { + assert(!upb_symtab_isfrozen(s)); + upb_def **add_defs = NULL; + upb_strtable addtab; + if (!upb_strtable_init(&addtab, UPB_CTYPE_PTR)) { + upb_status_seterrmsg(status, "out of memory"); + return false; + } + + // Add new defs to our "add" set. + for (int i = 0; i < n; i++) { + upb_def *def = defs[i]; + if (upb_def_isfrozen(def)) { + upb_status_seterrmsg(status, "added defs must be mutable"); + goto err; + } + assert(!upb_def_isfrozen(def)); + const char *fullname = upb_def_fullname(def); + if (!fullname) { + upb_status_seterrmsg( + status, "Anonymous defs cannot be added to a symtab"); + goto err; + } + + upb_fielddef *f = upb_dyncast_fielddef_mutable(def); + + if (f) { + if (!upb_fielddef_containingtypename(f)) { + upb_status_seterrmsg(status, + "Standalone fielddefs must have a containing type " + "(extendee) name set"); + goto err; + } + } else { + if (upb_strtable_lookup(&addtab, fullname, NULL)) { + upb_status_seterrf(status, "Conflicting defs named '%s'", fullname); + goto err; + } + // We need this to back out properly, because if there is a failure we + // need to donate the ref back to the caller. + def->came_from_user = true; + upb_def_donateref(def, ref_donor, s); + if (!upb_strtable_insert(&addtab, fullname, upb_value_ptr(def))) + goto oom_err; + } + } + + // Add standalone fielddefs (ie. extensions) to the appropriate messages. + // If the appropriate message only exists in the existing symtab, duplicate + // it so we have a mutable copy we can add the fields to. + for (int i = 0; i < n; i++) { + upb_def *def = defs[i]; + upb_fielddef *f = upb_dyncast_fielddef_mutable(def); + if (!f) continue; + const char *msgname = upb_fielddef_containingtypename(f); + // We validated this earlier in this function. + assert(msgname); + + // If the extendee name is absolutely qualified, move past the initial ".". + // TODO(haberman): it is not obvious what it would mean if this was not + // absolutely qualified. + if (msgname[0] == '.') { + msgname++; + } + + upb_value v; + upb_msgdef *m; + if (upb_strtable_lookup(&addtab, msgname, &v)) { + // Extendee is in the set of defs the user asked us to add. + m = upb_value_getptr(v); + } else { + // Need to find and dup the extendee from the existing symtab. + const upb_msgdef *frozen_m = upb_symtab_lookupmsg(s, msgname); + if (!frozen_m) { + upb_status_seterrf(status, + "Tried to extend message %s that does not exist " + "in this SymbolTable.", + msgname); + goto err; + } + m = upb_msgdef_dup(frozen_m, s); + if (!m) goto oom_err; + if (!upb_strtable_insert(&addtab, msgname, upb_value_ptr(m))) { + upb_msgdef_unref(m, s); + goto oom_err; + } + } + + if (!upb_msgdef_addfield(m, f, ref_donor, status)) { + goto err; + } + } + + // Add dups of any existing def that can reach a def with the same name as + // anything in our "add" set. + upb_inttable seen; + if (!upb_inttable_init(&seen, UPB_CTYPE_BOOL)) goto oom_err; + upb_strtable_iter i; + upb_strtable_begin(&i, &s->symtab); + for (; !upb_strtable_done(&i); upb_strtable_next(&i)) { + upb_def *def = upb_value_getptr(upb_strtable_iter_value(&i)); + upb_resolve_dfs(def, &addtab, s, &seen, status); + if (!upb_ok(status)) goto err; + } + upb_inttable_uninit(&seen); + + // Now using the table, resolve symbolic references for subdefs. + upb_strtable_begin(&i, &addtab); + for (; !upb_strtable_done(&i); upb_strtable_next(&i)) { + upb_def *def = upb_value_getptr(upb_strtable_iter_value(&i)); + upb_msgdef *m = upb_dyncast_msgdef_mutable(def); + if (!m) continue; + // Type names are resolved relative to the message in which they appear. + const char *base = upb_msgdef_fullname(m); + + upb_msg_field_iter j; + for(upb_msg_field_begin(&j, m); + !upb_msg_field_done(&j); + upb_msg_field_next(&j)) { + upb_fielddef *f = upb_msg_iter_field(&j); + const char *name = upb_fielddef_subdefname(f); + if (name && !upb_fielddef_subdef(f)) { + // Try the lookup in the current set of to-be-added defs first. If not + // there, try existing defs. + upb_def *subdef = upb_resolvename(&addtab, base, name); + if (subdef == NULL) { + subdef = upb_resolvename(&s->symtab, base, name); + } + if (subdef == NULL) { + upb_status_seterrf( + status, "couldn't resolve name '%s' in message '%s'", name, base); + goto err; + } else if (!upb_fielddef_setsubdef(f, subdef, status)) { + goto err; + } + } + } + } + + // We need an array of the defs in addtab, for passing to upb_def_freeze. + add_defs = malloc(sizeof(void*) * upb_strtable_count(&addtab)); + if (add_defs == NULL) goto oom_err; + upb_strtable_begin(&i, &addtab); + for (n = 0; !upb_strtable_done(&i); upb_strtable_next(&i)) { + add_defs[n++] = upb_value_getptr(upb_strtable_iter_value(&i)); + } + + if (!upb_def_freeze(add_defs, n, status)) goto err; + + // This must be delayed until all errors have been detected, since error + // recovery code uses this table to cleanup defs. + upb_strtable_uninit(&addtab); + + // TODO(haberman) we don't properly handle errors after this point (like + // OOM in upb_strtable_insert() below). + for (int i = 0; i < n; i++) { + upb_def *def = add_defs[i]; + const char *name = upb_def_fullname(def); + upb_value v; + if (upb_strtable_remove(&s->symtab, name, &v)) { + const upb_def *def = upb_value_getptr(v); + upb_def_unref(def, s); + } + bool success = upb_strtable_insert(&s->symtab, name, upb_value_ptr(def)); + UPB_ASSERT_VAR(success, success == true); + } + free(add_defs); + return true; + +oom_err: + upb_status_seterrmsg(status, "out of memory"); +err: { + // For defs the user passed in, we need to donate the refs back. For defs + // we dup'd, we need to just unref them. + upb_strtable_iter i; + upb_strtable_begin(&i, &addtab); + for (; !upb_strtable_done(&i); upb_strtable_next(&i)) { + upb_def *def = upb_value_getptr(upb_strtable_iter_value(&i)); + bool came_from_user = def->came_from_user; + def->came_from_user = false; + if (came_from_user) { + upb_def_donateref(def, s, ref_donor); + } else { + upb_def_unref(def, s); + } + } + } + upb_strtable_uninit(&addtab); + free(add_defs); + assert(!upb_ok(status)); + return false; +} + +// Iteration. + +static void advance_to_matching(upb_symtab_iter *iter) { + if (iter->type == UPB_DEF_ANY) + return; + + while (!upb_strtable_done(&iter->iter) && + iter->type != upb_symtab_iter_def(iter)->type) { + upb_strtable_next(&iter->iter); + } +} + +void upb_symtab_begin(upb_symtab_iter *iter, const upb_symtab *s, + upb_deftype_t type) { + upb_strtable_begin(&iter->iter, &s->symtab); + iter->type = type; + advance_to_matching(iter); +} + +void upb_symtab_next(upb_symtab_iter *iter) { + upb_strtable_next(&iter->iter); + advance_to_matching(iter); +} + +bool upb_symtab_done(const upb_symtab_iter *iter) { + return upb_strtable_done(&iter->iter); +} + +const upb_def *upb_symtab_iter_def(const upb_symtab_iter *iter) { + return upb_value_getptr(upb_strtable_iter_value(&iter->iter)); +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Implementation is heavily inspired by Lua's ltable.c. + */ + + +#include +#include + +#define UPB_MAXARRSIZE 16 // 64k. + +// From Chromium. +#define ARRAY_SIZE(x) \ + ((sizeof(x)/sizeof(0[x])) / ((size_t)(!(sizeof(x) % sizeof(0[x]))))) + +static const double MAX_LOAD = 0.85; + +// The minimum utilization of the array part of a mixed hash/array table. This +// is a speed/memory-usage tradeoff (though it's not straightforward because of +// cache effects). The lower this is, the more memory we'll use. +static const double MIN_DENSITY = 0.1; + +bool is_pow2(uint64_t v) { return v == 0 || (v & (v - 1)) == 0; } + +int log2ceil(uint64_t v) { + int ret = 0; + bool pow2 = is_pow2(v); + while (v >>= 1) ret++; + ret = pow2 ? ret : ret + 1; // Ceiling. + return UPB_MIN(UPB_MAXARRSIZE, ret); +} + +char *upb_strdup(const char *s) { + return upb_strdup2(s, strlen(s)); +} + +char *upb_strdup2(const char *s, size_t len) { + // Prevent overflow errors. + if (len == SIZE_MAX) return NULL; + // Always null-terminate, even if binary data; but don't rely on the input to + // have a null-terminating byte since it may be a raw binary buffer. + size_t n = len + 1; + char *p = malloc(n); + if (p) { + memcpy(p, s, len); + p[len] = 0; + } + return p; +} + +// A type to represent the lookup key of either a strtable or an inttable. +typedef struct { + upb_tabkey key; +} lookupkey_t; + +static lookupkey_t strkey2(const char *str, size_t len) { + lookupkey_t k; + k.key.s.str = (char*)str; + k.key.s.length = len; + return k; +} + +static lookupkey_t intkey(uintptr_t key) { + lookupkey_t k; + k.key = upb_intkey(key); + return k; +} + +typedef uint32_t hashfunc_t(upb_tabkey key); +typedef bool eqlfunc_t(upb_tabkey k1, lookupkey_t k2); + +/* Base table (shared code) ***************************************************/ + +// For when we need to cast away const. +static upb_tabent *mutable_entries(upb_table *t) { + return (upb_tabent*)t->entries; +} + +static bool isfull(upb_table *t) { + return (double)(t->count + 1) / upb_table_size(t) > MAX_LOAD; +} + +static bool init(upb_table *t, upb_ctype_t ctype, uint8_t size_lg2) { + t->count = 0; + t->ctype = ctype; + t->size_lg2 = size_lg2; + t->mask = upb_table_size(t) ? upb_table_size(t) - 1 : 0; + size_t bytes = upb_table_size(t) * sizeof(upb_tabent); + if (bytes > 0) { + t->entries = malloc(bytes); + if (!t->entries) return false; + memset(mutable_entries(t), 0, bytes); + } else { + t->entries = NULL; + } + return true; +} + +static void uninit(upb_table *t) { free(mutable_entries(t)); } + +static upb_tabent *emptyent(upb_table *t) { + upb_tabent *e = mutable_entries(t) + upb_table_size(t); + while (1) { if (upb_tabent_isempty(--e)) return e; assert(e > t->entries); } +} + +static upb_tabent *getentry_mutable(upb_table *t, uint32_t hash) { + return (upb_tabent*)upb_getentry(t, hash); +} + +static const upb_tabent *findentry(const upb_table *t, lookupkey_t key, + uint32_t hash, eqlfunc_t *eql) { + if (t->size_lg2 == 0) return NULL; + const upb_tabent *e = upb_getentry(t, hash); + if (upb_tabent_isempty(e)) return NULL; + while (1) { + if (eql(e->key, key)) return e; + if ((e = e->next) == NULL) return NULL; + } +} + +static upb_tabent *findentry_mutable(upb_table *t, lookupkey_t key, + uint32_t hash, eqlfunc_t *eql) { + return (upb_tabent*)findentry(t, key, hash, eql); +} + +static bool lookup(const upb_table *t, lookupkey_t key, upb_value *v, + uint32_t hash, eqlfunc_t *eql) { + const upb_tabent *e = findentry(t, key, hash, eql); + if (e) { + if (v) { + _upb_value_setval(v, e->val, t->ctype); + } + return true; + } else { + return false; + } +} + +// The given key must not already exist in the table. +static void insert(upb_table *t, lookupkey_t key, upb_value val, + uint32_t hash, hashfunc_t *hashfunc, eqlfunc_t *eql) { + UPB_UNUSED(eql); + assert(findentry(t, key, hash, eql) == NULL); + assert(val.ctype == t->ctype); + t->count++; + upb_tabent *mainpos_e = getentry_mutable(t, hash); + upb_tabent *our_e = mainpos_e; + if (upb_tabent_isempty(mainpos_e)) { + // Our main position is empty; use it. + our_e->next = NULL; + } else { + // Collision. + upb_tabent *new_e = emptyent(t); + // Head of collider's chain. + upb_tabent *chain = getentry_mutable(t, hashfunc(mainpos_e->key)); + if (chain == mainpos_e) { + // Existing ent is in its main posisiton (it has the same hash as us, and + // is the head of our chain). Insert to new ent and append to this chain. + new_e->next = mainpos_e->next; + mainpos_e->next = new_e; + our_e = new_e; + } else { + // Existing ent is not in its main position (it is a node in some other + // chain). This implies that no existing ent in the table has our hash. + // Evict it (updating its chain) and use its ent for head of our chain. + *new_e = *mainpos_e; // copies next. + while (chain->next != mainpos_e) { + chain = (upb_tabent*)chain->next; + assert(chain); + } + chain->next = new_e; + our_e = mainpos_e; + our_e->next = NULL; + } + } + our_e->key = key.key; + our_e->val = val.val; + assert(findentry(t, key, hash, eql) == our_e); +} + +static bool rm(upb_table *t, lookupkey_t key, upb_value *val, + upb_tabkey *removed, uint32_t hash, eqlfunc_t *eql) { + upb_tabent *chain = getentry_mutable(t, hash); + if (upb_tabent_isempty(chain)) return false; + if (eql(chain->key, key)) { + // Element to remove is at the head of its chain. + t->count--; + if (val) { + _upb_value_setval(val, chain->val, t->ctype); + } + if (chain->next) { + upb_tabent *move = (upb_tabent*)chain->next; + *chain = *move; + if (removed) *removed = move->key; + move->key.num = 0; // Make the slot empty. + } else { + if (removed) *removed = chain->key; + chain->key.num = 0; // Make the slot empty. + } + return true; + } else { + // Element to remove is either in a non-head position or not in the table. + while (chain->next && !eql(chain->next->key, key)) + chain = (upb_tabent*)chain->next; + if (chain->next) { + // Found element to remove. + if (val) { + _upb_value_setval(val, chain->next->val, t->ctype); + } + upb_tabent *rm = (upb_tabent*)chain->next; + if (removed) *removed = rm->key; + rm->key.num = 0; + chain->next = rm->next; + t->count--; + return true; + } else { + return false; + } + } +} + +static size_t next(const upb_table *t, size_t i) { + do { + if (++i >= upb_table_size(t)) + return SIZE_MAX; + } while(upb_tabent_isempty(&t->entries[i])); + + return i; +} + +static size_t begin(const upb_table *t) { + return next(t, -1); +} + + +/* upb_strtable ***************************************************************/ + +// A simple "subclass" of upb_table that only adds a hash function for strings. + +static uint32_t strhash(upb_tabkey key) { + return MurmurHash2(key.s.str, key.s.length, 0); +} + +static bool streql(upb_tabkey k1, lookupkey_t k2) { + return k1.s.length == k2.key.s.length && + memcmp(k1.s.str, k2.key.s.str, k1.s.length) == 0; +} + +bool upb_strtable_init(upb_strtable *t, upb_ctype_t ctype) { + return init(&t->t, ctype, 2); +} + +void upb_strtable_uninit(upb_strtable *t) { + for (size_t i = 0; i < upb_table_size(&t->t); i++) + free((void*)t->t.entries[i].key.s.str); + uninit(&t->t); +} + +bool upb_strtable_resize(upb_strtable *t, size_t size_lg2) { + upb_strtable new_table; + if (!init(&new_table.t, t->t.ctype, size_lg2)) + return false; + upb_strtable_iter i; + upb_strtable_begin(&i, t); + for ( ; !upb_strtable_done(&i); upb_strtable_next(&i)) { + upb_strtable_insert2( + &new_table, + upb_strtable_iter_key(&i), + upb_strtable_iter_keylength(&i), + upb_strtable_iter_value(&i)); + } + upb_strtable_uninit(t); + *t = new_table; + return true; +} + +bool upb_strtable_insert2(upb_strtable *t, const char *k, size_t len, + upb_value v) { + if (isfull(&t->t)) { + // Need to resize. New table of double the size, add old elements to it. + if (!upb_strtable_resize(t, t->t.size_lg2 + 1)) { + return false; + } + } + if ((k = upb_strdup2(k, len)) == NULL) return false; + + lookupkey_t key = strkey2(k, len); + uint32_t hash = MurmurHash2(key.key.s.str, key.key.s.length, 0); + insert(&t->t, key, v, hash, &strhash, &streql); + return true; +} + +bool upb_strtable_lookup2(const upb_strtable *t, const char *key, size_t len, + upb_value *v) { + uint32_t hash = MurmurHash2(key, len, 0); + return lookup(&t->t, strkey2(key, len), v, hash, &streql); +} + +bool upb_strtable_remove2(upb_strtable *t, const char *key, size_t len, + upb_value *val) { + uint32_t hash = MurmurHash2(key, strlen(key), 0); + upb_tabkey tabkey; + if (rm(&t->t, strkey2(key, len), val, &tabkey, hash, &streql)) { + free((void*)tabkey.s.str); + return true; + } else { + return false; + } +} + +// Iteration + +static const upb_tabent *str_tabent(const upb_strtable_iter *i) { + return &i->t->t.entries[i->index]; +} + +void upb_strtable_begin(upb_strtable_iter *i, const upb_strtable *t) { + i->t = t; + i->index = begin(&t->t); +} + +void upb_strtable_next(upb_strtable_iter *i) { + i->index = next(&i->t->t, i->index); +} + +bool upb_strtable_done(const upb_strtable_iter *i) { + return i->index >= upb_table_size(&i->t->t) || + upb_tabent_isempty(str_tabent(i)); +} + +const char *upb_strtable_iter_key(upb_strtable_iter *i) { + assert(!upb_strtable_done(i)); + return str_tabent(i)->key.s.str; +} + +size_t upb_strtable_iter_keylength(upb_strtable_iter *i) { + assert(!upb_strtable_done(i)); + return str_tabent(i)->key.s.length; +} + +upb_value upb_strtable_iter_value(const upb_strtable_iter *i) { + assert(!upb_strtable_done(i)); + return _upb_value_val(str_tabent(i)->val, i->t->t.ctype); +} + +void upb_strtable_iter_setdone(upb_strtable_iter *i) { + i->index = SIZE_MAX; +} + +bool upb_strtable_iter_isequal(const upb_strtable_iter *i1, + const upb_strtable_iter *i2) { + if (upb_strtable_done(i1) && upb_strtable_done(i2)) + return true; + return i1->t == i2->t && i1->index == i2->index; +} + + +/* upb_inttable ***************************************************************/ + +// For inttables we use a hybrid structure where small keys are kept in an +// array and large keys are put in the hash table. + +static uint32_t inthash(upb_tabkey key) { return upb_inthash(key.num); } + +static bool inteql(upb_tabkey k1, lookupkey_t k2) { + return k1.num == k2.key.num; +} + +static _upb_value *mutable_array(upb_inttable *t) { + return (_upb_value*)t->array; +} + +static _upb_value *inttable_val(upb_inttable *t, uintptr_t key) { + if (key < t->array_size) { + return upb_arrhas(t->array[key]) ? &(mutable_array(t)[key]) : NULL; + } else { + upb_tabent *e = + findentry_mutable(&t->t, intkey(key), upb_inthash(key), &inteql); + return e ? &e->val : NULL; + } +} + +static const _upb_value *inttable_val_const(const upb_inttable *t, + uintptr_t key) { + return inttable_val((upb_inttable*)t, key); +} + +size_t upb_inttable_count(const upb_inttable *t) { + return t->t.count + t->array_count; +} + +static void check(upb_inttable *t) { + UPB_UNUSED(t); +#if defined(UPB_DEBUG_TABLE) && !defined(NDEBUG) + // This check is very expensive (makes inserts/deletes O(N)). + size_t count = 0; + upb_inttable_iter i; + upb_inttable_begin(&i, t); + for(; !upb_inttable_done(&i); upb_inttable_next(&i), count++) { + assert(upb_inttable_lookup(t, upb_inttable_iter_key(&i), NULL)); + } + assert(count == upb_inttable_count(t)); +#endif +} + +bool upb_inttable_sizedinit(upb_inttable *t, upb_ctype_t ctype, + size_t asize, int hsize_lg2) { + if (!init(&t->t, ctype, hsize_lg2)) return false; + // Always make the array part at least 1 long, so that we know key 0 + // won't be in the hash part, which simplifies things. + t->array_size = UPB_MAX(1, asize); + t->array_count = 0; + size_t array_bytes = t->array_size * sizeof(upb_value); + t->array = malloc(array_bytes); + if (!t->array) { + uninit(&t->t); + return false; + } + memset(mutable_array(t), 0xff, array_bytes); + check(t); + return true; +} + +bool upb_inttable_init(upb_inttable *t, upb_ctype_t ctype) { + return upb_inttable_sizedinit(t, ctype, 0, 4); +} + +void upb_inttable_uninit(upb_inttable *t) { + uninit(&t->t); + free(mutable_array(t)); +} + +bool upb_inttable_insert(upb_inttable *t, uintptr_t key, upb_value val) { + assert(upb_arrhas(val.val)); + if (key < t->array_size) { + assert(!upb_arrhas(t->array[key])); + t->array_count++; + mutable_array(t)[key] = val.val; + } else { + if (isfull(&t->t)) { + // Need to resize the hash part, but we re-use the array part. + upb_table new_table; + if (!init(&new_table, t->t.ctype, t->t.size_lg2 + 1)) + return false; + size_t i; + for (i = begin(&t->t); i < upb_table_size(&t->t); i = next(&t->t, i)) { + const upb_tabent *e = &t->t.entries[i]; + upb_value v; + _upb_value_setval(&v, e->val, t->t.ctype); + uint32_t hash = upb_inthash(e->key.num); + insert(&new_table, intkey(e->key.num), v, hash, &inthash, &inteql); + } + + assert(t->t.count == new_table.count); + + uninit(&t->t); + t->t = new_table; + } + insert(&t->t, intkey(key), val, upb_inthash(key), &inthash, &inteql); + } + check(t); + return true; +} + +bool upb_inttable_lookup(const upb_inttable *t, uintptr_t key, upb_value *v) { + const _upb_value *table_v = inttable_val_const(t, key); + if (!table_v) return false; + if (v) _upb_value_setval(v, *table_v, t->t.ctype); + return true; +} + +bool upb_inttable_replace(upb_inttable *t, uintptr_t key, upb_value val) { + _upb_value *table_v = inttable_val(t, key); + if (!table_v) return false; + *table_v = val.val; + return true; +} + +bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val) { + bool success; + if (key < t->array_size) { + if (upb_arrhas(t->array[key])) { + t->array_count--; + if (val) { + _upb_value_setval(val, t->array[key], t->t.ctype); + } + _upb_value empty = UPB_ARRAY_EMPTYENT; + mutable_array(t)[key] = empty; + success = true; + } else { + success = false; + } + } else { + upb_tabkey removed; + uint32_t hash = upb_inthash(key); + success = rm(&t->t, intkey(key), val, &removed, hash, &inteql); + } + check(t); + return success; +} + +bool upb_inttable_push(upb_inttable *t, upb_value val) { + return upb_inttable_insert(t, upb_inttable_count(t), val); +} + +upb_value upb_inttable_pop(upb_inttable *t) { + upb_value val; + bool ok = upb_inttable_remove(t, upb_inttable_count(t) - 1, &val); + UPB_ASSERT_VAR(ok, ok); + return val; +} + +bool upb_inttable_insertptr(upb_inttable *t, const void *key, upb_value val) { + return upb_inttable_insert(t, (uintptr_t)key, val); +} + +bool upb_inttable_lookupptr(const upb_inttable *t, const void *key, + upb_value *v) { + return upb_inttable_lookup(t, (uintptr_t)key, v); +} + +bool upb_inttable_removeptr(upb_inttable *t, const void *key, upb_value *val) { + return upb_inttable_remove(t, (uintptr_t)key, val); +} + +void upb_inttable_compact(upb_inttable *t) { + // Create a power-of-two histogram of the table keys. + int counts[UPB_MAXARRSIZE + 1] = {0}; + uintptr_t max_key = 0; + upb_inttable_iter i; + upb_inttable_begin(&i, t); + for (; !upb_inttable_done(&i); upb_inttable_next(&i)) { + uintptr_t key = upb_inttable_iter_key(&i); + if (key > max_key) { + max_key = key; + } + counts[log2ceil(key)]++; + } + + int arr_size; + int arr_count = upb_inttable_count(t); + + if (upb_inttable_count(t) >= max_key * MIN_DENSITY) { + // We can put 100% of the entries in the array part. + arr_size = max_key + 1; + } else { + // Find the largest power of two that satisfies the MIN_DENSITY definition. + for (int size_lg2 = ARRAY_SIZE(counts) - 1; size_lg2 > 1; size_lg2--) { + arr_size = 1 << size_lg2; + arr_count -= counts[size_lg2]; + if (arr_count >= arr_size * MIN_DENSITY) { + break; + } + } + } + + // Array part must always be at least 1 entry large to catch lookups of key + // 0. Key 0 must always be in the array part because "0" in the hash part + // denotes an empty entry. + arr_size = UPB_MAX(arr_size, 1); + + // Insert all elements into new, perfectly-sized table. + int hash_count = upb_inttable_count(t) - arr_count; + int hash_size = hash_count ? (hash_count / MAX_LOAD) + 1 : 0; + int hashsize_lg2 = log2ceil(hash_size); + assert(hash_count >= 0); + + upb_inttable new_t; + upb_inttable_sizedinit(&new_t, t->t.ctype, arr_size, hashsize_lg2); + upb_inttable_begin(&i, t); + for (; !upb_inttable_done(&i); upb_inttable_next(&i)) { + uintptr_t k = upb_inttable_iter_key(&i); + upb_inttable_insert(&new_t, k, upb_inttable_iter_value(&i)); + } + assert(new_t.array_size == arr_size); + assert(new_t.t.size_lg2 == hashsize_lg2); + upb_inttable_uninit(t); + *t = new_t; +} + +// Iteration. + +static const upb_tabent *int_tabent(const upb_inttable_iter *i) { + assert(!i->array_part); + return &i->t->t.entries[i->index]; +} + +static _upb_value int_arrent(const upb_inttable_iter *i) { + assert(i->array_part); + return i->t->array[i->index]; +} + +void upb_inttable_begin(upb_inttable_iter *i, const upb_inttable *t) { + i->t = t; + i->index = -1; + i->array_part = true; + upb_inttable_next(i); +} + +void upb_inttable_next(upb_inttable_iter *iter) { + const upb_inttable *t = iter->t; + if (iter->array_part) { + while (++iter->index < t->array_size) { + if (upb_arrhas(int_arrent(iter))) { + return; + } + } + iter->array_part = false; + iter->index = begin(&t->t); + } else { + iter->index = next(&t->t, iter->index); + } +} + +bool upb_inttable_done(const upb_inttable_iter *i) { + if (i->array_part) { + return i->index >= i->t->array_size || + !upb_arrhas(int_arrent(i)); + } else { + return i->index >= upb_table_size(&i->t->t) || + upb_tabent_isempty(int_tabent(i)); + } +} + +uintptr_t upb_inttable_iter_key(const upb_inttable_iter *i) { + assert(!upb_inttable_done(i)); + return i->array_part ? i->index : int_tabent(i)->key.num; +} + +upb_value upb_inttable_iter_value(const upb_inttable_iter *i) { + assert(!upb_inttable_done(i)); + return _upb_value_val( + i->array_part ? i->t->array[i->index] : int_tabent(i)->val, + i->t->t.ctype); +} + +void upb_inttable_iter_setdone(upb_inttable_iter *i) { + i->index = SIZE_MAX; + i->array_part = false; +} + +bool upb_inttable_iter_isequal(const upb_inttable_iter *i1, + const upb_inttable_iter *i2) { + if (upb_inttable_done(i1) && upb_inttable_done(i2)) + return true; + return i1->t == i2->t && i1->index == i2->index && + i1->array_part == i2->array_part; +} + +#ifdef UPB_UNALIGNED_READS_OK +//----------------------------------------------------------------------------- +// MurmurHash2, by Austin Appleby (released as public domain). +// Reformatted and C99-ified by Joshua Haberman. +// Note - This code makes a few assumptions about how your machine behaves - +// 1. We can read a 4-byte value from any address without crashing +// 2. sizeof(int) == 4 (in upb this limitation is removed by using uint32_t +// And it has a few limitations - +// 1. It will not work incrementally. +// 2. It will not produce the same results on little-endian and big-endian +// machines. +uint32_t MurmurHash2(const void *key, size_t len, uint32_t seed) { + // 'm' and 'r' are mixing constants generated offline. + // They're not really 'magic', they just happen to work well. + const uint32_t m = 0x5bd1e995; + const int32_t r = 24; + + // Initialize the hash to a 'random' value + uint32_t h = seed ^ len; + + // Mix 4 bytes at a time into the hash + const uint8_t * data = (const uint8_t *)key; + while(len >= 4) { + uint32_t k = *(uint32_t *)data; + + k *= m; + k ^= k >> r; + k *= m; + + h *= m; + h ^= k; + + data += 4; + len -= 4; + } + + // Handle the last few bytes of the input array + switch(len) { + case 3: h ^= data[2] << 16; + case 2: h ^= data[1] << 8; + case 1: h ^= data[0]; h *= m; + }; + + // Do a few final mixes of the hash to ensure the last few + // bytes are well-incorporated. + h ^= h >> 13; + h *= m; + h ^= h >> 15; + + return h; +} + +#else // !UPB_UNALIGNED_READS_OK + +//----------------------------------------------------------------------------- +// MurmurHashAligned2, by Austin Appleby +// Same algorithm as MurmurHash2, but only does aligned reads - should be safer +// on certain platforms. +// Performance will be lower than MurmurHash2 + +#define MIX(h,k,m) { k *= m; k ^= k >> r; k *= m; h *= m; h ^= k; } + +uint32_t MurmurHash2(const void * key, size_t len, uint32_t seed) { + const uint32_t m = 0x5bd1e995; + const int32_t r = 24; + const uint8_t * data = (const uint8_t *)key; + uint32_t h = seed ^ len; + uint8_t align = (uintptr_t)data & 3; + + if(align && (len >= 4)) { + // Pre-load the temp registers + uint32_t t = 0, d = 0; + + switch(align) { + case 1: t |= data[2] << 16; + case 2: t |= data[1] << 8; + case 3: t |= data[0]; + } + + t <<= (8 * align); + + data += 4-align; + len -= 4-align; + + int32_t sl = 8 * (4-align); + int32_t sr = 8 * align; + + // Mix + + while(len >= 4) { + d = *(uint32_t *)data; + t = (t >> sr) | (d << sl); + + uint32_t k = t; + + MIX(h,k,m); + + t = d; + + data += 4; + len -= 4; + } + + // Handle leftover data in temp registers + + d = 0; + + if(len >= align) { + switch(align) { + case 3: d |= data[2] << 16; + case 2: d |= data[1] << 8; + case 1: d |= data[0]; + } + + uint32_t k = (t >> sr) | (d << sl); + MIX(h,k,m); + + data += align; + len -= align; + + //---------- + // Handle tail bytes + + switch(len) { + case 3: h ^= data[2] << 16; + case 2: h ^= data[1] << 8; + case 1: h ^= data[0]; h *= m; + }; + } else { + switch(len) { + case 3: d |= data[2] << 16; + case 2: d |= data[1] << 8; + case 1: d |= data[0]; + case 0: h ^= (t >> sr) | (d << sl); h *= m; + } + } + + h ^= h >> 13; + h *= m; + h ^= h >> 15; + + return h; + } else { + while(len >= 4) { + uint32_t k = *(uint32_t *)data; + + MIX(h,k,m); + + data += 4; + len -= 4; + } + + //---------- + // Handle tail bytes + + switch(len) { + case 3: h ^= data[2] << 16; + case 2: h ^= data[1] << 8; + case 1: h ^= data[0]; h *= m; + }; + + h ^= h >> 13; + h *= m; + h ^= h >> 15; + + return h; + } +} +#undef MIX + +#endif // UPB_UNALIGNED_READS_OK +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + +#include +#include +#include +#include +#include +#include +#include + +bool upb_dumptostderr(void *closure, const upb_status* status) { + UPB_UNUSED(closure); + fprintf(stderr, "%s\n", upb_status_errmsg(status)); + return false; +} + +// Guarantee null-termination and provide ellipsis truncation. +// It may be tempting to "optimize" this by initializing these final +// four bytes up-front and then being careful never to overwrite them, +// this is safer and simpler. +static void nullz(upb_status *status) { + const char *ellipsis = "..."; + size_t len = strlen(ellipsis); + assert(sizeof(status->msg) > len); + memcpy(status->msg + sizeof(status->msg) - len, ellipsis, len); +} + +void upb_status_clear(upb_status *status) { + if (!status) return; + status->ok_ = true; + status->code_ = 0; + status->msg[0] = '\0'; +} + +bool upb_ok(const upb_status *status) { return status->ok_; } + +upb_errorspace *upb_status_errspace(const upb_status *status) { + return status->error_space_; +} + +int upb_status_errcode(const upb_status *status) { return status->code_; } + +const char *upb_status_errmsg(const upb_status *status) { return status->msg; } + +void upb_status_seterrmsg(upb_status *status, const char *msg) { + if (!status) return; + status->ok_ = false; + strncpy(status->msg, msg, sizeof(status->msg)); + nullz(status); +} + +void upb_status_seterrf(upb_status *status, const char *fmt, ...) { + va_list args; + va_start(args, fmt); + upb_status_vseterrf(status, fmt, args); + va_end(args); +} + +void upb_status_vseterrf(upb_status *status, const char *fmt, va_list args) { + if (!status) return; + status->ok_ = false; + vsnprintf(status->msg, sizeof(status->msg), fmt, args); + nullz(status); +} + +void upb_status_seterrcode(upb_status *status, upb_errorspace *space, + int code) { + if (!status) return; + status->ok_ = false; + status->error_space_ = space; + status->code_ = code; + space->set_message(status, code); +} + +void upb_status_copy(upb_status *to, const upb_status *from) { + if (!to) return; + *to = *from; +} +// This file was generated by upbc (the upb compiler). +// Do not edit -- your changes will be discarded when the file is +// regenerated. + + +static const upb_msgdef msgs[20]; +static const upb_fielddef fields[81]; +static const upb_enumdef enums[4]; +static const upb_tabent strentries[236]; +static const upb_tabent intentries[14]; +static const _upb_value arrays[232]; + +#ifdef UPB_DEBUG_REFS +static upb_inttable reftables[212]; +#endif + +static const upb_msgdef msgs[20] = { + UPB_MSGDEF_INIT("google.protobuf.DescriptorProto", 27, 6, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[0], 8, 7), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[0]),&reftables[0], &reftables[1]), + UPB_MSGDEF_INIT("google.protobuf.DescriptorProto.ExtensionRange", 4, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[8], 3, 2), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[16]),&reftables[2], &reftables[3]), + UPB_MSGDEF_INIT("google.protobuf.EnumDescriptorProto", 11, 2, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[11], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[20]),&reftables[4], &reftables[5]), + UPB_MSGDEF_INIT("google.protobuf.EnumOptions", 7, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[0], &arrays[15], 8, 1), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[24]),&reftables[6], &reftables[7]), + UPB_MSGDEF_INIT("google.protobuf.EnumValueDescriptorProto", 8, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[23], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[28]),&reftables[8], &reftables[9]), + UPB_MSGDEF_INIT("google.protobuf.EnumValueOptions", 6, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[2], &arrays[27], 4, 0), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[32]),&reftables[10], &reftables[11]), + UPB_MSGDEF_INIT("google.protobuf.FieldDescriptorProto", 19, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[31], 9, 8), UPB_STRTABLE_INIT(8, 15, UPB_CTYPE_PTR, 4, &strentries[36]),&reftables[12], &reftables[13]), + UPB_MSGDEF_INIT("google.protobuf.FieldOptions", 14, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[4], &arrays[40], 32, 6), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[52]),&reftables[14], &reftables[15]), + UPB_MSGDEF_INIT("google.protobuf.FileDescriptorProto", 39, 6, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[72], 12, 11), UPB_STRTABLE_INIT(11, 15, UPB_CTYPE_PTR, 4, &strentries[68]),&reftables[16], &reftables[17]), + UPB_MSGDEF_INIT("google.protobuf.FileDescriptorSet", 6, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[84], 2, 1), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[84]),&reftables[18], &reftables[19]), + UPB_MSGDEF_INIT("google.protobuf.FileOptions", 21, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[6], &arrays[86], 64, 9), UPB_STRTABLE_INIT(10, 15, UPB_CTYPE_PTR, 4, &strentries[88]),&reftables[20], &reftables[21]), + UPB_MSGDEF_INIT("google.protobuf.MessageOptions", 8, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[8], &arrays[150], 16, 2), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[104]),&reftables[22], &reftables[23]), + UPB_MSGDEF_INIT("google.protobuf.MethodDescriptorProto", 13, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[166], 5, 4), UPB_STRTABLE_INIT(4, 7, UPB_CTYPE_PTR, 3, &strentries[108]),&reftables[24], &reftables[25]), + UPB_MSGDEF_INIT("google.protobuf.MethodOptions", 6, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[10], &arrays[171], 4, 0), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[116]),&reftables[26], &reftables[27]), + UPB_MSGDEF_INIT("google.protobuf.ServiceDescriptorProto", 11, 2, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[175], 4, 3), UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_PTR, 2, &strentries[120]),&reftables[28], &reftables[29]), + UPB_MSGDEF_INIT("google.protobuf.ServiceOptions", 6, 1, UPB_INTTABLE_INIT(1, 1, UPB_CTYPE_PTR, 1, &intentries[12], &arrays[179], 4, 0), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[124]),&reftables[30], &reftables[31]), + UPB_MSGDEF_INIT("google.protobuf.SourceCodeInfo", 6, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[183], 2, 1), UPB_STRTABLE_INIT(1, 3, UPB_CTYPE_PTR, 2, &strentries[128]),&reftables[32], &reftables[33]), + UPB_MSGDEF_INIT("google.protobuf.SourceCodeInfo.Location", 14, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[185], 5, 4), UPB_STRTABLE_INIT(4, 7, UPB_CTYPE_PTR, 3, &strentries[132]),&reftables[34], &reftables[35]), + UPB_MSGDEF_INIT("google.protobuf.UninterpretedOption", 18, 1, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[190], 9, 7), UPB_STRTABLE_INIT(7, 15, UPB_CTYPE_PTR, 4, &strentries[140]),&reftables[36], &reftables[37]), + UPB_MSGDEF_INIT("google.protobuf.UninterpretedOption.NamePart", 6, 0, UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_PTR, 0, NULL, &arrays[199], 3, 2), UPB_STRTABLE_INIT(2, 3, UPB_CTYPE_PTR, 2, &strentries[156]),&reftables[38], &reftables[39]), +}; + +static const upb_fielddef fields[81] = { + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "aggregate_value", 8, &msgs[18], NULL, 15, 6, {0},&reftables[40], &reftables[41]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "allow_alias", 2, &msgs[3], NULL, 6, 1, {0},&reftables[42], &reftables[43]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "cc_generic_services", 16, &msgs[10], NULL, 17, 6, {0},&reftables[44], &reftables[45]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "ctype", 1, &msgs[7], UPB_UPCAST(&enums[2]), 6, 1, {0},&reftables[46], &reftables[47]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "default_value", 7, &msgs[6], NULL, 16, 7, {0},&reftables[48], &reftables[49]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_STRING, 0, false, false, false, false, "dependency", 3, &msgs[8], NULL, 30, 8, {0},&reftables[50], &reftables[51]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "deprecated", 3, &msgs[7], NULL, 8, 3, {0},&reftables[52], &reftables[53]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_DOUBLE, 0, false, false, false, false, "double_value", 6, &msgs[18], NULL, 11, 4, {0},&reftables[54], &reftables[55]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "end", 2, &msgs[1], NULL, 3, 1, {0},&reftables[56], &reftables[57]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "enum_type", 4, &msgs[0], UPB_UPCAST(&msgs[2]), 16, 2, {0},&reftables[58], &reftables[59]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "enum_type", 5, &msgs[8], UPB_UPCAST(&msgs[2]), 13, 1, {0},&reftables[60], &reftables[61]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "experimental_map_key", 9, &msgs[7], NULL, 10, 5, {0},&reftables[62], &reftables[63]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "extendee", 2, &msgs[6], NULL, 7, 2, {0},&reftables[64], &reftables[65]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension", 7, &msgs[8], UPB_UPCAST(&msgs[6]), 19, 3, {0},&reftables[66], &reftables[67]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension", 6, &msgs[0], UPB_UPCAST(&msgs[6]), 22, 4, {0},&reftables[68], &reftables[69]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "extension_range", 5, &msgs[0], UPB_UPCAST(&msgs[1]), 19, 3, {0},&reftables[70], &reftables[71]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "field", 2, &msgs[0], UPB_UPCAST(&msgs[6]), 10, 0, {0},&reftables[72], &reftables[73]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "file", 1, &msgs[9], UPB_UPCAST(&msgs[8]), 5, 0, {0},&reftables[74], &reftables[75]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "go_package", 11, &msgs[10], NULL, 14, 5, {0},&reftables[76], &reftables[77]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "identifier_value", 3, &msgs[18], NULL, 6, 1, {0},&reftables[78], &reftables[79]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "input_type", 2, &msgs[12], NULL, 7, 2, {0},&reftables[80], &reftables[81]), + UPB_FIELDDEF_INIT(UPB_LABEL_REQUIRED, UPB_TYPE_BOOL, 0, false, false, false, false, "is_extension", 2, &msgs[19], NULL, 5, 1, {0},&reftables[82], &reftables[83]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_generate_equals_and_hash", 20, &msgs[10], NULL, 20, 9, {0},&reftables[84], &reftables[85]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_generic_services", 17, &msgs[10], NULL, 18, 7, {0},&reftables[86], &reftables[87]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "java_multiple_files", 10, &msgs[10], NULL, 13, 4, {0},&reftables[88], &reftables[89]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "java_outer_classname", 8, &msgs[10], NULL, 9, 2, {0},&reftables[90], &reftables[91]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "java_package", 1, &msgs[10], NULL, 6, 1, {0},&reftables[92], &reftables[93]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "label", 4, &msgs[6], UPB_UPCAST(&enums[0]), 11, 4, {0},&reftables[94], &reftables[95]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "lazy", 5, &msgs[7], NULL, 9, 4, {0},&reftables[96], &reftables[97]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "leading_comments", 3, &msgs[17], NULL, 8, 2, {0},&reftables[98], &reftables[99]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "location", 1, &msgs[16], UPB_UPCAST(&msgs[17]), 5, 0, {0},&reftables[100], &reftables[101]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "message_set_wire_format", 1, &msgs[11], NULL, 6, 1, {0},&reftables[102], &reftables[103]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "message_type", 4, &msgs[8], UPB_UPCAST(&msgs[0]), 10, 0, {0},&reftables[104], &reftables[105]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "method", 2, &msgs[14], UPB_UPCAST(&msgs[12]), 6, 0, {0},&reftables[106], &reftables[107]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[8], NULL, 22, 6, {0},&reftables[108], &reftables[109]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[14], NULL, 8, 2, {0},&reftables[110], &reftables[111]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "name", 2, &msgs[18], UPB_UPCAST(&msgs[19]), 5, 0, {0},&reftables[112], &reftables[113]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[4], NULL, 4, 1, {0},&reftables[114], &reftables[115]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[0], NULL, 24, 6, {0},&reftables[116], &reftables[117]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[12], NULL, 4, 1, {0},&reftables[118], &reftables[119]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[2], NULL, 8, 2, {0},&reftables[120], &reftables[121]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "name", 1, &msgs[6], NULL, 4, 1, {0},&reftables[122], &reftables[123]), + UPB_FIELDDEF_INIT(UPB_LABEL_REQUIRED, UPB_TYPE_STRING, 0, false, false, false, false, "name_part", 1, &msgs[19], NULL, 2, 0, {0},&reftables[124], &reftables[125]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT64, UPB_INTFMT_VARIABLE, false, false, false, false, "negative_int_value", 5, &msgs[18], NULL, 10, 3, {0},&reftables[126], &reftables[127]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "nested_type", 3, &msgs[0], UPB_UPCAST(&msgs[0]), 13, 1, {0},&reftables[128], &reftables[129]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "no_standard_descriptor_accessor", 2, &msgs[11], NULL, 7, 2, {0},&reftables[130], &reftables[131]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "number", 3, &msgs[6], NULL, 10, 3, {0},&reftables[132], &reftables[133]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "number", 2, &msgs[4], NULL, 7, 2, {0},&reftables[134], &reftables[135]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "optimize_for", 9, &msgs[10], UPB_UPCAST(&enums[3]), 12, 3, {0},&reftables[136], &reftables[137]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 7, &msgs[0], UPB_UPCAST(&msgs[11]), 23, 5, {0},&reftables[138], &reftables[139]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[2], UPB_UPCAST(&msgs[3]), 7, 1, {0},&reftables[140], &reftables[141]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 8, &msgs[6], UPB_UPCAST(&msgs[7]), 3, 0, {0},&reftables[142], &reftables[143]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[4], UPB_UPCAST(&msgs[5]), 3, 0, {0},&reftables[144], &reftables[145]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 8, &msgs[8], UPB_UPCAST(&msgs[10]), 20, 4, {0},&reftables[146], &reftables[147]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 3, &msgs[14], UPB_UPCAST(&msgs[15]), 7, 1, {0},&reftables[148], &reftables[149]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "options", 4, &msgs[12], UPB_UPCAST(&msgs[13]), 3, 0, {0},&reftables[150], &reftables[151]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "output_type", 3, &msgs[12], NULL, 10, 3, {0},&reftables[152], &reftables[153]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "package", 2, &msgs[8], NULL, 25, 7, {0},&reftables[154], &reftables[155]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "packed", 2, &msgs[7], NULL, 7, 2, {0},&reftables[156], &reftables[157]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, true, "path", 1, &msgs[17], NULL, 4, 0, {0},&reftables[158], &reftables[159]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_UINT64, UPB_INTFMT_VARIABLE, false, false, false, false, "positive_int_value", 4, &msgs[18], NULL, 9, 2, {0},&reftables[160], &reftables[161]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "public_dependency", 10, &msgs[8], NULL, 35, 9, {0},&reftables[162], &reftables[163]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "py_generic_services", 18, &msgs[10], NULL, 19, 8, {0},&reftables[164], &reftables[165]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "service", 6, &msgs[8], UPB_UPCAST(&msgs[14]), 16, 2, {0},&reftables[166], &reftables[167]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_MESSAGE, 0, false, false, false, false, "source_code_info", 9, &msgs[8], UPB_UPCAST(&msgs[16]), 21, 5, {0},&reftables[168], &reftables[169]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, true, "span", 2, &msgs[17], NULL, 7, 1, {0},&reftables[170], &reftables[171]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "start", 1, &msgs[1], NULL, 2, 0, {0},&reftables[172], &reftables[173]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BYTES, 0, false, false, false, false, "string_value", 7, &msgs[18], NULL, 12, 5, {0},&reftables[174], &reftables[175]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "trailing_comments", 4, &msgs[17], NULL, 11, 3, {0},&reftables[176], &reftables[177]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_ENUM, 0, false, false, false, false, "type", 5, &msgs[6], UPB_UPCAST(&enums[1]), 12, 5, {0},&reftables[178], &reftables[179]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_STRING, 0, false, false, false, false, "type_name", 6, &msgs[6], NULL, 13, 6, {0},&reftables[180], &reftables[181]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[5], UPB_UPCAST(&msgs[18]), 5, 0, {0},&reftables[182], &reftables[183]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[15], UPB_UPCAST(&msgs[18]), 5, 0, {0},&reftables[184], &reftables[185]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[3], UPB_UPCAST(&msgs[18]), 5, 0, {0},&reftables[186], &reftables[187]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[13], UPB_UPCAST(&msgs[18]), 5, 0, {0},&reftables[188], &reftables[189]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[10], UPB_UPCAST(&msgs[18]), 5, 0, {0},&reftables[190], &reftables[191]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[11], UPB_UPCAST(&msgs[18]), 5, 0, {0},&reftables[192], &reftables[193]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "uninterpreted_option", 999, &msgs[7], UPB_UPCAST(&msgs[18]), 5, 0, {0},&reftables[194], &reftables[195]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_MESSAGE, 0, false, false, false, false, "value", 2, &msgs[2], UPB_UPCAST(&msgs[4]), 6, 0, {0},&reftables[196], &reftables[197]), + UPB_FIELDDEF_INIT(UPB_LABEL_OPTIONAL, UPB_TYPE_BOOL, 0, false, false, false, false, "weak", 10, &msgs[7], NULL, 13, 6, {0},&reftables[198], &reftables[199]), + UPB_FIELDDEF_INIT(UPB_LABEL_REPEATED, UPB_TYPE_INT32, UPB_INTFMT_VARIABLE, false, false, false, false, "weak_dependency", 11, &msgs[8], NULL, 38, 10, {0},&reftables[200], &reftables[201]), +}; + +static const upb_enumdef enums[4] = { + UPB_ENUMDEF_INIT("google.protobuf.FieldDescriptorProto.Label", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[160]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[202], 4, 3), 0, &reftables[202], &reftables[203]), + UPB_ENUMDEF_INIT("google.protobuf.FieldDescriptorProto.Type", UPB_STRTABLE_INIT(18, 31, UPB_CTYPE_INT32, 5, &strentries[164]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[206], 19, 18), 0, &reftables[204], &reftables[205]), + UPB_ENUMDEF_INIT("google.protobuf.FieldOptions.CType", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[196]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[225], 3, 3), 0, &reftables[206], &reftables[207]), + UPB_ENUMDEF_INIT("google.protobuf.FileOptions.OptimizeMode", UPB_STRTABLE_INIT(3, 3, UPB_CTYPE_INT32, 2, &strentries[200]), UPB_INTTABLE_INIT(0, 0, UPB_CTYPE_CSTR, 0, NULL, &arrays[228], 4, 3), 0, &reftables[208], &reftables[209]), +}; + +static const upb_tabent strentries[236] = { + {UPB_TABKEY_STR("extension"), UPB_VALUE_INIT_CONSTPTR(&fields[14]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[38]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("field"), UPB_VALUE_INIT_CONSTPTR(&fields[16]), NULL}, + {UPB_TABKEY_STR("extension_range"), UPB_VALUE_INIT_CONSTPTR(&fields[15]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("nested_type"), UPB_VALUE_INIT_CONSTPTR(&fields[44]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("options"), UPB_VALUE_INIT_CONSTPTR(&fields[49]), NULL}, + {UPB_TABKEY_STR("enum_type"), UPB_VALUE_INIT_CONSTPTR(&fields[9]), &strentries[14]}, + {UPB_TABKEY_STR("start"), UPB_VALUE_INIT_CONSTPTR(&fields[66]), NULL}, + {UPB_TABKEY_STR("end"), UPB_VALUE_INIT_CONSTPTR(&fields[8]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("value"), UPB_VALUE_INIT_CONSTPTR(&fields[78]), NULL}, + {UPB_TABKEY_STR("options"), UPB_VALUE_INIT_CONSTPTR(&fields[50]), NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[40]), &strentries[22]}, + {UPB_TABKEY_STR("uninterpreted_option"), UPB_VALUE_INIT_CONSTPTR(&fields[73]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("allow_alias"), UPB_VALUE_INIT_CONSTPTR(&fields[1]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("number"), UPB_VALUE_INIT_CONSTPTR(&fields[47]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("options"), UPB_VALUE_INIT_CONSTPTR(&fields[52]), NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[37]), &strentries[30]}, + {UPB_TABKEY_STR("uninterpreted_option"), UPB_VALUE_INIT_CONSTPTR(&fields[71]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("label"), UPB_VALUE_INIT_CONSTPTR(&fields[27]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[41]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("number"), UPB_VALUE_INIT_CONSTPTR(&fields[46]), &strentries[49]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("type_name"), UPB_VALUE_INIT_CONSTPTR(&fields[70]), NULL}, + {UPB_TABKEY_STR("extendee"), UPB_VALUE_INIT_CONSTPTR(&fields[12]), NULL}, + {UPB_TABKEY_STR("type"), UPB_VALUE_INIT_CONSTPTR(&fields[69]), &strentries[48]}, + {UPB_TABKEY_STR("default_value"), UPB_VALUE_INIT_CONSTPTR(&fields[4]), NULL}, + {UPB_TABKEY_STR("options"), UPB_VALUE_INIT_CONSTPTR(&fields[51]), NULL}, + {UPB_TABKEY_STR("experimental_map_key"), UPB_VALUE_INIT_CONSTPTR(&fields[11]), &strentries[67]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("weak"), UPB_VALUE_INIT_CONSTPTR(&fields[79]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("packed"), UPB_VALUE_INIT_CONSTPTR(&fields[58]), NULL}, + {UPB_TABKEY_STR("lazy"), UPB_VALUE_INIT_CONSTPTR(&fields[28]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("ctype"), UPB_VALUE_INIT_CONSTPTR(&fields[3]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("deprecated"), UPB_VALUE_INIT_CONSTPTR(&fields[6]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("uninterpreted_option"), UPB_VALUE_INIT_CONSTPTR(&fields[77]), NULL}, + {UPB_TABKEY_STR("extension"), UPB_VALUE_INIT_CONSTPTR(&fields[13]), NULL}, + {UPB_TABKEY_STR("weak_dependency"), UPB_VALUE_INIT_CONSTPTR(&fields[80]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[34]), NULL}, + {UPB_TABKEY_STR("service"), UPB_VALUE_INIT_CONSTPTR(&fields[63]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("source_code_info"), UPB_VALUE_INIT_CONSTPTR(&fields[64]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("dependency"), UPB_VALUE_INIT_CONSTPTR(&fields[5]), NULL}, + {UPB_TABKEY_STR("message_type"), UPB_VALUE_INIT_CONSTPTR(&fields[32]), NULL}, + {UPB_TABKEY_STR("package"), UPB_VALUE_INIT_CONSTPTR(&fields[57]), NULL}, + {UPB_TABKEY_STR("options"), UPB_VALUE_INIT_CONSTPTR(&fields[53]), &strentries[82]}, + {UPB_TABKEY_STR("enum_type"), UPB_VALUE_INIT_CONSTPTR(&fields[10]), NULL}, + {UPB_TABKEY_STR("public_dependency"), UPB_VALUE_INIT_CONSTPTR(&fields[61]), &strentries[81]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("file"), UPB_VALUE_INIT_CONSTPTR(&fields[17]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("uninterpreted_option"), UPB_VALUE_INIT_CONSTPTR(&fields[75]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("cc_generic_services"), UPB_VALUE_INIT_CONSTPTR(&fields[2]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("java_multiple_files"), UPB_VALUE_INIT_CONSTPTR(&fields[24]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("java_generic_services"), UPB_VALUE_INIT_CONSTPTR(&fields[23]), &strentries[102]}, + {UPB_TABKEY_STR("java_generate_equals_and_hash"), UPB_VALUE_INIT_CONSTPTR(&fields[22]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("go_package"), UPB_VALUE_INIT_CONSTPTR(&fields[18]), NULL}, + {UPB_TABKEY_STR("java_package"), UPB_VALUE_INIT_CONSTPTR(&fields[26]), NULL}, + {UPB_TABKEY_STR("optimize_for"), UPB_VALUE_INIT_CONSTPTR(&fields[48]), NULL}, + {UPB_TABKEY_STR("py_generic_services"), UPB_VALUE_INIT_CONSTPTR(&fields[62]), NULL}, + {UPB_TABKEY_STR("java_outer_classname"), UPB_VALUE_INIT_CONSTPTR(&fields[25]), NULL}, + {UPB_TABKEY_STR("message_set_wire_format"), UPB_VALUE_INIT_CONSTPTR(&fields[31]), &strentries[106]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("uninterpreted_option"), UPB_VALUE_INIT_CONSTPTR(&fields[76]), NULL}, + {UPB_TABKEY_STR("no_standard_descriptor_accessor"), UPB_VALUE_INIT_CONSTPTR(&fields[45]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[39]), NULL}, + {UPB_TABKEY_STR("input_type"), UPB_VALUE_INIT_CONSTPTR(&fields[20]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("output_type"), UPB_VALUE_INIT_CONSTPTR(&fields[56]), NULL}, + {UPB_TABKEY_STR("options"), UPB_VALUE_INIT_CONSTPTR(&fields[55]), NULL}, + {UPB_TABKEY_STR("uninterpreted_option"), UPB_VALUE_INIT_CONSTPTR(&fields[74]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("options"), UPB_VALUE_INIT_CONSTPTR(&fields[54]), &strentries[122]}, + {UPB_TABKEY_STR("method"), UPB_VALUE_INIT_CONSTPTR(&fields[33]), NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[35]), &strentries[121]}, + {UPB_TABKEY_STR("uninterpreted_option"), UPB_VALUE_INIT_CONSTPTR(&fields[72]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("location"), UPB_VALUE_INIT_CONSTPTR(&fields[30]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("span"), UPB_VALUE_INIT_CONSTPTR(&fields[65]), &strentries[139]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("trailing_comments"), UPB_VALUE_INIT_CONSTPTR(&fields[68]), NULL}, + {UPB_TABKEY_STR("leading_comments"), UPB_VALUE_INIT_CONSTPTR(&fields[29]), &strentries[137]}, + {UPB_TABKEY_STR("path"), UPB_VALUE_INIT_CONSTPTR(&fields[59]), NULL}, + {UPB_TABKEY_STR("double_value"), UPB_VALUE_INIT_CONSTPTR(&fields[7]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("name"), UPB_VALUE_INIT_CONSTPTR(&fields[36]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("negative_int_value"), UPB_VALUE_INIT_CONSTPTR(&fields[43]), NULL}, + {UPB_TABKEY_STR("aggregate_value"), UPB_VALUE_INIT_CONSTPTR(&fields[0]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("positive_int_value"), UPB_VALUE_INIT_CONSTPTR(&fields[60]), NULL}, + {UPB_TABKEY_STR("identifier_value"), UPB_VALUE_INIT_CONSTPTR(&fields[19]), NULL}, + {UPB_TABKEY_STR("string_value"), UPB_VALUE_INIT_CONSTPTR(&fields[67]), &strentries[154]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("is_extension"), UPB_VALUE_INIT_CONSTPTR(&fields[21]), NULL}, + {UPB_TABKEY_STR("name_part"), UPB_VALUE_INIT_CONSTPTR(&fields[42]), NULL}, + {UPB_TABKEY_STR("LABEL_REQUIRED"), UPB_VALUE_INIT_INT32(2), &strentries[162]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("LABEL_REPEATED"), UPB_VALUE_INIT_INT32(3), NULL}, + {UPB_TABKEY_STR("LABEL_OPTIONAL"), UPB_VALUE_INIT_INT32(1), NULL}, + {UPB_TABKEY_STR("TYPE_FIXED64"), UPB_VALUE_INIT_INT32(6), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("TYPE_STRING"), UPB_VALUE_INIT_INT32(9), NULL}, + {UPB_TABKEY_STR("TYPE_FLOAT"), UPB_VALUE_INIT_INT32(2), &strentries[193]}, + {UPB_TABKEY_STR("TYPE_DOUBLE"), UPB_VALUE_INIT_INT32(1), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("TYPE_INT32"), UPB_VALUE_INIT_INT32(5), NULL}, + {UPB_TABKEY_STR("TYPE_SFIXED32"), UPB_VALUE_INIT_INT32(15), NULL}, + {UPB_TABKEY_STR("TYPE_FIXED32"), UPB_VALUE_INIT_INT32(7), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("TYPE_MESSAGE"), UPB_VALUE_INIT_INT32(11), &strentries[194]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("TYPE_INT64"), UPB_VALUE_INIT_INT32(3), &strentries[191]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("TYPE_ENUM"), UPB_VALUE_INIT_INT32(14), NULL}, + {UPB_TABKEY_STR("TYPE_UINT32"), UPB_VALUE_INIT_INT32(13), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("TYPE_UINT64"), UPB_VALUE_INIT_INT32(4), &strentries[190]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("TYPE_SFIXED64"), UPB_VALUE_INIT_INT32(16), NULL}, + {UPB_TABKEY_STR("TYPE_BYTES"), UPB_VALUE_INIT_INT32(12), NULL}, + {UPB_TABKEY_STR("TYPE_SINT64"), UPB_VALUE_INIT_INT32(18), NULL}, + {UPB_TABKEY_STR("TYPE_BOOL"), UPB_VALUE_INIT_INT32(8), NULL}, + {UPB_TABKEY_STR("TYPE_GROUP"), UPB_VALUE_INIT_INT32(10), NULL}, + {UPB_TABKEY_STR("TYPE_SINT32"), UPB_VALUE_INIT_INT32(17), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("CORD"), UPB_VALUE_INIT_INT32(1), NULL}, + {UPB_TABKEY_STR("STRING"), UPB_VALUE_INIT_INT32(0), &strentries[197]}, + {UPB_TABKEY_STR("STRING_PIECE"), UPB_VALUE_INIT_INT32(2), NULL}, + {UPB_TABKEY_STR("CODE_SIZE"), UPB_VALUE_INIT_INT32(2), NULL}, + {UPB_TABKEY_STR("SPEED"), UPB_VALUE_INIT_INT32(1), &strentries[203]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("LITE_RUNTIME"), UPB_VALUE_INIT_INT32(3), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("google.protobuf.SourceCodeInfo.Location"), UPB_VALUE_INIT_CONSTPTR(&msgs[17]), NULL}, + {UPB_TABKEY_STR("google.protobuf.UninterpretedOption"), UPB_VALUE_INIT_CONSTPTR(&msgs[18]), NULL}, + {UPB_TABKEY_STR("google.protobuf.FileDescriptorProto"), UPB_VALUE_INIT_CONSTPTR(&msgs[8]), NULL}, + {UPB_TABKEY_STR("google.protobuf.MethodDescriptorProto"), UPB_VALUE_INIT_CONSTPTR(&msgs[12]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("google.protobuf.EnumValueOptions"), UPB_VALUE_INIT_CONSTPTR(&msgs[5]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("google.protobuf.DescriptorProto"), UPB_VALUE_INIT_CONSTPTR(&msgs[0]), &strentries[228]}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("google.protobuf.SourceCodeInfo"), UPB_VALUE_INIT_CONSTPTR(&msgs[16]), NULL}, + {UPB_TABKEY_STR("google.protobuf.FieldDescriptorProto.Type"), UPB_VALUE_INIT_CONSTPTR(&enums[1]), NULL}, + {UPB_TABKEY_STR("google.protobuf.DescriptorProto.ExtensionRange"), UPB_VALUE_INIT_CONSTPTR(&msgs[1]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_STR("google.protobuf.EnumValueDescriptorProto"), UPB_VALUE_INIT_CONSTPTR(&msgs[4]), NULL}, + {UPB_TABKEY_STR("google.protobuf.FieldOptions"), UPB_VALUE_INIT_CONSTPTR(&msgs[7]), NULL}, + {UPB_TABKEY_STR("google.protobuf.FileOptions"), UPB_VALUE_INIT_CONSTPTR(&msgs[10]), NULL}, + {UPB_TABKEY_STR("google.protobuf.EnumDescriptorProto"), UPB_VALUE_INIT_CONSTPTR(&msgs[2]), &strentries[233]}, + {UPB_TABKEY_STR("google.protobuf.FieldDescriptorProto.Label"), UPB_VALUE_INIT_CONSTPTR(&enums[0]), NULL}, + {UPB_TABKEY_STR("google.protobuf.ServiceDescriptorProto"), UPB_VALUE_INIT_CONSTPTR(&msgs[14]), NULL}, + {UPB_TABKEY_STR("google.protobuf.FieldOptions.CType"), UPB_VALUE_INIT_CONSTPTR(&enums[2]), &strentries[229]}, + {UPB_TABKEY_STR("google.protobuf.FileDescriptorSet"), UPB_VALUE_INIT_CONSTPTR(&msgs[9]), &strentries[235]}, + {UPB_TABKEY_STR("google.protobuf.EnumOptions"), UPB_VALUE_INIT_CONSTPTR(&msgs[3]), NULL}, + {UPB_TABKEY_STR("google.protobuf.FieldDescriptorProto"), UPB_VALUE_INIT_CONSTPTR(&msgs[6]), NULL}, + {UPB_TABKEY_STR("google.protobuf.FileOptions.OptimizeMode"), UPB_VALUE_INIT_CONSTPTR(&enums[3]), &strentries[221]}, + {UPB_TABKEY_STR("google.protobuf.ServiceOptions"), UPB_VALUE_INIT_CONSTPTR(&msgs[15]), NULL}, + {UPB_TABKEY_STR("google.protobuf.MessageOptions"), UPB_VALUE_INIT_CONSTPTR(&msgs[11]), NULL}, + {UPB_TABKEY_STR("google.protobuf.MethodOptions"), UPB_VALUE_INIT_CONSTPTR(&msgs[13]), &strentries[226]}, + {UPB_TABKEY_STR("google.protobuf.UninterpretedOption.NamePart"), UPB_VALUE_INIT_CONSTPTR(&msgs[19]), NULL}, +}; + +static const upb_tabent intentries[14] = { + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NUM(999), UPB_VALUE_INIT_CONSTPTR(&fields[73]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NUM(999), UPB_VALUE_INIT_CONSTPTR(&fields[71]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NUM(999), UPB_VALUE_INIT_CONSTPTR(&fields[77]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NUM(999), UPB_VALUE_INIT_CONSTPTR(&fields[75]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NUM(999), UPB_VALUE_INIT_CONSTPTR(&fields[76]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NUM(999), UPB_VALUE_INIT_CONSTPTR(&fields[74]), NULL}, + {UPB_TABKEY_NONE, UPB__VALUE_INIT_NONE, NULL}, + {UPB_TABKEY_NUM(999), UPB_VALUE_INIT_CONSTPTR(&fields[72]), NULL}, +}; + +static const _upb_value arrays[232] = { + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[38]), + UPB_VALUE_INIT_CONSTPTR(&fields[16]), + UPB_VALUE_INIT_CONSTPTR(&fields[44]), + UPB_VALUE_INIT_CONSTPTR(&fields[9]), + UPB_VALUE_INIT_CONSTPTR(&fields[15]), + UPB_VALUE_INIT_CONSTPTR(&fields[14]), + UPB_VALUE_INIT_CONSTPTR(&fields[49]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[66]), + UPB_VALUE_INIT_CONSTPTR(&fields[8]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[40]), + UPB_VALUE_INIT_CONSTPTR(&fields[78]), + UPB_VALUE_INIT_CONSTPTR(&fields[50]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[1]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[37]), + UPB_VALUE_INIT_CONSTPTR(&fields[47]), + UPB_VALUE_INIT_CONSTPTR(&fields[52]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[41]), + UPB_VALUE_INIT_CONSTPTR(&fields[12]), + UPB_VALUE_INIT_CONSTPTR(&fields[46]), + UPB_VALUE_INIT_CONSTPTR(&fields[27]), + UPB_VALUE_INIT_CONSTPTR(&fields[69]), + UPB_VALUE_INIT_CONSTPTR(&fields[70]), + UPB_VALUE_INIT_CONSTPTR(&fields[4]), + UPB_VALUE_INIT_CONSTPTR(&fields[51]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[3]), + UPB_VALUE_INIT_CONSTPTR(&fields[58]), + UPB_VALUE_INIT_CONSTPTR(&fields[6]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[28]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[11]), + UPB_VALUE_INIT_CONSTPTR(&fields[79]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[34]), + UPB_VALUE_INIT_CONSTPTR(&fields[57]), + UPB_VALUE_INIT_CONSTPTR(&fields[5]), + UPB_VALUE_INIT_CONSTPTR(&fields[32]), + UPB_VALUE_INIT_CONSTPTR(&fields[10]), + UPB_VALUE_INIT_CONSTPTR(&fields[63]), + UPB_VALUE_INIT_CONSTPTR(&fields[13]), + UPB_VALUE_INIT_CONSTPTR(&fields[53]), + UPB_VALUE_INIT_CONSTPTR(&fields[64]), + UPB_VALUE_INIT_CONSTPTR(&fields[61]), + UPB_VALUE_INIT_CONSTPTR(&fields[80]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[17]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[26]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[25]), + UPB_VALUE_INIT_CONSTPTR(&fields[48]), + UPB_VALUE_INIT_CONSTPTR(&fields[24]), + UPB_VALUE_INIT_CONSTPTR(&fields[18]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[2]), + UPB_VALUE_INIT_CONSTPTR(&fields[23]), + UPB_VALUE_INIT_CONSTPTR(&fields[62]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[22]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[31]), + UPB_VALUE_INIT_CONSTPTR(&fields[45]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[39]), + UPB_VALUE_INIT_CONSTPTR(&fields[20]), + UPB_VALUE_INIT_CONSTPTR(&fields[56]), + UPB_VALUE_INIT_CONSTPTR(&fields[55]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[35]), + UPB_VALUE_INIT_CONSTPTR(&fields[33]), + UPB_VALUE_INIT_CONSTPTR(&fields[54]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[30]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[59]), + UPB_VALUE_INIT_CONSTPTR(&fields[65]), + UPB_VALUE_INIT_CONSTPTR(&fields[29]), + UPB_VALUE_INIT_CONSTPTR(&fields[68]), + UPB_ARRAY_EMPTYENT, + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[36]), + UPB_VALUE_INIT_CONSTPTR(&fields[19]), + UPB_VALUE_INIT_CONSTPTR(&fields[60]), + UPB_VALUE_INIT_CONSTPTR(&fields[43]), + UPB_VALUE_INIT_CONSTPTR(&fields[7]), + UPB_VALUE_INIT_CONSTPTR(&fields[67]), + UPB_VALUE_INIT_CONSTPTR(&fields[0]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR(&fields[42]), + UPB_VALUE_INIT_CONSTPTR(&fields[21]), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR("LABEL_OPTIONAL"), + UPB_VALUE_INIT_CONSTPTR("LABEL_REQUIRED"), + UPB_VALUE_INIT_CONSTPTR("LABEL_REPEATED"), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR("TYPE_DOUBLE"), + UPB_VALUE_INIT_CONSTPTR("TYPE_FLOAT"), + UPB_VALUE_INIT_CONSTPTR("TYPE_INT64"), + UPB_VALUE_INIT_CONSTPTR("TYPE_UINT64"), + UPB_VALUE_INIT_CONSTPTR("TYPE_INT32"), + UPB_VALUE_INIT_CONSTPTR("TYPE_FIXED64"), + UPB_VALUE_INIT_CONSTPTR("TYPE_FIXED32"), + UPB_VALUE_INIT_CONSTPTR("TYPE_BOOL"), + UPB_VALUE_INIT_CONSTPTR("TYPE_STRING"), + UPB_VALUE_INIT_CONSTPTR("TYPE_GROUP"), + UPB_VALUE_INIT_CONSTPTR("TYPE_MESSAGE"), + UPB_VALUE_INIT_CONSTPTR("TYPE_BYTES"), + UPB_VALUE_INIT_CONSTPTR("TYPE_UINT32"), + UPB_VALUE_INIT_CONSTPTR("TYPE_ENUM"), + UPB_VALUE_INIT_CONSTPTR("TYPE_SFIXED32"), + UPB_VALUE_INIT_CONSTPTR("TYPE_SFIXED64"), + UPB_VALUE_INIT_CONSTPTR("TYPE_SINT32"), + UPB_VALUE_INIT_CONSTPTR("TYPE_SINT64"), + UPB_VALUE_INIT_CONSTPTR("STRING"), + UPB_VALUE_INIT_CONSTPTR("CORD"), + UPB_VALUE_INIT_CONSTPTR("STRING_PIECE"), + UPB_ARRAY_EMPTYENT, + UPB_VALUE_INIT_CONSTPTR("SPEED"), + UPB_VALUE_INIT_CONSTPTR("CODE_SIZE"), + UPB_VALUE_INIT_CONSTPTR("LITE_RUNTIME"), +}; + +static const upb_symtab symtab = UPB_SYMTAB_INIT(UPB_STRTABLE_INIT(24, 31, UPB_CTYPE_PTR, 5, &strentries[204]), &reftables[210], &reftables[211]); + +const upb_symtab *upbdefs_google_protobuf_descriptor(const void *owner) { + upb_symtab_ref(&symtab, owner); + return &symtab; +} + +#ifdef UPB_DEBUG_REFS +static upb_inttable reftables[212] = { + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), + UPB_EMPTY_INTTABLE_INIT(UPB_CTYPE_PTR), +}; +#endif + +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2008-2009 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * XXX: The routines in this file that consume a string do not currently + * support having the string span buffers. In the future, as upb_sink and + * its buffering/sharing functionality evolve there should be an easy and + * idiomatic way of correctly handling this case. For now, we accept this + * limitation since we currently only parse descriptors from single strings. + */ + + +#include +#include +#include + +static char *upb_strndup(const char *buf, size_t n) { + char *ret = malloc(n + 1); + if (!ret) return NULL; + memcpy(ret, buf, n); + ret[n] = '\0'; + return ret; +} + +// Returns a newly allocated string that joins input strings together, for +// example: +// join("Foo.Bar", "Baz") -> "Foo.Bar.Baz" +// join("", "Baz") -> "Baz" +// Caller owns a ref on the returned string. +static char *upb_join(const char *base, const char *name) { + if (!base || strlen(base) == 0) { + return upb_strdup(name); + } else { + char *ret = malloc(strlen(base) + strlen(name) + 2); + ret[0] = '\0'; + strcat(ret, base); + strcat(ret, "."); + strcat(ret, name); + return ret; + } +} + + +/* upb_deflist ****************************************************************/ + +void upb_deflist_init(upb_deflist *l) { + l->size = 0; + l->defs = NULL; + l->len = 0; + l->owned = true; +} + +void upb_deflist_uninit(upb_deflist *l) { + if (l->owned) + for(size_t i = 0; i < l->len; i++) + upb_def_unref(l->defs[i], l); + free(l->defs); +} + +bool upb_deflist_push(upb_deflist *l, upb_def *d) { + if(++l->len >= l->size) { + size_t new_size = UPB_MAX(l->size, 4); + new_size *= 2; + l->defs = realloc(l->defs, new_size * sizeof(void *)); + if (!l->defs) return false; + l->size = new_size; + } + l->defs[l->len - 1] = d; + return true; +} + +void upb_deflist_donaterefs(upb_deflist *l, void *owner) { + assert(l->owned); + for (size_t i = 0; i < l->len; i++) + upb_def_donateref(l->defs[i], l, owner); + l->owned = false; +} + +static upb_def *upb_deflist_last(upb_deflist *l) { + return l->defs[l->len-1]; +} + +// Qualify the defname for all defs starting with offset "start" with "str". +static void upb_deflist_qualify(upb_deflist *l, char *str, int32_t start) { + for (uint32_t i = start; i < l->len; i++) { + upb_def *def = l->defs[i]; + char *name = upb_join(str, upb_def_fullname(def)); + upb_def_setfullname(def, name, NULL); + free(name); + } +} + + +/* upb_descreader ************************************************************/ + +void upb_descreader_init(upb_descreader *r, const upb_handlers *handlers, + upb_status *status) { + UPB_UNUSED(status); + upb_deflist_init(&r->defs); + upb_sink_reset(upb_descreader_input(r), handlers, r); + r->stack_len = 0; + r->name = NULL; + r->default_string = NULL; +} + +void upb_descreader_uninit(upb_descreader *r) { + free(r->name); + upb_deflist_uninit(&r->defs); + free(r->default_string); + while (r->stack_len > 0) { + upb_descreader_frame *f = &r->stack[--r->stack_len]; + free(f->name); + } +} + +upb_def **upb_descreader_getdefs(upb_descreader *r, void *owner, int *n) { + *n = r->defs.len; + upb_deflist_donaterefs(&r->defs, owner); + return r->defs.defs; +} + +upb_sink *upb_descreader_input(upb_descreader *r) { + return &r->sink; +} + +static upb_msgdef *upb_descreader_top(upb_descreader *r) { + assert(r->stack_len > 1); + int index = r->stack[r->stack_len-1].start - 1; + assert(index >= 0); + return upb_downcast_msgdef_mutable(r->defs.defs[index]); +} + +static upb_def *upb_descreader_last(upb_descreader *r) { + return upb_deflist_last(&r->defs); +} + +// Start/end handlers for FileDescriptorProto and DescriptorProto (the two +// entities that have names and can contain sub-definitions. +void upb_descreader_startcontainer(upb_descreader *r) { + upb_descreader_frame *f = &r->stack[r->stack_len++]; + f->start = r->defs.len; + f->name = NULL; +} + +void upb_descreader_endcontainer(upb_descreader *r) { + upb_descreader_frame *f = &r->stack[--r->stack_len]; + upb_deflist_qualify(&r->defs, f->name, f->start); + free(f->name); + f->name = NULL; +} + +void upb_descreader_setscopename(upb_descreader *r, char *str) { + upb_descreader_frame *f = &r->stack[r->stack_len-1]; + free(f->name); + f->name = str; +} + +// Handlers for google.protobuf.FileDescriptorProto. +static bool file_startmsg(void *r, const void *hd) { + UPB_UNUSED(hd); + upb_descreader_startcontainer(r); + return true; +} + +static bool file_endmsg(void *closure, const void *hd, upb_status *status) { + UPB_UNUSED(hd); + UPB_UNUSED(status); + upb_descreader *r = closure; + upb_descreader_endcontainer(r); + return true; +} + +static size_t file_onpackage(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + // XXX: see comment at the top of the file. + upb_descreader_setscopename(r, upb_strndup(buf, n)); + return n; +} + +// Handlers for google.protobuf.EnumValueDescriptorProto. +static bool enumval_startmsg(void *closure, const void *hd) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + r->saw_number = false; + r->saw_name = false; + return true; +} + +static size_t enumval_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + // XXX: see comment at the top of the file. + free(r->name); + r->name = upb_strndup(buf, n); + r->saw_name = true; + return n; +} + +static bool enumval_onnumber(void *closure, const void *hd, int32_t val) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + r->number = val; + r->saw_number = true; + return true; +} + +static bool enumval_endmsg(void *closure, const void *hd, upb_status *status) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + if(!r->saw_number || !r->saw_name) { + upb_status_seterrmsg(status, "Enum value missing name or number."); + return false; + } + upb_enumdef *e = upb_downcast_enumdef_mutable(upb_descreader_last(r)); + upb_enumdef_addval(e, r->name, r->number, status); + free(r->name); + r->name = NULL; + return true; +} + + +// Handlers for google.protobuf.EnumDescriptorProto. +static bool enum_startmsg(void *closure, const void *hd) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_deflist_push(&r->defs, UPB_UPCAST(upb_enumdef_new(&r->defs))); + return true; +} + +static bool enum_endmsg(void *closure, const void *hd, upb_status *status) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_enumdef *e = upb_downcast_enumdef_mutable(upb_descreader_last(r)); + if (upb_def_fullname(upb_descreader_last(r)) == NULL) { + upb_status_seterrmsg(status, "Enum had no name."); + return false; + } + if (upb_enumdef_numvals(e) == 0) { + upb_status_seterrmsg(status, "Enum had no values."); + return false; + } + return true; +} + +static size_t enum_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + // XXX: see comment at the top of the file. + char *fullname = upb_strndup(buf, n); + upb_def_setfullname(upb_descreader_last(r), fullname, NULL); + free(fullname); + return n; +} + +// Handlers for google.protobuf.FieldDescriptorProto +static bool field_startmsg(void *closure, const void *hd) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + r->f = upb_fielddef_new(&r->defs); + free(r->default_string); + r->default_string = NULL; + + // fielddefs default to packed, but descriptors default to non-packed. + upb_fielddef_setpacked(r->f, false); + return true; +} + +// Converts the default value in string "str" into "d". Passes a ref on str. +// Returns true on success. +static bool parse_default(char *str, upb_fielddef *f) { + bool success = true; + char *end; + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: { + long val = strtol(str, &end, 0); + if (val > INT32_MAX || val < INT32_MIN || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultint32(f, val); + break; + } + case UPB_TYPE_INT64: { + long long val = strtoll(str, &end, 0); + if (val > INT64_MAX || val < INT64_MIN || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultint64(f, val); + break; + } + case UPB_TYPE_UINT32: { + long val = strtoul(str, &end, 0); + if (val > UINT32_MAX || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultuint32(f, val); + break; + } + case UPB_TYPE_UINT64: { + unsigned long long val = strtoull(str, &end, 0); + if (val > UINT64_MAX || errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultuint64(f, val); + break; + } + case UPB_TYPE_DOUBLE: { + double val = strtod(str, &end); + if (errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultdouble(f, val); + break; + } + case UPB_TYPE_FLOAT: { + float val = strtof(str, &end); + if (errno == ERANGE || *end) + success = false; + else + upb_fielddef_setdefaultfloat(f, val); + break; + } + case UPB_TYPE_BOOL: { + if (strcmp(str, "false") == 0) + upb_fielddef_setdefaultbool(f, false); + else if (strcmp(str, "true") == 0) + upb_fielddef_setdefaultbool(f, true); + else + success = false; + break; + } + default: abort(); + } + return success; +} + +static bool field_endmsg(void *closure, const void *hd, upb_status *status) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_fielddef *f = r->f; + // TODO: verify that all required fields were present. + assert(upb_fielddef_number(f) != 0); + assert(upb_fielddef_name(f) != NULL); + assert((upb_fielddef_subdefname(f) != NULL) == upb_fielddef_hassubdef(f)); + + if (r->default_string) { + if (upb_fielddef_issubmsg(f)) { + upb_status_seterrmsg(status, "Submessages cannot have defaults."); + return false; + } + if (upb_fielddef_isstring(f) || upb_fielddef_type(f) == UPB_TYPE_ENUM) { + upb_fielddef_setdefaultcstr(f, r->default_string, NULL); + } else { + if (r->default_string && !parse_default(r->default_string, f)) { + // We don't worry too much about giving a great error message since the + // compiler should have ensured this was correct. + upb_status_seterrmsg(status, "Error converting default value."); + return false; + } + } + } + return true; +} + +static bool field_onlazy(void *closure, const void *hd, bool val) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_fielddef_setlazy(r->f, val); + return true; +} + +static bool field_onpacked(void *closure, const void *hd, bool val) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_fielddef_setpacked(r->f, val); + return true; +} + +static bool field_ontype(void *closure, const void *hd, int32_t val) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_fielddef_setdescriptortype(r->f, val); + return true; +} + +static bool field_onlabel(void *closure, const void *hd, int32_t val) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_fielddef_setlabel(r->f, val); + return true; +} + +static bool field_onnumber(void *closure, const void *hd, int32_t val) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + bool ok = upb_fielddef_setnumber(r->f, val, NULL); + UPB_ASSERT_VAR(ok, ok); + return true; +} + +static size_t field_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + // XXX: see comment at the top of the file. + char *name = upb_strndup(buf, n); + upb_fielddef_setname(r->f, name, NULL); + free(name); + return n; +} + +static size_t field_ontypename(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + // XXX: see comment at the top of the file. + char *name = upb_strndup(buf, n); + upb_fielddef_setsubdefname(r->f, name, NULL); + free(name); + return n; +} + +static size_t field_onextendee(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + // XXX: see comment at the top of the file. + char *name = upb_strndup(buf, n); + upb_fielddef_setcontainingtypename(r->f, name, NULL); + free(name); + return n; +} + +static size_t field_ondefaultval(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + // Have to convert from string to the correct type, but we might not know the + // type yet, so we save it as a string until the end of the field. + // XXX: see comment at the top of the file. + free(r->default_string); + r->default_string = upb_strndup(buf, n); + return n; +} + +// Handlers for google.protobuf.DescriptorProto (representing a message). +static bool msg_startmsg(void *closure, const void *hd) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_deflist_push(&r->defs, UPB_UPCAST(upb_msgdef_new(&r->defs))); + upb_descreader_startcontainer(r); + return true; +} + +static bool msg_endmsg(void *closure, const void *hd, upb_status *status) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_msgdef *m = upb_descreader_top(r); + if(!upb_def_fullname(UPB_UPCAST(m))) { + upb_status_seterrmsg(status, "Encountered message with no name."); + return false; + } + upb_descreader_endcontainer(r); + return true; +} + +static size_t msg_onname(void *closure, const void *hd, const char *buf, + size_t n, const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_descreader *r = closure; + upb_msgdef *m = upb_descreader_top(r); + // XXX: see comment at the top of the file. + char *name = upb_strndup(buf, n); + upb_def_setfullname(UPB_UPCAST(m), name, NULL); + upb_descreader_setscopename(r, name); // Passes ownership of name. + return n; +} + +static bool msg_onendfield(void *closure, const void *hd) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + upb_msgdef *m = upb_descreader_top(r); + upb_msgdef_addfield(m, r->f, &r->defs, NULL); + r->f = NULL; + return true; +} + +static bool pushextension(void *closure, const void *hd) { + UPB_UNUSED(hd); + upb_descreader *r = closure; + assert(upb_fielddef_containingtypename(r->f)); + upb_fielddef_setisextension(r->f, true); + upb_deflist_push(&r->defs, UPB_UPCAST(r->f)); + r->f = NULL; + return true; +} + +#define D(name) upbdefs_google_protobuf_ ## name(s) + +static void reghandlers(const void *closure, upb_handlers *h) { + const upb_symtab *s = closure; + const upb_msgdef *m = upb_handlers_msgdef(h); + + if (m == D(DescriptorProto)) { + upb_handlers_setstartmsg(h, &msg_startmsg, NULL); + upb_handlers_setendmsg(h, &msg_endmsg, NULL); + upb_handlers_setstring(h, D(DescriptorProto_name), &msg_onname, NULL); + upb_handlers_setendsubmsg(h, D(DescriptorProto_field), &msg_onendfield, + NULL); + upb_handlers_setendsubmsg(h, D(DescriptorProto_extension), &pushextension, + NULL); + } else if (m == D(FileDescriptorProto)) { + upb_handlers_setstartmsg(h, &file_startmsg, NULL); + upb_handlers_setendmsg(h, &file_endmsg, NULL); + upb_handlers_setstring(h, D(FileDescriptorProto_package), &file_onpackage, + NULL); + upb_handlers_setendsubmsg(h, D(FileDescriptorProto_extension), &pushextension, + NULL); + } else if (m == D(EnumValueDescriptorProto)) { + upb_handlers_setstartmsg(h, &enumval_startmsg, NULL); + upb_handlers_setendmsg(h, &enumval_endmsg, NULL); + upb_handlers_setstring(h, D(EnumValueDescriptorProto_name), &enumval_onname, NULL); + upb_handlers_setint32(h, D(EnumValueDescriptorProto_number), &enumval_onnumber, + NULL); + } else if (m == D(EnumDescriptorProto)) { + upb_handlers_setstartmsg(h, &enum_startmsg, NULL); + upb_handlers_setendmsg(h, &enum_endmsg, NULL); + upb_handlers_setstring(h, D(EnumDescriptorProto_name), &enum_onname, NULL); + } else if (m == D(FieldDescriptorProto)) { + upb_handlers_setstartmsg(h, &field_startmsg, NULL); + upb_handlers_setendmsg(h, &field_endmsg, NULL); + upb_handlers_setint32(h, D(FieldDescriptorProto_type), &field_ontype, + NULL); + upb_handlers_setint32(h, D(FieldDescriptorProto_label), &field_onlabel, + NULL); + upb_handlers_setint32(h, D(FieldDescriptorProto_number), &field_onnumber, + NULL); + upb_handlers_setstring(h, D(FieldDescriptorProto_name), &field_onname, + NULL); + upb_handlers_setstring(h, D(FieldDescriptorProto_type_name), + &field_ontypename, NULL); + upb_handlers_setstring(h, D(FieldDescriptorProto_extendee), + &field_onextendee, NULL); + upb_handlers_setstring(h, D(FieldDescriptorProto_default_value), + &field_ondefaultval, NULL); + } else if (m == D(FieldOptions)) { + upb_handlers_setbool(h, D(FieldOptions_lazy), &field_onlazy, NULL); + upb_handlers_setbool(h, D(FieldOptions_packed), &field_onpacked, NULL); + } +} + +#undef D + +const upb_handlers *upb_descreader_newhandlers(const void *owner) { + const upb_symtab *s = upbdefs_google_protobuf_descriptor(&s); + const upb_handlers *h = upb_handlers_newfrozen( + upbdefs_google_protobuf_FileDescriptorSet(s), owner, reghandlers, s); + upb_symtab_unref(s, &s); + return h; +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2013 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Code to compile a upb::Handlers into bytecode for decoding a protobuf + * according to that specific schema and destination handlers. + * + * Compiling to bytecode is always the first step. If we are using the + * interpreted decoder we leave it as bytecode and interpret that. If we are + * using a JIT decoder we use a code generator to turn the bytecode into native + * code, LLVM IR, etc. + * + * Bytecode definition is in decoder.int.h. + */ + +#include + +#ifdef UPB_DUMP_BYTECODE +#include +#endif + +#define MAXLABEL 5 +#define EMPTYLABEL -1 + +/* mgroup *********************************************************************/ + +static void freegroup(upb_refcounted *r) { + mgroup *g = (mgroup*)r; + upb_inttable_uninit(&g->methods); +#ifdef UPB_USE_JIT_X64 + upb_pbdecoder_freejit(g); +#endif + free(g->bytecode); + free(g); +} + +static void visitgroup(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const mgroup *g = (const mgroup*)r; + upb_inttable_iter i; + upb_inttable_begin(&i, &g->methods); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_pbdecodermethod *method = upb_value_getptr(upb_inttable_iter_value(&i)); + visit(r, UPB_UPCAST(method), closure); + } +} + +mgroup *newgroup(const void *owner) { + mgroup *g = malloc(sizeof(*g)); + static const struct upb_refcounted_vtbl vtbl = {visitgroup, freegroup}; + upb_refcounted_init(UPB_UPCAST(g), &vtbl, owner); + upb_inttable_init(&g->methods, UPB_CTYPE_PTR); + g->bytecode = NULL; + g->bytecode_end = NULL; + return g; +} + + +/* upb_pbdecodermethod ********************************************************/ + +static void freemethod(upb_refcounted *r) { + upb_pbdecodermethod *method = (upb_pbdecodermethod*)r; + upb_byteshandler_uninit(&method->input_handler_); + + if (method->dest_handlers_) { + upb_handlers_unref(method->dest_handlers_, method); + } + + upb_inttable_uninit(&method->dispatch); + free(method); +} + +static void visitmethod(const upb_refcounted *r, upb_refcounted_visit *visit, + void *closure) { + const upb_pbdecodermethod *m = (const upb_pbdecodermethod*)r; + visit(r, m->group, closure); +} + +static upb_pbdecodermethod *newmethod(const upb_handlers *dest_handlers, + mgroup *group) { + static const struct upb_refcounted_vtbl vtbl = {visitmethod, freemethod}; + upb_pbdecodermethod *ret = malloc(sizeof(*ret)); + upb_refcounted_init(UPB_UPCAST(ret), &vtbl, &ret); + upb_byteshandler_init(&ret->input_handler_); + + // The method references the group and vice-versa, in a circular reference. + upb_ref2(ret, group); + upb_ref2(group, ret); + upb_inttable_insertptr(&group->methods, dest_handlers, upb_value_ptr(ret)); + upb_refcounted_unref(UPB_UPCAST(ret), &ret); + + ret->group = UPB_UPCAST(group); + ret->dest_handlers_ = dest_handlers; + ret->is_native_ = false; // If we JIT, it will update this later. + upb_inttable_init(&ret->dispatch, UPB_CTYPE_UINT64); + + if (ret->dest_handlers_) { + upb_handlers_ref(ret->dest_handlers_, ret); + } + return ret; +} + +void upb_pbdecodermethod_ref(const upb_pbdecodermethod *m, const void *owner) { + upb_refcounted_ref(UPB_UPCAST(m), owner); +} + +void upb_pbdecodermethod_unref(const upb_pbdecodermethod *m, + const void *owner) { + upb_refcounted_unref(UPB_UPCAST(m), owner); +} + +void upb_pbdecodermethod_donateref(const upb_pbdecodermethod *m, + const void *from, const void *to) { + upb_refcounted_donateref(UPB_UPCAST(m), from, to); +} + +void upb_pbdecodermethod_checkref(const upb_pbdecodermethod *m, + const void *owner) { + upb_refcounted_checkref(UPB_UPCAST(m), owner); +} + +const upb_handlers *upb_pbdecodermethod_desthandlers( + const upb_pbdecodermethod *m) { + return m->dest_handlers_; +} + +const upb_byteshandler *upb_pbdecodermethod_inputhandler( + const upb_pbdecodermethod *m) { + return &m->input_handler_; +} + +bool upb_pbdecodermethod_isnative(const upb_pbdecodermethod *m) { + return m->is_native_; +} + +const upb_pbdecodermethod *upb_pbdecodermethod_new( + const upb_pbdecodermethodopts *opts, const void *owner) { + upb_pbcodecache cache; + upb_pbcodecache_init(&cache); + const upb_pbdecodermethod *ret = + upb_pbcodecache_getdecodermethod(&cache, opts); + upb_pbdecodermethod_ref(ret, owner); + upb_pbcodecache_uninit(&cache); + return ret; +} + + +/* bytecode compiler **********************************************************/ + +// Data used only at compilation time. +typedef struct { + mgroup *group; + + uint32_t *pc; + int fwd_labels[MAXLABEL]; + int back_labels[MAXLABEL]; + + // For fields marked "lazy", parse them lazily or eagerly? + bool lazy; +} compiler; + +static compiler *newcompiler(mgroup *group, bool lazy) { + compiler *ret = malloc(sizeof(*ret)); + ret->group = group; + ret->lazy = lazy; + for (int i = 0; i < MAXLABEL; i++) { + ret->fwd_labels[i] = EMPTYLABEL; + ret->back_labels[i] = EMPTYLABEL; + } + return ret; +} + +static void freecompiler(compiler *c) { + free(c); +} + +const size_t ptr_words = sizeof(void*) / sizeof(uint32_t); + +// How many words an instruction is. +static int instruction_len(uint32_t instr) { + switch (getop(instr)) { + case OP_SETDISPATCH: return 1 + ptr_words; + case OP_TAGN: return 3; + case OP_SETBIGGROUPNUM: return 2; + default: return 1; + } +} + +bool op_has_longofs(int32_t instruction) { + switch (getop(instruction)) { + case OP_CALL: + case OP_BRANCH: + case OP_CHECKDELIM: + return true; + // The "tag" instructions only have 8 bytes available for the jump target, + // but that is ok because these opcodes only require short jumps. + case OP_TAG1: + case OP_TAG2: + case OP_TAGN: + return false; + default: + assert(false); + return false; + } +} + +static int32_t getofs(uint32_t instruction) { + if (op_has_longofs(instruction)) { + return (int32_t)instruction >> 8; + } else { + return (int8_t)(instruction >> 8); + } +} + +static void setofs(uint32_t *instruction, int32_t ofs) { + if (op_has_longofs(*instruction)) { + *instruction = getop(*instruction) | ofs << 8; + } else { + *instruction = (*instruction & ~0xff00) | ((ofs & 0xff) << 8); + } + assert(getofs(*instruction) == ofs); // Would fail in cases of overflow. +} + +static uint32_t pcofs(compiler *c) { return c->pc - c->group->bytecode; } + +// Defines a local label at the current PC location. All previous forward +// references are updated to point to this location. The location is noted +// for any future backward references. +static void label(compiler *c, unsigned int label) { + assert(label < MAXLABEL); + int val = c->fwd_labels[label]; + uint32_t *codep = (val == EMPTYLABEL) ? NULL : c->group->bytecode + val; + while (codep) { + int ofs = getofs(*codep); + setofs(codep, c->pc - codep - instruction_len(*codep)); + codep = ofs ? codep + ofs : NULL; + } + c->fwd_labels[label] = EMPTYLABEL; + c->back_labels[label] = pcofs(c); +} + +// Creates a reference to a numbered label; either a forward reference +// (positive arg) or backward reference (negative arg). For forward references +// the value returned now is actually a "next" pointer into a linked list of all +// instructions that use this label and will be patched later when the label is +// defined with label(). +// +// The returned value is the offset that should be written into the instruction. +static int32_t labelref(compiler *c, int label) { + assert(label < MAXLABEL); + if (label == LABEL_DISPATCH) { + // No resolving required. + return 0; + } else if (label < 0) { + // Backward local label. Relative to the next instruction. + uint32_t from = (c->pc + 1) - c->group->bytecode; + return c->back_labels[-label] - from; + } else { + // Forward local label: prepend to (possibly-empty) linked list. + int *lptr = &c->fwd_labels[label]; + int32_t ret = (*lptr == EMPTYLABEL) ? 0 : *lptr - pcofs(c); + *lptr = pcofs(c); + return ret; + } +} + +static void put32(compiler *c, uint32_t v) { + mgroup *g = c->group; + if (c->pc == g->bytecode_end) { + int ofs = pcofs(c); + size_t oldsize = g->bytecode_end - g->bytecode; + size_t newsize = UPB_MAX(oldsize * 2, 64); + // TODO(haberman): handle OOM. + g->bytecode = realloc(g->bytecode, newsize * sizeof(uint32_t)); + g->bytecode_end = g->bytecode + newsize; + c->pc = g->bytecode + ofs; + } + *c->pc++ = v; +} + +static void putop(compiler *c, opcode op, ...) { + va_list ap; + va_start(ap, op); + + switch (op) { + case OP_SETDISPATCH: { + uintptr_t ptr = (uintptr_t)va_arg(ap, void*); + put32(c, OP_SETDISPATCH); + put32(c, ptr); + if (sizeof(uintptr_t) > sizeof(uint32_t)) + put32(c, (uint64_t)ptr >> 32); + break; + } + case OP_STARTMSG: + case OP_ENDMSG: + case OP_PUSHLENDELIM: + case OP_POP: + case OP_SETDELIM: + case OP_HALT: + case OP_RET: + case OP_DISPATCH: + put32(c, op); + break; + case OP_PARSE_DOUBLE: + case OP_PARSE_FLOAT: + case OP_PARSE_INT64: + case OP_PARSE_UINT64: + case OP_PARSE_INT32: + case OP_PARSE_FIXED64: + case OP_PARSE_FIXED32: + case OP_PARSE_BOOL: + case OP_PARSE_UINT32: + case OP_PARSE_SFIXED32: + case OP_PARSE_SFIXED64: + case OP_PARSE_SINT32: + case OP_PARSE_SINT64: + case OP_STARTSEQ: + case OP_ENDSEQ: + case OP_STARTSUBMSG: + case OP_ENDSUBMSG: + case OP_STARTSTR: + case OP_STRING: + case OP_ENDSTR: + case OP_PUSHTAGDELIM: + put32(c, op | va_arg(ap, upb_selector_t) << 8); + break; + case OP_SETBIGGROUPNUM: + put32(c, op); + put32(c, va_arg(ap, int)); + break; + case OP_CALL: { + const upb_pbdecodermethod *method = va_arg(ap, upb_pbdecodermethod *); + put32(c, op | (method->code_base.ofs - (pcofs(c) + 1)) << 8); + break; + } + case OP_CHECKDELIM: + case OP_BRANCH: { + uint32_t instruction = op; + int label = va_arg(ap, int); + setofs(&instruction, labelref(c, label)); + put32(c, instruction); + break; + } + case OP_TAG1: + case OP_TAG2: { + int label = va_arg(ap, int); + uint64_t tag = va_arg(ap, uint64_t); + uint32_t instruction = op | (tag << 16); + assert(tag <= 0xffff); + setofs(&instruction, labelref(c, label)); + put32(c, instruction); + break; + } + case OP_TAGN: { + int label = va_arg(ap, int); + uint64_t tag = va_arg(ap, uint64_t); + uint32_t instruction = op | (upb_value_size(tag) << 16); + setofs(&instruction, labelref(c, label)); + put32(c, instruction); + put32(c, tag); + put32(c, tag >> 32); + break; + } + } + + va_end(ap); +} + +#if defined(UPB_USE_JIT_X64) || defined(UPB_DUMP_BYTECODE) + +const char *upb_pbdecoder_getopname(unsigned int op) { +#define OP(op) [OP_ ## op] = "OP_" #op +#define T(op) OP(PARSE_##op) + static const char *names[] = { + "", + T(DOUBLE), T(FLOAT), T(INT64), T(UINT64), T(INT32), T(FIXED64), T(FIXED32), + T(BOOL), T(UINT32), T(SFIXED32), T(SFIXED64), T(SINT32), T(SINT64), + OP(STARTMSG), OP(ENDMSG), OP(STARTSEQ), OP(ENDSEQ), OP(STARTSUBMSG), + OP(ENDSUBMSG), OP(STARTSTR), OP(STRING), OP(ENDSTR), OP(CALL), OP(RET), + OP(PUSHLENDELIM), OP(PUSHTAGDELIM), OP(SETDELIM), OP(CHECKDELIM), + OP(BRANCH), OP(TAG1), OP(TAG2), OP(TAGN), OP(SETDISPATCH), OP(POP), + OP(SETBIGGROUPNUM), OP(DISPATCH), OP(HALT), + }; + return op > OP_HALT ? names[0] : names[op]; +#undef OP +#undef T +} + +#endif + +#ifdef UPB_DUMP_BYTECODE + +static void dumpbc(uint32_t *p, uint32_t *end, FILE *f) { + + uint32_t *begin = p; + + while (p < end) { + fprintf(f, "%p %8tx", p, p - begin); + uint32_t instr = *p++; + uint8_t op = getop(instr); + fprintf(f, " %s", upb_pbdecoder_getopname(op)); + switch ((opcode)op) { + case OP_SETDISPATCH: { + const upb_inttable *dispatch; + memcpy(&dispatch, p, sizeof(void*)); + p += ptr_words; + const upb_pbdecodermethod *method = + (void *)((char *)dispatch - + offsetof(upb_pbdecodermethod, dispatch)); + fprintf(f, " %s", upb_msgdef_fullname( + upb_handlers_msgdef(method->dest_handlers_))); + break; + } + case OP_DISPATCH: + case OP_STARTMSG: + case OP_ENDMSG: + case OP_PUSHLENDELIM: + case OP_POP: + case OP_SETDELIM: + case OP_HALT: + case OP_RET: + break; + case OP_PARSE_DOUBLE: + case OP_PARSE_FLOAT: + case OP_PARSE_INT64: + case OP_PARSE_UINT64: + case OP_PARSE_INT32: + case OP_PARSE_FIXED64: + case OP_PARSE_FIXED32: + case OP_PARSE_BOOL: + case OP_PARSE_UINT32: + case OP_PARSE_SFIXED32: + case OP_PARSE_SFIXED64: + case OP_PARSE_SINT32: + case OP_PARSE_SINT64: + case OP_STARTSEQ: + case OP_ENDSEQ: + case OP_STARTSUBMSG: + case OP_ENDSUBMSG: + case OP_STARTSTR: + case OP_STRING: + case OP_ENDSTR: + case OP_PUSHTAGDELIM: + fprintf(f, " %d", instr >> 8); + break; + case OP_SETBIGGROUPNUM: + fprintf(f, " %d", *p++); + break; + case OP_CHECKDELIM: + case OP_CALL: + case OP_BRANCH: + fprintf(f, " =>0x%tx", p + getofs(instr) - begin); + break; + case OP_TAG1: + case OP_TAG2: { + fprintf(f, " tag:0x%x", instr >> 16); + if (getofs(instr)) { + fprintf(f, " =>0x%tx", p + getofs(instr) - begin); + } + break; + } + case OP_TAGN: { + uint64_t tag = *p++; + tag |= (uint64_t)*p++ << 32; + fprintf(f, " tag:0x%llx", (long long)tag); + fprintf(f, " n:%d", instr >> 16); + if (getofs(instr)) { + fprintf(f, " =>0x%tx", p + getofs(instr) - begin); + } + break; + } + } + fputs("\n", f); + } +} + +#endif + +static uint64_t get_encoded_tag(const upb_fielddef *f, int wire_type) { + uint32_t tag = (upb_fielddef_number(f) << 3) | wire_type; + uint64_t encoded_tag = upb_vencode32(tag); + // No tag should be greater than 5 bytes. + assert(encoded_tag <= 0xffffffffff); + return encoded_tag; +} + +static void putchecktag(compiler *c, const upb_fielddef *f, + int wire_type, int dest) { + uint64_t tag = get_encoded_tag(f, wire_type); + switch (upb_value_size(tag)) { + case 1: + putop(c, OP_TAG1, dest, tag); + break; + case 2: + putop(c, OP_TAG2, dest, tag); + break; + default: + putop(c, OP_TAGN, dest, tag); + break; + } +} + +static upb_selector_t getsel(const upb_fielddef *f, upb_handlertype_t type) { + upb_selector_t selector; + bool ok = upb_handlers_getselector(f, type, &selector); + UPB_ASSERT_VAR(ok, ok); + return selector; +} + +// Takes an existing, primary dispatch table entry and repacks it with a +// different alternate wire type. Called when we are inserting a secondary +// dispatch table entry for an alternate wire type. +static uint64_t repack(uint64_t dispatch, int new_wt2) { + uint64_t ofs; + uint8_t wt1; + uint8_t old_wt2; + upb_pbdecoder_unpackdispatch(dispatch, &ofs, &wt1, &old_wt2); + assert(old_wt2 == NO_WIRE_TYPE); // wt2 should not be set yet. + return upb_pbdecoder_packdispatch(ofs, wt1, new_wt2); +} + +// Marks the current bytecode position as the dispatch target for this message, +// field, and wire type. +static void dispatchtarget(compiler *c, upb_pbdecodermethod *method, + const upb_fielddef *f, int wire_type) { + // Offset is relative to msg base. + uint64_t ofs = pcofs(c) - method->code_base.ofs; + uint32_t fn = upb_fielddef_number(f); + upb_inttable *d = &method->dispatch; + upb_value v; + if (upb_inttable_remove(d, fn, &v)) { + // TODO: prioritize based on packed setting in .proto file. + uint64_t repacked = repack(upb_value_getuint64(v), wire_type); + upb_inttable_insert(d, fn, upb_value_uint64(repacked)); + upb_inttable_insert(d, fn + UPB_MAX_FIELDNUMBER, upb_value_uint64(ofs)); + } else { + uint64_t val = upb_pbdecoder_packdispatch(ofs, wire_type, NO_WIRE_TYPE); + upb_inttable_insert(d, fn, upb_value_uint64(val)); + } +} + +static void putpush(compiler *c, const upb_fielddef *f) { + if (upb_fielddef_descriptortype(f) == UPB_DESCRIPTOR_TYPE_MESSAGE) { + putop(c, OP_PUSHLENDELIM); + } else { + uint32_t fn = upb_fielddef_number(f); + if (fn >= 1 << 24) { + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_SETBIGGROUPNUM, fn); + } else { + putop(c, OP_PUSHTAGDELIM, fn); + } + } +} + +static upb_pbdecodermethod *find_submethod(const compiler *c, + const upb_pbdecodermethod *method, + const upb_fielddef *f) { + const upb_handlers *sub = + upb_handlers_getsubhandlers(method->dest_handlers_, f); + upb_value v; + return upb_inttable_lookupptr(&c->group->methods, sub, &v) + ? upb_value_getptr(v) + : NULL; +} + +static void putsel(compiler *c, opcode op, upb_selector_t sel, + const upb_handlers *h) { + if (upb_handlers_gethandler(h, sel)) { + putop(c, op, sel); + } +} + +// Puts an opcode to call a callback, but only if a callback actually exists for +// this field and handler type. +static void maybeput(compiler *c, opcode op, const upb_handlers *h, + const upb_fielddef *f, upb_handlertype_t type) { + putsel(c, op, getsel(f, type), h); +} + +static bool haslazyhandlers(const upb_handlers *h, const upb_fielddef *f) { + if (!upb_fielddef_lazy(f)) + return false; + + return upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_STARTSTR)) || + upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_STRING)) || + upb_handlers_gethandler(h, getsel(f, UPB_HANDLER_ENDSTR)); +} + + +/* bytecode compiler code generation ******************************************/ + +// Symbolic names for our local labels. +#define LABEL_LOOPSTART 1 // Top of a repeated field loop. +#define LABEL_LOOPBREAK 2 // To jump out of a repeated loop +#define LABEL_FIELD 3 // Jump backward to find the most recent field. +#define LABEL_ENDMSG 4 // To reach the OP_ENDMSG instr for this msg. + +// Generates bytecode to parse a single non-lazy message field. +static void generate_msgfield(compiler *c, const upb_fielddef *f, + upb_pbdecodermethod *method) { + const upb_handlers *h = upb_pbdecodermethod_desthandlers(method); + const upb_pbdecodermethod *sub_m = find_submethod(c, method, f); + + if (!sub_m) { + // Don't emit any code for this field at all; it will be parsed as an + // unknown field. + return; + } + + label(c, LABEL_FIELD); + + int wire_type = + (upb_fielddef_descriptortype(f) == UPB_DESCRIPTOR_TYPE_MESSAGE) + ? UPB_WIRE_TYPE_DELIMITED + : UPB_WIRE_TYPE_START_GROUP; + + if (upb_fielddef_isseq(f)) { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, wire_type, LABEL_DISPATCH); + dispatchtarget(c, method, f, wire_type); + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); + label(c, LABEL_LOOPSTART); + putpush(c, f); + putop(c, OP_STARTSUBMSG, getsel(f, UPB_HANDLER_STARTSUBMSG)); + putop(c, OP_CALL, sub_m); + putop(c, OP_POP); + maybeput(c, OP_ENDSUBMSG, h, f, UPB_HANDLER_ENDSUBMSG); + if (wire_type == UPB_WIRE_TYPE_DELIMITED) { + putop(c, OP_SETDELIM); + } + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putchecktag(c, f, wire_type, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + label(c, LABEL_LOOPBREAK); + putop(c, OP_POP); + maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ); + } else { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, wire_type, LABEL_DISPATCH); + dispatchtarget(c, method, f, wire_type); + putpush(c, f); + putop(c, OP_STARTSUBMSG, getsel(f, UPB_HANDLER_STARTSUBMSG)); + putop(c, OP_CALL, sub_m); + putop(c, OP_POP); + maybeput(c, OP_ENDSUBMSG, h, f, UPB_HANDLER_ENDSUBMSG); + if (wire_type == UPB_WIRE_TYPE_DELIMITED) { + putop(c, OP_SETDELIM); + } + } +} + +// Generates bytecode to parse a single string or lazy submessage field. +static void generate_delimfield(compiler *c, const upb_fielddef *f, + upb_pbdecodermethod *method) { + const upb_handlers *h = upb_pbdecodermethod_desthandlers(method); + + label(c, LABEL_FIELD); + if (upb_fielddef_isseq(f)) { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH); + dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED); + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); + label(c, LABEL_LOOPSTART); + putop(c, OP_PUSHLENDELIM); + putop(c, OP_STARTSTR, getsel(f, UPB_HANDLER_STARTSTR)); + // Need to emit even if no handler to skip past the string. + putop(c, OP_STRING, getsel(f, UPB_HANDLER_STRING)); + putop(c, OP_POP); + maybeput(c, OP_ENDSTR, h, f, UPB_HANDLER_ENDSTR); + putop(c, OP_SETDELIM); + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + label(c, LABEL_LOOPBREAK); + putop(c, OP_POP); + maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ); + } else { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH); + dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED); + putop(c, OP_PUSHLENDELIM); + putop(c, OP_STARTSTR, getsel(f, UPB_HANDLER_STARTSTR)); + putop(c, OP_STRING, getsel(f, UPB_HANDLER_STRING)); + putop(c, OP_POP); + maybeput(c, OP_ENDSTR, h, f, UPB_HANDLER_ENDSTR); + putop(c, OP_SETDELIM); + } +} + +// Generates bytecode to parse a single primitive field. +static void generate_primitivefield(compiler *c, const upb_fielddef *f, + upb_pbdecodermethod *method) { + label(c, LABEL_FIELD); + + const upb_handlers *h = upb_pbdecodermethod_desthandlers(method); + upb_descriptortype_t descriptor_type = upb_fielddef_descriptortype(f); + + // From a decoding perspective, ENUM is the same as INT32. + if (descriptor_type == UPB_DESCRIPTOR_TYPE_ENUM) + descriptor_type = UPB_DESCRIPTOR_TYPE_INT32; + + opcode parse_type = (opcode)descriptor_type; + + // TODO(haberman): generate packed or non-packed first depending on "packed" + // setting in the fielddef. This will favor (in speed) whichever was + // specified. + + assert((int)parse_type >= 0 && parse_type <= OP_MAX); + upb_selector_t sel = getsel(f, upb_handlers_getprimitivehandlertype(f)); + int wire_type = upb_pb_native_wire_types[upb_fielddef_descriptortype(f)]; + if (upb_fielddef_isseq(f)) { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, UPB_WIRE_TYPE_DELIMITED, LABEL_DISPATCH); + dispatchtarget(c, method, f, UPB_WIRE_TYPE_DELIMITED); + putop(c, OP_PUSHLENDELIM); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); // Packed + label(c, LABEL_LOOPSTART); + putop(c, parse_type, sel); + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + dispatchtarget(c, method, f, wire_type); + putop(c, OP_PUSHTAGDELIM, 0); + putop(c, OP_STARTSEQ, getsel(f, UPB_HANDLER_STARTSEQ)); // Non-packed + label(c, LABEL_LOOPSTART); + putop(c, parse_type, sel); + putop(c, OP_CHECKDELIM, LABEL_LOOPBREAK); + putchecktag(c, f, wire_type, LABEL_LOOPBREAK); + putop(c, OP_BRANCH, -LABEL_LOOPSTART); + label(c, LABEL_LOOPBREAK); + putop(c, OP_POP); // Packed and non-packed join. + maybeput(c, OP_ENDSEQ, h, f, UPB_HANDLER_ENDSEQ); + putop(c, OP_SETDELIM); // Could remove for non-packed by dup ENDSEQ. + } else { + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + putchecktag(c, f, wire_type, LABEL_DISPATCH); + dispatchtarget(c, method, f, wire_type); + putop(c, parse_type, sel); + } +} + +// Adds bytecode for parsing the given message to the given decoderplan, +// while adding all dispatch targets to this message's dispatch table. +static void compile_method(compiler *c, upb_pbdecodermethod *method) { + assert(method); + + // Clear all entries in the dispatch table. + upb_inttable_uninit(&method->dispatch); + upb_inttable_init(&method->dispatch, UPB_CTYPE_UINT64); + + const upb_handlers *h = upb_pbdecodermethod_desthandlers(method); + const upb_msgdef *md = upb_handlers_msgdef(h); + + method->code_base.ofs = pcofs(c); + putop(c, OP_SETDISPATCH, &method->dispatch); + putsel(c, OP_STARTMSG, UPB_STARTMSG_SELECTOR, h); + label(c, LABEL_FIELD); + uint32_t* start_pc = c->pc; + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, md); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + upb_fieldtype_t type = upb_fielddef_type(f); + + if (type == UPB_TYPE_MESSAGE && !(haslazyhandlers(h, f) && c->lazy)) { + generate_msgfield(c, f, method); + } else if (type == UPB_TYPE_STRING || type == UPB_TYPE_BYTES || + type == UPB_TYPE_MESSAGE) { + generate_delimfield(c, f, method); + } else { + generate_primitivefield(c, f, method); + } + } + + // If there were no fields, or if no handlers were defined, we need to + // generate a non-empty loop body so that we can at least dispatch for unknown + // fields and check for the end of the message. + if (c->pc == start_pc) { + // Check for end-of-message. + putop(c, OP_CHECKDELIM, LABEL_ENDMSG); + // Unconditionally dispatch. + putop(c, OP_DISPATCH, 0); + } + + // For now we just loop back to the last field of the message (or if none, + // the DISPATCH opcode for the message). + putop(c, OP_BRANCH, -LABEL_FIELD); + + // Insert both a label and a dispatch table entry for this end-of-msg. + label(c, LABEL_ENDMSG); + upb_value val = upb_value_uint64(pcofs(c) - method->code_base.ofs); + upb_inttable_insert(&method->dispatch, DISPATCH_ENDMSG, val); + + putsel(c, OP_ENDMSG, UPB_ENDMSG_SELECTOR, h); + putop(c, OP_RET); + + upb_inttable_compact(&method->dispatch); +} + +// Populate "methods" with new upb_pbdecodermethod objects reachable from "h". +// Returns the method for these handlers. +// +// Generates a new method for every destination handlers reachable from "h". +static void find_methods(compiler *c, const upb_handlers *h) { + upb_value v; + if (upb_inttable_lookupptr(&c->group->methods, h, &v)) + return; + newmethod(h, c->group); + + // Find submethods. + upb_msg_field_iter i; + const upb_msgdef *md = upb_handlers_msgdef(h); + for(upb_msg_field_begin(&i, md); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + const upb_handlers *sub_h; + if (upb_fielddef_type(f) == UPB_TYPE_MESSAGE && + (sub_h = upb_handlers_getsubhandlers(h, f)) != NULL) { + // We only generate a decoder method for submessages with handlers. + // Others will be parsed as unknown fields. + find_methods(c, sub_h); + } + } +} + +// (Re-)compile bytecode for all messages in "msgs." +// Overwrites any existing bytecode in "c". +static void compile_methods(compiler *c) { + // Start over at the beginning of the bytecode. + c->pc = c->group->bytecode; + + upb_inttable_iter i; + upb_inttable_begin(&i, &c->group->methods); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_pbdecodermethod *method = upb_value_getptr(upb_inttable_iter_value(&i)); + compile_method(c, method); + } +} + +static void set_bytecode_handlers(mgroup *g) { + upb_inttable_iter i; + upb_inttable_begin(&i, &g->methods); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + upb_pbdecodermethod *m = upb_value_getptr(upb_inttable_iter_value(&i)); + + m->code_base.ptr = g->bytecode + m->code_base.ofs; + + upb_byteshandler *h = &m->input_handler_; + upb_byteshandler_setstartstr(h, upb_pbdecoder_startbc, m->code_base.ptr); + upb_byteshandler_setstring(h, upb_pbdecoder_decode, g); + upb_byteshandler_setendstr(h, upb_pbdecoder_end, m); + } +} + + +/* JIT setup. *****************************************************************/ + +#ifdef UPB_USE_JIT_X64 + +static void sethandlers(mgroup *g, bool allowjit) { + g->jit_code = NULL; + if (allowjit) { + // Compile byte-code into machine code, create handlers. + upb_pbdecoder_jit(g); + } else { + set_bytecode_handlers(g); + } +} + +#else // UPB_USE_JIT_X64 + +static void sethandlers(mgroup *g, bool allowjit) { + // No JIT compiled in; use bytecode handlers unconditionally. + UPB_UNUSED(allowjit); + set_bytecode_handlers(g); +} + +#endif // UPB_USE_JIT_X64 + + +// TODO(haberman): allow this to be constructed for an arbitrary set of dest +// handlers and other mgroups (but verify we have a transitive closure). +const mgroup *mgroup_new(const upb_handlers *dest, bool allowjit, bool lazy, + const void *owner) { + UPB_UNUSED(allowjit); + assert(upb_handlers_isfrozen(dest)); + + mgroup *g = newgroup(owner); + compiler *c = newcompiler(g, lazy); + find_methods(c, dest); + + // We compile in two passes: + // 1. all messages are assigned relative offsets from the beginning of the + // bytecode (saved in method->code_base). + // 2. forwards OP_CALL instructions can be correctly linked since message + // offsets have been previously assigned. + // + // Could avoid the second pass by linking OP_CALL instructions somehow. + compile_methods(c); + compile_methods(c); + g->bytecode_end = c->pc; + freecompiler(c); + +#ifdef UPB_DUMP_BYTECODE + FILE *f = fopen("/tmp/upb-bytecode", "wb"); + assert(f); + dumpbc(g->bytecode, g->bytecode_end, stderr); + dumpbc(g->bytecode, g->bytecode_end, f); + fclose(f); +#endif + + sethandlers(g, allowjit); + return g; +} + + +/* upb_pbcodecache ************************************************************/ + +void upb_pbcodecache_init(upb_pbcodecache *c) { + upb_inttable_init(&c->groups, UPB_CTYPE_CONSTPTR); + c->allow_jit_ = true; +} + +void upb_pbcodecache_uninit(upb_pbcodecache *c) { + upb_inttable_iter i; + upb_inttable_begin(&i, &c->groups); + for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { + const mgroup *group = upb_value_getconstptr(upb_inttable_iter_value(&i)); + upb_refcounted_unref(UPB_UPCAST(group), c); + } + upb_inttable_uninit(&c->groups); +} + +bool upb_pbcodecache_allowjit(const upb_pbcodecache *c) { + return c->allow_jit_; +} + +bool upb_pbcodecache_setallowjit(upb_pbcodecache *c, bool allow) { + if (upb_inttable_count(&c->groups) > 0) + return false; + c->allow_jit_ = allow; + return true; +} + +const upb_pbdecodermethod *upb_pbcodecache_getdecodermethod( + upb_pbcodecache *c, const upb_pbdecodermethodopts *opts) { + // Right now we build a new DecoderMethod every time. + // TODO(haberman): properly cache methods by their true key. + const mgroup *g = mgroup_new(opts->handlers, c->allow_jit_, opts->lazy, c); + upb_inttable_push(&c->groups, upb_value_constptr(g)); + + upb_value v; + bool ok = upb_inttable_lookupptr(&g->methods, opts->handlers, &v); + UPB_ASSERT_VAR(ok, ok); + return upb_value_getptr(v); +} + + +/* upb_pbdecodermethodopts ****************************************************/ + +void upb_pbdecodermethodopts_init(upb_pbdecodermethodopts *opts, + const upb_handlers *h) { + opts->handlers = h; + opts->lazy = false; +} + +void upb_pbdecodermethodopts_setlazy(upb_pbdecodermethodopts *opts, bool lazy) { + opts->lazy = lazy; +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2008-2013 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * This file implements a VM for the interpreted (bytecode) decoder. + * + * Bytecode must previously have been generated using the bytecode compiler in + * compile_decoder.c. This decoder then walks through the bytecode op-by-op to + * parse the input. + * + * Decoding is fully resumable; we just keep a pointer to the current bytecode + * instruction and resume from there. A fair amount of the logic here is to + * handle the fact that values can span buffer seams and we have to be able to + * be capable of suspending/resuming from any byte in the stream. This + * sometimes requires keeping a few trailing bytes from the last buffer around + * in the "residual" buffer. + */ + +#include +#include +#include +#include +#include + +#ifdef UPB_DUMP_BYTECODE +#include +#endif + +#define CHECK_SUSPEND(x) if (!(x)) return upb_pbdecoder_suspend(d); + +// Error messages that are shared between the bytecode and JIT decoders. +const char *kPbDecoderStackOverflow = "Nesting too deep."; + +// Error messages shared within this file. +static const char *kUnterminatedVarint = "Unterminated varint."; + +/* upb_pbdecoder **************************************************************/ + +static opcode halt = OP_HALT; + +// Whether an op consumes any of the input buffer. +static bool consumes_input(opcode op) { + switch (op) { + case OP_SETDISPATCH: + case OP_STARTMSG: + case OP_ENDMSG: + case OP_STARTSEQ: + case OP_ENDSEQ: + case OP_STARTSUBMSG: + case OP_ENDSUBMSG: + case OP_STARTSTR: + case OP_ENDSTR: + case OP_PUSHTAGDELIM: + case OP_POP: + case OP_SETDELIM: + case OP_SETBIGGROUPNUM: + case OP_CHECKDELIM: + case OP_CALL: + case OP_RET: + case OP_BRANCH: + return false; + default: + return true; + } +} + +static bool in_residual_buf(const upb_pbdecoder *d, const char *p); + +// It's unfortunate that we have to micro-manage the compiler this way, +// especially since this tuning is necessarily specific to one hardware +// configuration. But emperically on a Core i7, performance increases 30-50% +// with these annotations. Every instance where these appear, gcc 4.2.1 made +// the wrong decision and degraded performance in benchmarks. +#define FORCEINLINE static inline __attribute__((always_inline)) +#define NOINLINE __attribute__((noinline)) + +static void seterr(upb_pbdecoder *d, const char *msg) { + // TODO(haberman): encapsulate this access to pipeline->status, but not sure + // exactly what that interface should look like. + upb_status_seterrmsg(d->status, msg); +} + +void upb_pbdecoder_seterr(upb_pbdecoder *d, const char *msg) { + seterr(d, msg); +} + + +/* Buffering ******************************************************************/ + +// We operate on one buffer at a time, which is either the user's buffer passed +// to our "decode" callback or some residual bytes from the previous buffer. + +// How many bytes can be safely read from d->ptr without reading past end-of-buf +// or past the current delimited end. +static size_t curbufleft(const upb_pbdecoder *d) { + assert(d->data_end >= d->ptr); + return d->data_end - d->ptr; +} + +// Overall stream offset of d->ptr. +uint64_t offset(const upb_pbdecoder *d) { + return d->bufstart_ofs + (d->ptr - d->buf); +} + +// Advances d->ptr. +static void advance(upb_pbdecoder *d, size_t len) { + assert(curbufleft(d) >= len); + d->ptr += len; +} + +static bool in_buf(const char *p, const char *buf, const char *end) { + return p >= buf && p <= end; +} + +static bool in_residual_buf(const upb_pbdecoder *d, const char *p) { + return in_buf(p, d->residual, d->residual_end); +} + +// Calculates the delim_end value, which is affected by both the current buffer +// and the parsing stack, so must be called whenever either is updated. +static void set_delim_end(upb_pbdecoder *d) { + size_t delim_ofs = d->top->end_ofs - d->bufstart_ofs; + if (delim_ofs <= (d->end - d->buf)) { + d->delim_end = d->buf + delim_ofs; + d->data_end = d->delim_end; + } else { + d->data_end = d->end; + d->delim_end = NULL; + } +} + +static void switchtobuf(upb_pbdecoder *d, const char *buf, const char *end) { + d->ptr = buf; + d->buf = buf; + d->end = end; + set_delim_end(d); +} + +static void advancetobuf(upb_pbdecoder *d, const char *buf, size_t len) { + assert(curbufleft(d) == 0); + d->bufstart_ofs += (d->end - d->buf); + switchtobuf(d, buf, buf + len); +} + +static void checkpoint(upb_pbdecoder *d) { + // The assertion here is in the interests of efficiency, not correctness. + // We are trying to ensure that we don't checkpoint() more often than + // necessary. + assert(d->checkpoint != d->ptr); + d->checkpoint = d->ptr; +} + +// Resumes the decoder from an initial state or from a previous suspend. +int32_t upb_pbdecoder_resume(upb_pbdecoder *d, void *p, const char *buf, + size_t size, const upb_bufhandle *handle) { + UPB_UNUSED(p); // Useless; just for the benefit of the JIT. + d->buf_param = buf; + d->size_param = size; + d->handle = handle; + if (d->residual_end > d->residual) { + // We have residual bytes from the last buffer. + assert(d->ptr == d->residual); + } else { + switchtobuf(d, buf, buf + size); + } + d->checkpoint = d->ptr; + if (d->top->groupnum < 0) { + CHECK_RETURN(upb_pbdecoder_skipunknown(d, -1, 0)); + d->checkpoint = d->ptr; + } + return DECODE_OK; +} + +// Suspends the decoder at the last checkpoint, without saving any residual +// bytes. If there are any unconsumed bytes, returns a short byte count. +size_t upb_pbdecoder_suspend(upb_pbdecoder *d) { + d->pc = d->last; + if (d->checkpoint == d->residual) { + // Checkpoint was in residual buf; no user bytes were consumed. + d->ptr = d->residual; + return 0; + } else { + assert(!in_residual_buf(d, d->checkpoint)); + assert(d->buf == d->buf_param); + size_t consumed = d->checkpoint - d->buf; + d->bufstart_ofs += consumed; + d->residual_end = d->residual; + switchtobuf(d, d->residual, d->residual_end); + return consumed; + } +} + +// Suspends the decoder at the last checkpoint, and saves any unconsumed +// bytes in our residual buffer. This is necessary if we need more user +// bytes to form a complete value, which might not be contiguous in the +// user's buffers. Always consumes all user bytes. +static size_t suspend_save(upb_pbdecoder *d) { + // We hit end-of-buffer before we could parse a full value. + // Save any unconsumed bytes (if any) to the residual buffer. + d->pc = d->last; + + if (d->checkpoint == d->residual) { + // Checkpoint was in residual buf; append user byte(s) to residual buf. + assert((d->residual_end - d->residual) + d->size_param <= + sizeof(d->residual)); + if (!in_residual_buf(d, d->ptr)) { + d->bufstart_ofs -= (d->residual_end - d->residual); + } + memcpy(d->residual_end, d->buf_param, d->size_param); + d->residual_end += d->size_param; + } else { + // Checkpoint was in user buf; old residual bytes not needed. + assert(!in_residual_buf(d, d->checkpoint)); + d->ptr = d->checkpoint; + size_t save = curbufleft(d); + assert(save <= sizeof(d->residual)); + memcpy(d->residual, d->ptr, save); + d->residual_end = d->residual + save; + d->bufstart_ofs = offset(d); + } + + switchtobuf(d, d->residual, d->residual_end); + return d->size_param; +} + +// Skips "bytes" bytes in the stream, which may be more than available. If we +// skip more bytes than are available, we return a long read count to the caller +// indicating how many bytes the caller should skip before passing a new buffer. +static int32_t skip(upb_pbdecoder *d, size_t bytes) { + assert(!in_residual_buf(d, d->ptr) || d->size_param == 0); + if (curbufleft(d) >= bytes) { + // Skipped data is all in current buffer. + advance(d, bytes); + return DECODE_OK; + } else { + // Skipped data extends beyond currently available buffers. + d->pc = d->last; + size_t skip = bytes - curbufleft(d); + d->bufstart_ofs += (d->end - d->buf) + skip; + d->residual_end = d->residual; + switchtobuf(d, d->residual, d->residual_end); + return d->size_param + skip; + } +} + +// Copies the next "bytes" bytes into "buf" and advances the stream. +// Requires that this many bytes are available in the current buffer. +FORCEINLINE void consumebytes(upb_pbdecoder *d, void *buf, size_t bytes) { + assert(bytes <= curbufleft(d)); + memcpy(buf, d->ptr, bytes); + advance(d, bytes); +} + +// Slow path for getting the next "bytes" bytes, regardless of whether they are +// available in the current buffer or not. Returns a status code as described +// in decoder.int.h. +static NOINLINE int32_t getbytes_slow(upb_pbdecoder *d, void *buf, + size_t bytes) { + const size_t avail = curbufleft(d); + consumebytes(d, buf, avail); + bytes -= avail; + assert(bytes > 0); + if (in_residual_buf(d, d->ptr)) { + advancetobuf(d, d->buf_param, d->size_param); + } + if (curbufleft(d) >= bytes) { + consumebytes(d, buf + avail, bytes); + return DECODE_OK; + } else if (d->data_end == d->delim_end) { + seterr(d, "Submessage ended in the middle of a value or group"); + return upb_pbdecoder_suspend(d); + } else { + return suspend_save(d); + } +} + +// Gets the next "bytes" bytes, regardless of whether they are available in the +// current buffer or not. Returns a status code as described in decoder.int.h. +FORCEINLINE int32_t getbytes(upb_pbdecoder *d, void *buf, size_t bytes) { + if (curbufleft(d) >= bytes) { + // Buffer has enough data to satisfy. + consumebytes(d, buf, bytes); + return DECODE_OK; + } else { + return getbytes_slow(d, buf, bytes); + } +} + +static NOINLINE size_t peekbytes_slow(upb_pbdecoder *d, void *buf, + size_t bytes) { + size_t ret = curbufleft(d); + memcpy(buf, d->ptr, ret); + if (in_residual_buf(d, d->ptr)) { + size_t copy = UPB_MIN(bytes - ret, d->size_param); + memcpy(buf + ret, d->buf_param, copy); + ret += copy; + } + return ret; +} + +FORCEINLINE size_t peekbytes(upb_pbdecoder *d, void *buf, size_t bytes) { + if (curbufleft(d) >= bytes) { + memcpy(buf, d->ptr, bytes); + return bytes; + } else { + return peekbytes_slow(d, buf, bytes); + } +} + + +/* Decoding of wire types *****************************************************/ + +// Slow path for decoding a varint from the current buffer position. +// Returns a status code as described in decoder.int.h. +NOINLINE int32_t upb_pbdecoder_decode_varint_slow(upb_pbdecoder *d, + uint64_t *u64) { + *u64 = 0; + uint8_t byte = 0x80; + int bitpos; + for(bitpos = 0; bitpos < 70 && (byte & 0x80); bitpos += 7) { + int32_t ret = getbytes(d, &byte, 1); + if (ret >= 0) return ret; + *u64 |= (uint64_t)(byte & 0x7F) << bitpos; + } + if(bitpos == 70 && (byte & 0x80)) { + seterr(d, kUnterminatedVarint); + return upb_pbdecoder_suspend(d); + } + return DECODE_OK; +} + +// Decodes a varint from the current buffer position. +// Returns a status code as described in decoder.int.h. +FORCEINLINE int32_t decode_varint(upb_pbdecoder *d, uint64_t *u64) { + if (curbufleft(d) > 0 && !(*d->ptr & 0x80)) { + *u64 = *d->ptr; + advance(d, 1); + return DECODE_OK; + } else if (curbufleft(d) >= 10) { + // Fast case. + upb_decoderet r = upb_vdecode_fast(d->ptr); + if (r.p == NULL) { + seterr(d, kUnterminatedVarint); + return upb_pbdecoder_suspend(d); + } + advance(d, r.p - d->ptr); + *u64 = r.val; + return DECODE_OK; + } else { + // Slow case -- varint spans buffer seam. + return upb_pbdecoder_decode_varint_slow(d, u64); + } +} + +// Decodes a 32-bit varint from the current buffer position. +// Returns a status code as described in decoder.int.h. +FORCEINLINE int32_t decode_v32(upb_pbdecoder *d, uint32_t *u32) { + uint64_t u64; + int32_t ret = decode_varint(d, &u64); + if (ret >= 0) return ret; + if (u64 > UINT32_MAX) { + seterr(d, "Unterminated 32-bit varint"); + // TODO(haberman) guarantee that this function return is >= 0 somehow, + // so we know this path will always be treated as error by our caller. + // Right now the size_t -> int32_t can overflow and produce negative values. + *u32 = 0; + return upb_pbdecoder_suspend(d); + } + *u32 = u64; + return DECODE_OK; +} + +// Decodes a fixed32 from the current buffer position. +// Returns a status code as described in decoder.int.h. +// TODO: proper byte swapping for big-endian machines. +FORCEINLINE int32_t decode_fixed32(upb_pbdecoder *d, uint32_t *u32) { + return getbytes(d, u32, 4); +} + +// Decodes a fixed64 from the current buffer position. +// Returns a status code as described in decoder.int.h. +// TODO: proper byte swapping for big-endian machines. +FORCEINLINE int32_t decode_fixed64(upb_pbdecoder *d, uint64_t *u64) { + return getbytes(d, u64, 8); +} + +// Non-static versions of the above functions. +// These are called by the JIT for fallback paths. +int32_t upb_pbdecoder_decode_f32(upb_pbdecoder *d, uint32_t *u32) { + return decode_fixed32(d, u32); +} + +int32_t upb_pbdecoder_decode_f64(upb_pbdecoder *d, uint64_t *u64) { + return decode_fixed64(d, u64); +} + +static double as_double(uint64_t n) { double d; memcpy(&d, &n, 8); return d; } +static float as_float(uint32_t n) { float f; memcpy(&f, &n, 4); return f; } + +// Pushes a frame onto the decoder stack. +static bool decoder_push(upb_pbdecoder *d, uint64_t end) { + upb_pbdecoder_frame *fr = d->top; + + if (end > fr->end_ofs) { + seterr(d, "Submessage end extends past enclosing submessage."); + return false; + } else if ((fr + 1) == d->limit) { + seterr(d, kPbDecoderStackOverflow); + return false; + } + + fr++; + fr->end_ofs = end; + fr->dispatch = NULL; + fr->groupnum = 0; + d->top = fr; + return true; +} + +static bool pushtagdelim(upb_pbdecoder *d, uint32_t arg) { + // While we expect to see an "end" tag (either ENDGROUP or a non-sequence + // field number) prior to hitting any enclosing submessage end, pushing our + // existing delim end prevents us from continuing to parse values from a + // corrupt proto that doesn't give us an END tag in time. + if (!decoder_push(d, d->top->end_ofs)) + return false; + d->top->groupnum = arg; + return true; +} + +// Pops a frame from the decoder stack. +static void decoder_pop(upb_pbdecoder *d) { d->top--; } + +NOINLINE int32_t upb_pbdecoder_checktag_slow(upb_pbdecoder *d, + uint64_t expected) { + uint64_t data = 0; + size_t bytes = upb_value_size(expected); + size_t read = peekbytes(d, &data, bytes); + if (read == bytes && data == expected) { + // Advance past matched bytes. + int32_t ok = getbytes(d, &data, read); + UPB_ASSERT_VAR(ok, ok < 0); + return DECODE_OK; + } else if (read < bytes && memcmp(&data, &expected, read) == 0) { + return suspend_save(d); + } else { + return DECODE_MISMATCH; + } +} + +int32_t upb_pbdecoder_skipunknown(upb_pbdecoder *d, int32_t fieldnum, + uint8_t wire_type) { + if (fieldnum >= 0) + goto have_tag; + + while (true) { + uint32_t tag; + CHECK_RETURN(decode_v32(d, &tag)); + wire_type = tag & 0x7; + fieldnum = tag >> 3; + +have_tag: + if (fieldnum == 0) { + seterr(d, "Saw invalid field number (0)"); + return upb_pbdecoder_suspend(d); + } + + // TODO: deliver to unknown field callback. + switch (wire_type) { + case UPB_WIRE_TYPE_32BIT: + CHECK_RETURN(skip(d, 4)); + break; + case UPB_WIRE_TYPE_64BIT: + CHECK_RETURN(skip(d, 8)); + break; + case UPB_WIRE_TYPE_VARINT: { + uint64_t u64; + CHECK_RETURN(decode_varint(d, &u64)); + break; + } + case UPB_WIRE_TYPE_DELIMITED: { + uint32_t len; + CHECK_RETURN(decode_v32(d, &len)); + CHECK_RETURN(skip(d, len)); + break; + } + case UPB_WIRE_TYPE_START_GROUP: + CHECK_SUSPEND(pushtagdelim(d, -fieldnum)); + break; + case UPB_WIRE_TYPE_END_GROUP: + if (fieldnum == -d->top->groupnum) { + decoder_pop(d); + } else if (fieldnum == d->top->groupnum) { + return DECODE_ENDGROUP; + } else { + seterr(d, "Unmatched ENDGROUP tag."); + return upb_pbdecoder_suspend(d); + } + break; + default: + seterr(d, "Invalid wire type"); + return upb_pbdecoder_suspend(d); + } + + if (d->top->groupnum >= 0) { + return DECODE_OK; + } + + if (d->ptr == d->delim_end) { + seterr(d, "Enclosing submessage ended in the middle of value or group"); + // Unlike most errors we notice during parsing, right now we have consumed + // all of the user's input. + // + // There are three different options for how to handle this case: + // + // 1. decode() = short count, error = set + // 2. decode() = full count, error = set + // 3. decode() = full count, error NOT set, short count and error will + // be reported on next call to decode() (or end()) + // + // (1) and (3) have the advantage that they preserve the invariant that an + // error occurs iff decode() returns a short count. + // + // (2) and (3) have the advantage of reflecting the fact that all of the + // bytes were in fact parsed (and possibly delivered to the unknown field + // handler, in the future when that is supported). + // + // (3) requires extra state in the decode (a place to store the "permanent + // error" that we should return for all subsequent attempts to decode). + // But we likely want this anyway. + // + // Right now we do (1), thanks to the fact that we checkpoint *after* this + // check. (3) may be a better choice long term; unclear at the moment. + return upb_pbdecoder_suspend(d); + } + + checkpoint(d); + } +} + +static void goto_endmsg(upb_pbdecoder *d) { + upb_value v; + bool found = upb_inttable_lookup32(d->top->dispatch, DISPATCH_ENDMSG, &v); + UPB_ASSERT_VAR(found, found); + d->pc = d->top->base + upb_value_getuint64(v); +} + +// Parses a tag and jumps to the corresponding bytecode instruction for this +// field. +// +// If the tag is unknown (or the wire type doesn't match), parses the field as +// unknown. If the tag is a valid ENDGROUP tag, jumps to the bytecode +// instruction for the end of message. +static int32_t dispatch(upb_pbdecoder *d) { + upb_inttable *dispatch = d->top->dispatch; + + // Decode tag. + uint32_t tag; + CHECK_RETURN(decode_v32(d, &tag)); + uint8_t wire_type = tag & 0x7; + uint32_t fieldnum = tag >> 3; + + // Lookup tag. Because of packed/non-packed compatibility, we have to + // check the wire type against two possibilities. + upb_value val; + if (fieldnum != DISPATCH_ENDMSG && + upb_inttable_lookup32(dispatch, fieldnum, &val)) { + uint64_t v = upb_value_getuint64(val); + if (wire_type == (v & 0xff)) { + d->pc = d->top->base + (v >> 16); + return DECODE_OK; + } else if (wire_type == ((v >> 8) & 0xff)) { + bool found = + upb_inttable_lookup(dispatch, fieldnum + UPB_MAX_FIELDNUMBER, &val); + UPB_ASSERT_VAR(found, found); + d->pc = d->top->base + upb_value_getuint64(val); + return DECODE_OK; + } + } + + // Unknown field or ENDGROUP. + int32_t ret = upb_pbdecoder_skipunknown(d, fieldnum, wire_type); + + if (ret == DECODE_ENDGROUP) { + goto_endmsg(d); + return DECODE_OK; + } else { + d->pc = d->last - 1; // Rewind to CHECKDELIM. + return ret; + } +} + +// Callers know that the stack is more than one deep because the opcodes that +// call this only occur after PUSH operations. +upb_pbdecoder_frame *outer_frame(upb_pbdecoder *d) { + assert(d->top != d->stack); + return d->top - 1; +} + + +/* The main decoding loop *****************************************************/ + +// The main decoder VM function. Uses traditional bytecode dispatch loop with a +// switch() statement. +size_t upb_pbdecoder_decode(void *closure, const void *hd, const char *buf, + size_t size, const upb_bufhandle *handle) { + upb_pbdecoder *d = closure; + const mgroup *group = hd; + assert(buf); + int32_t result = upb_pbdecoder_resume(d, NULL, buf, size, handle); + if (result == DECODE_ENDGROUP) { + goto_endmsg(d); + } + CHECK_RETURN(result); + UPB_UNUSED(group); + +#define VMCASE(op, code) \ + case op: { code; if (consumes_input(op)) checkpoint(d); break; } +#define PRIMITIVE_OP(type, wt, name, convfunc, ctype) \ + VMCASE(OP_PARSE_ ## type, { \ + ctype val; \ + CHECK_RETURN(decode_ ## wt(d, &val)); \ + upb_sink_put ## name(&d->top->sink, arg, (convfunc)(val)); \ + }) + + while(1) { + d->last = d->pc; + int32_t instruction = *d->pc++; + opcode op = getop(instruction); + uint32_t arg = instruction >> 8; + int32_t longofs = arg; + assert(d->ptr != d->residual_end); +#ifdef UPB_DUMP_BYTECODE + fprintf(stderr, "s_ofs=%d buf_ofs=%d data_rem=%d buf_rem=%d delim_rem=%d " + "%x %s (%d)\n", + (int)offset(d), + (int)(d->ptr - d->buf), + (int)(d->data_end - d->ptr), + (int)(d->end - d->ptr), + (int)((d->top->end_ofs - d->bufstart_ofs) - (d->ptr - d->buf)), + (int)(d->pc - 1 - group->bytecode), + upb_pbdecoder_getopname(op), + arg); +#endif + switch (op) { + // Technically, we are losing data if we see a 32-bit varint that is not + // properly sign-extended. We could detect this and error about the data + // loss, but proto2 does not do this, so we pass. + PRIMITIVE_OP(INT32, varint, int32, int32_t, uint64_t) + PRIMITIVE_OP(INT64, varint, int64, int64_t, uint64_t) + PRIMITIVE_OP(UINT32, varint, uint32, uint32_t, uint64_t) + PRIMITIVE_OP(UINT64, varint, uint64, uint64_t, uint64_t) + PRIMITIVE_OP(FIXED32, fixed32, uint32, uint32_t, uint32_t) + PRIMITIVE_OP(FIXED64, fixed64, uint64, uint64_t, uint64_t) + PRIMITIVE_OP(SFIXED32, fixed32, int32, int32_t, uint32_t) + PRIMITIVE_OP(SFIXED64, fixed64, int64, int64_t, uint64_t) + PRIMITIVE_OP(BOOL, varint, bool, bool, uint64_t) + PRIMITIVE_OP(DOUBLE, fixed64, double, as_double, uint64_t) + PRIMITIVE_OP(FLOAT, fixed32, float, as_float, uint32_t) + PRIMITIVE_OP(SINT32, varint, int32, upb_zzdec_32, uint64_t) + PRIMITIVE_OP(SINT64, varint, int64, upb_zzdec_64, uint64_t) + + VMCASE(OP_SETDISPATCH, + d->top->base = d->pc - 1; + memcpy(&d->top->dispatch, d->pc, sizeof(void*)); + d->pc += sizeof(void*) / sizeof(uint32_t); + ) + VMCASE(OP_STARTMSG, + CHECK_SUSPEND(upb_sink_startmsg(&d->top->sink)); + ) + VMCASE(OP_ENDMSG, + CHECK_SUSPEND(upb_sink_endmsg(&d->top->sink, d->status)); + ) + VMCASE(OP_STARTSEQ, + upb_pbdecoder_frame *outer = outer_frame(d); + CHECK_SUSPEND(upb_sink_startseq(&outer->sink, arg, &d->top->sink)); + ) + VMCASE(OP_ENDSEQ, + CHECK_SUSPEND(upb_sink_endseq(&d->top->sink, arg)); + ) + VMCASE(OP_STARTSUBMSG, + upb_pbdecoder_frame *outer = outer_frame(d); + CHECK_SUSPEND(upb_sink_startsubmsg(&outer->sink, arg, &d->top->sink)); + ) + VMCASE(OP_ENDSUBMSG, + CHECK_SUSPEND(upb_sink_endsubmsg(&d->top->sink, arg)); + ) + VMCASE(OP_STARTSTR, + uint32_t len = d->top->end_ofs - offset(d); + upb_pbdecoder_frame *outer = outer_frame(d); + CHECK_SUSPEND(upb_sink_startstr(&outer->sink, arg, len, &d->top->sink)); + if (len == 0) { + d->pc++; // Skip OP_STRING. + } + ) + VMCASE(OP_STRING, + uint32_t len = curbufleft(d); + size_t n = upb_sink_putstring(&d->top->sink, arg, d->ptr, len, handle); + if (n > len) { + if (n > d->top->end_ofs - offset(d)) { + seterr(d, "Tried to skip past end of string."); + return upb_pbdecoder_suspend(d); + } else { + int32_t ret = skip(d, n); + // This shouldn't return DECODE_OK, because n > len. + assert(ret >= 0); + return ret; + } + } + advance(d, n); + if (n < len || d->delim_end == NULL) { + // We aren't finished with this string yet. + d->pc--; // Repeat OP_STRING. + if (n > 0) checkpoint(d); + return upb_pbdecoder_suspend(d); + } + ) + VMCASE(OP_ENDSTR, + CHECK_SUSPEND(upb_sink_endstr(&d->top->sink, arg)); + ) + VMCASE(OP_PUSHTAGDELIM, + CHECK_SUSPEND(pushtagdelim(d, arg)); + ) + VMCASE(OP_SETBIGGROUPNUM, + d->top->groupnum = *d->pc++; + ) + VMCASE(OP_POP, + assert(d->top > d->stack); + decoder_pop(d); + ) + VMCASE(OP_PUSHLENDELIM, + uint32_t len; + CHECK_RETURN(decode_v32(d, &len)); + CHECK_SUSPEND(decoder_push(d, offset(d) + len)); + set_delim_end(d); + ) + VMCASE(OP_SETDELIM, + set_delim_end(d); + ) + VMCASE(OP_CHECKDELIM, + // We are guaranteed of this assert because we never allow ourselves to + // consume bytes beyond data_end, which covers delim_end when non-NULL. + assert(!(d->delim_end && d->ptr > d->delim_end)); + if (d->ptr == d->delim_end) + d->pc += longofs; + ) + VMCASE(OP_CALL, + d->callstack[d->call_len++] = d->pc; + d->pc += longofs; + ) + VMCASE(OP_RET, + assert(d->call_len > 0); + d->pc = d->callstack[--d->call_len]; + ) + VMCASE(OP_BRANCH, + d->pc += longofs; + ) + VMCASE(OP_TAG1, + CHECK_SUSPEND(curbufleft(d) > 0); + uint8_t expected = (arg >> 8) & 0xff; + if (*d->ptr == expected) { + advance(d, 1); + } else { + int8_t shortofs; + badtag: + shortofs = arg; + if (shortofs == LABEL_DISPATCH) { + CHECK_RETURN(dispatch(d)); + } else { + d->pc += shortofs; + break; // Avoid checkpoint(). + } + } + ) + VMCASE(OP_TAG2, + CHECK_SUSPEND(curbufleft(d) > 0); + uint16_t expected = (arg >> 8) & 0xffff; + if (curbufleft(d) >= 2) { + uint16_t actual; + memcpy(&actual, d->ptr, 2); + if (expected == actual) { + advance(d, 2); + } else { + goto badtag; + } + } else { + int32_t result = upb_pbdecoder_checktag_slow(d, expected); + if (result == DECODE_MISMATCH) goto badtag; + if (result >= 0) return result; + } + ) + VMCASE(OP_TAGN, { + uint64_t expected; + memcpy(&expected, d->pc, 8); + d->pc += 2; + int32_t result = upb_pbdecoder_checktag_slow(d, expected); + if (result == DECODE_MISMATCH) goto badtag; + if (result >= 0) return result; + }) + VMCASE(OP_DISPATCH, { + CHECK_RETURN(dispatch(d)); + }) + VMCASE(OP_HALT, { + return size; + }) + } + } +} + +void *upb_pbdecoder_startbc(void *closure, const void *pc, size_t size_hint) { + upb_pbdecoder *d = closure; + UPB_UNUSED(size_hint); + d->call_len = 1; + d->pc = pc; + return d; +} + +void *upb_pbdecoder_startjit(void *closure, const void *hd, size_t size_hint) { + UPB_UNUSED(hd); + UPB_UNUSED(size_hint); + upb_pbdecoder *d = closure; + d->call_len = 0; + return d; +} + +bool upb_pbdecoder_end(void *closure, const void *handler_data) { + upb_pbdecoder *d = closure; + const upb_pbdecodermethod *method = handler_data; + + if (d->residual_end > d->residual) { + seterr(d, "Unexpected EOF"); + return false; + } + + if (d->top->end_ofs != UINT64_MAX) { + seterr(d, "Unexpected EOF inside delimited string"); + return false; + } + + // Message ends here. + uint64_t end = offset(d); + d->top->end_ofs = end; + + char dummy; +#ifdef UPB_USE_JIT_X64 + const mgroup *group = (const mgroup*)method->group; + if (group->jit_code) { + if (d->top != d->stack) + d->stack->end_ofs = 0; + group->jit_code(closure, method->code_base.ptr, &dummy, 0, NULL); + } else { +#endif + d->stack->end_ofs = end; + const uint32_t *p = d->pc; + // Check the previous bytecode, but guard against beginning. + if (p != method->code_base.ptr) p--; + if (getop(*p) == OP_CHECKDELIM) { + // Rewind from OP_TAG* to OP_CHECKDELIM. + assert(getop(*d->pc) == OP_TAG1 || + getop(*d->pc) == OP_TAG2 || + getop(*d->pc) == OP_TAGN || + getop(*d->pc == OP_DISPATCH)); + d->pc = p; + } + upb_pbdecoder_decode(closure, handler_data, &dummy, 0, NULL); +#ifdef UPB_USE_JIT_X64 + } +#endif + + if (d->call_len != 0) { + seterr(d, "Unexpected EOF"); + return false; + } + + return true; +} + +void upb_pbdecoder_init(upb_pbdecoder *d, const upb_pbdecodermethod *m, + upb_status *s) { + d->limit = &d->stack[UPB_DECODER_MAX_NESTING]; + upb_bytessink_reset(&d->input_, &m->input_handler_, d); + d->method_ = m; + d->callstack[0] = &halt; + d->status = s; + upb_pbdecoder_reset(d); +} + +void upb_pbdecoder_reset(upb_pbdecoder *d) { + d->top = d->stack; + d->top->end_ofs = UINT64_MAX; + d->top->groupnum = 0; + d->bufstart_ofs = 0; + d->ptr = d->residual; + d->buf = d->residual; + d->end = d->residual; + d->residual_end = d->residual; + d->call_len = 1; +} + +uint64_t upb_pbdecoder_bytesparsed(const upb_pbdecoder *d) { + return offset(d); +} + +// Not currently required, but to support outgrowing the static stack we need +// this. +void upb_pbdecoder_uninit(upb_pbdecoder *d) { + UPB_UNUSED(d); +} + +const upb_pbdecodermethod *upb_pbdecoder_method(const upb_pbdecoder *d) { + return d->method_; +} + +bool upb_pbdecoder_resetoutput(upb_pbdecoder *d, upb_sink* sink) { + // TODO(haberman): do we need to test whether the decoder is already on the + // stack (like calling this from within a callback)? Should we support + // rebinding the output at all? + assert(sink); + if (d->method_->dest_handlers_) { + if (sink->handlers != d->method_->dest_handlers_) + return false; + } + upb_sink_reset(&d->top->sink, sink->handlers, sink->closure); + return true; +} + +upb_bytessink *upb_pbdecoder_input(upb_pbdecoder *d) { + return &d->input_; +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2014 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Since we are implementing pure handlers (ie. without any out-of-band access + * to pre-computed lengths), we have to buffer all submessages before we can + * emit even their first byte. + * + * Not knowing the size of submessages also means we can't write a perfect + * zero-copy implementation, even with buffering. Lengths are stored as + * varints, which means that we don't know how many bytes to reserve for the + * length until we know what the length is. + * + * This leaves us with three main choices: + * + * 1. buffer all submessage data in a temporary buffer, then copy it exactly + * once into the output buffer. + * + * 2. attempt to buffer data directly into the output buffer, estimating how + * many bytes each length will take. When our guesses are wrong, use + * memmove() to grow or shrink the allotted space. + * + * 3. buffer directly into the output buffer, allocating a max length + * ahead-of-time for each submessage length. If we overallocated, we waste + * space, but no memcpy() or memmove() is required. This approach requires + * defining a maximum size for submessages and rejecting submessages that + * exceed that size. + * + * (2) and (3) have the potential to have better performance, but they are more + * complicated and subtle to implement: + * + * (3) requires making an arbitrary choice of the maximum message size; it + * wastes space when submessages are shorter than this and fails + * completely when they are longer. This makes it more finicky and + * requires configuration based on the input. It also makes it impossible + * to perfectly match the output of reference encoders that always use the + * optimal amount of space for each length. + * + * (2) requires guessing the the size upfront, and if multiple lengths are + * guessed wrong the minimum required number of memmove() operations may + * be complicated to compute correctly. Implemented properly, it may have + * a useful amortized or average cost, but more investigation is required + * to determine this and what the optimal algorithm is to achieve it. + * + * (1) makes you always pay for exactly one copy, but its implementation is + * the simplest and its performance is predictable. + * + * So for now, we implement (1) only. If we wish to optimize later, we should + * be able to do it without affecting users. + * + * The strategy is to buffer the segments of data that do *not* depend on + * unknown lengths in one buffer, and keep a separate buffer of segment pointers + * and lengths. When the top-level submessage ends, we can go beginning to end, + * alternating the writing of lengths with memcpy() of the rest of the data. + * At the top level though, no buffering is required. + */ + + +#include + +/* low-level buffering ********************************************************/ + +// Low-level functions for interacting with the output buffer. + +// TODO(haberman): handle pushback +static void putbuf(upb_pb_encoder *e, const char *buf, size_t len) { + size_t n = upb_bytessink_putbuf(e->output_, e->subc, buf, len, NULL); + UPB_ASSERT_VAR(n, n == len); +} + +static upb_pb_encoder_segment *top(upb_pb_encoder *e) { + return &e->segbuf[*e->top]; +} + +// Call to ensure that at least "bytes" bytes are available for writing at +// e->ptr. Returns false if the bytes could not be allocated. +static bool reserve(upb_pb_encoder *e, size_t bytes) { + if ((e->limit - e->ptr) < bytes) { + size_t needed = bytes + (e->ptr - e->buf); + size_t old_size = e->limit - e->buf; + size_t new_size = old_size; + while (new_size < needed) { + new_size *= 2; + } + + char *realloc_from = (e->buf == e->initbuf) ? NULL : e->buf; + char *new_buf = realloc(realloc_from, new_size); + + if (new_buf == NULL) { + return false; + } + + if (realloc_from == NULL) { + memcpy(new_buf, e->initbuf, old_size); + } + + e->ptr = new_buf + (e->ptr - e->buf); + e->runbegin = new_buf + (e->runbegin - e->buf); + e->limit = new_buf + new_size; + e->buf = new_buf; + } + + return true; +} + +// Call when "bytes" bytes have been writte at e->ptr. The caller *must* have +// previously called reserve() with at least this many bytes. +static void encoder_advance(upb_pb_encoder *e, size_t bytes) { + assert((e->limit - e->ptr) >= bytes); + e->ptr += bytes; +} + +// Call when all of the bytes for a handler have been written. Flushes the +// bytes if possible and necessary, returning false if this failed. +static bool commit(upb_pb_encoder *e) { + if (!e->top) { + // We aren't inside a delimited region. Flush our accumulated bytes to + // the output. + // + // TODO(haberman): in the future we may want to delay flushing for + // efficiency reasons. + putbuf(e, e->buf, e->ptr - e->buf); + e->ptr = e->buf; + } + + return true; +} + +// Writes the given bytes to the buffer, handling reserve/advance. +static bool encode_bytes(upb_pb_encoder *e, const void *data, size_t len) { + if (!reserve(e, len)) { + return false; + } + + memcpy(e->ptr, data, len); + encoder_advance(e, len); + return true; +} + +// Finish the current run by adding the run totals to the segment and message +// length. +static void accumulate(upb_pb_encoder *e) { + assert(e->ptr >= e->runbegin); + size_t run_len = e->ptr - e->runbegin; + e->segptr->seglen += run_len; + top(e)->msglen += run_len; + e->runbegin = e->ptr; +} + +// Call to indicate the start of delimited region for which the full length is +// not yet known. All data will be buffered until the length is known. +// Delimited regions may be nested; their lengths will all be tracked properly. +static bool start_delim(upb_pb_encoder *e) { + if (e->top) { + // We are already buffering, advance to the next segment and push it on the + // stack. + accumulate(e); + + if (++e->top == e->stacklimit) { + // TODO(haberman): grow stack? + return false; + } + + if (++e->segptr == e->seglimit) { + upb_pb_encoder_segment *realloc_from = + (e->segbuf == e->seginitbuf) ? NULL : e->segbuf; + size_t old_size = + (e->seglimit - e->segbuf) * sizeof(upb_pb_encoder_segment); + size_t new_size = old_size * 2; + upb_pb_encoder_segment *new_buf = realloc(realloc_from, new_size); + + if (new_buf == NULL) { + return false; + } + + if (realloc_from == NULL) { + memcpy(new_buf, e->seginitbuf, old_size); + } + + e->segptr = new_buf + (e->segptr - e->segbuf); + e->seglimit = new_buf + (new_size / sizeof(upb_pb_encoder_segment)); + e->segbuf = new_buf; + } + } else { + // We were previously at the top level, start buffering. + e->segptr = e->segbuf; + e->top = e->stack; + e->runbegin = e->ptr; + } + + *e->top = e->segptr - e->segbuf; + e->segptr->seglen = 0; + e->segptr->msglen = 0; + + return true; +} + +// Call to indicate the end of a delimited region. We now know the length of +// the delimited region. If we are not nested inside any other delimited +// regions, we can now emit all of the buffered data we accumulated. +static bool end_delim(upb_pb_encoder *e) { + accumulate(e); + size_t msglen = top(e)->msglen; + + if (e->top == e->stack) { + // All lengths are now available, emit all buffered data. + char buf[UPB_PB_VARINT_MAX_LEN]; + upb_pb_encoder_segment *s; + const char *ptr = e->buf; + for (s = e->segbuf; s <= e->segptr; s++) { + size_t lenbytes = upb_vencode64(s->msglen, buf); + putbuf(e, buf, lenbytes); + putbuf(e, ptr, s->seglen); + ptr += s->seglen; + } + + e->ptr = e->buf; + e->top = NULL; + } else { + // Need to keep buffering; propagate length info into enclosing submessages. + --e->top; + top(e)->msglen += msglen + upb_varint_size(msglen); + } + + return true; +} + + +/* tag_t **********************************************************************/ + +// A precomputed (pre-encoded) tag and length. + +typedef struct { + uint8_t bytes; + char tag[7]; +} tag_t; + +// Allocates a new tag for this field, and sets it in these handlerattr. +static void new_tag(upb_handlers *h, const upb_fielddef *f, upb_wiretype_t wt, + upb_handlerattr *attr) { + uint32_t n = upb_fielddef_number(f); + + tag_t *tag = malloc(sizeof(tag_t)); + tag->bytes = upb_vencode64((n << 3) | wt, tag->tag); + + upb_handlerattr_init(attr); + upb_handlerattr_sethandlerdata(attr, tag); + upb_handlers_addcleanup(h, tag, free); +} + +static bool encode_tag(upb_pb_encoder *e, const tag_t *tag) { + return encode_bytes(e, tag->tag, tag->bytes); +} + + +/* encoding of wire types *****************************************************/ + +static bool encode_fixed64(upb_pb_encoder *e, uint64_t val) { + // TODO(haberman): byte-swap for big endian. + return encode_bytes(e, &val, sizeof(uint64_t)); +} + +static bool encode_fixed32(upb_pb_encoder *e, uint32_t val) { + // TODO(haberman): byte-swap for big endian. + return encode_bytes(e, &val, sizeof(uint32_t)); +} + +static bool encode_varint(upb_pb_encoder *e, uint64_t val) { + if (!reserve(e, UPB_PB_VARINT_MAX_LEN)) { + return false; + } + + encoder_advance(e, upb_vencode64(val, e->ptr)); + return true; +} + +static uint64_t dbl2uint64(double d) { + uint64_t ret; + memcpy(&ret, &d, sizeof(uint64_t)); + return ret; +} + +static uint32_t flt2uint32(float d) { + uint32_t ret; + memcpy(&ret, &d, sizeof(uint32_t)); + return ret; +} + + +/* encoding of proto types ****************************************************/ + +static bool startmsg(void *c, const void *hd) { + upb_pb_encoder *e = c; + UPB_UNUSED(hd); + if (e->depth++ == 0) { + upb_bytessink_start(e->output_, 0, &e->subc); + } + return true; +} + +static bool endmsg(void *c, const void *hd, upb_status *status) { + upb_pb_encoder *e = c; + UPB_UNUSED(hd); + UPB_UNUSED(status); + if (--e->depth == 0) { + upb_bytessink_end(e->output_); + } + return true; +} + +static void *encode_startdelimfield(void *c, const void *hd) { + bool ok = encode_tag(c, hd) && commit(c) && start_delim(c); + return ok ? c : UPB_BREAK; +} + +static bool encode_enddelimfield(void *c, const void *hd) { + UPB_UNUSED(hd); + return end_delim(c); +} + +static void *encode_startgroup(void *c, const void *hd) { + return (encode_tag(c, hd) && commit(c)) ? c : UPB_BREAK; +} + +static bool encode_endgroup(void *c, const void *hd) { + return encode_tag(c, hd) && commit(c); +} + +static void *encode_startstr(void *c, const void *hd, size_t size_hint) { + UPB_UNUSED(size_hint); + return encode_startdelimfield(c, hd); +} + +static size_t encode_strbuf(void *c, const void *hd, const char *buf, + size_t len, const upb_bufhandle *h) { + UPB_UNUSED(hd); + UPB_UNUSED(h); + return encode_bytes(c, buf, len) ? len : 0; +} + +#define T(type, ctype, convert, encode) \ + static bool encode_scalar_##type(void *e, const void *hd, ctype val) { \ + return encode_tag(e, hd) && encode(e, (convert)(val)) && commit(e); \ + } \ + static bool encode_packed_##type(void *e, const void *hd, ctype val) { \ + UPB_UNUSED(hd); \ + return encode(e, (convert)(val)); \ + } + +T(double, double, dbl2uint64, encode_fixed64) +T(float, float, flt2uint32, encode_fixed32); +T(int64, int64_t, uint64_t, encode_varint); +T(int32, int32_t, uint32_t, encode_varint); +T(fixed64, uint64_t, uint64_t, encode_fixed64); +T(fixed32, uint32_t, uint32_t, encode_fixed32); +T(bool, bool, bool, encode_varint); +T(uint32, uint32_t, uint32_t, encode_varint); +T(uint64, uint64_t, uint64_t, encode_varint); +T(enum, int32_t, uint32_t, encode_varint); +T(sfixed32, int32_t, uint32_t, encode_fixed32); +T(sfixed64, int64_t, uint64_t, encode_fixed64); +T(sint32, int32_t, upb_zzenc_32, encode_varint); +T(sint64, int64_t, upb_zzenc_64, encode_varint); + +#undef T + + +/* code to build the handlers *************************************************/ + +static void newhandlers_callback(const void *closure, upb_handlers *h) { + UPB_UNUSED(closure); + + upb_handlers_setstartmsg(h, startmsg, NULL); + upb_handlers_setendmsg(h, endmsg, NULL); + + const upb_msgdef *m = upb_handlers_msgdef(h); + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + bool packed = upb_fielddef_isseq(f) && upb_fielddef_isprimitive(f) && + upb_fielddef_packed(f); + upb_handlerattr attr; + upb_wiretype_t wt = + packed ? UPB_WIRE_TYPE_DELIMITED + : upb_pb_native_wire_types[upb_fielddef_descriptortype(f)]; + + // Pre-encode the tag for this field. + new_tag(h, f, wt, &attr); + + if (packed) { + upb_handlers_setstartseq(h, f, encode_startdelimfield, &attr); + upb_handlers_setendseq(h, f, encode_enddelimfield, &attr); + } + +#define T(upper, lower, upbtype) \ + case UPB_DESCRIPTOR_TYPE_##upper: \ + if (packed) { \ + upb_handlers_set##upbtype(h, f, encode_packed_##lower, &attr); \ + } else { \ + upb_handlers_set##upbtype(h, f, encode_scalar_##lower, &attr); \ + } \ + break; + + switch (upb_fielddef_descriptortype(f)) { + T(DOUBLE, double, double); + T(FLOAT, float, float); + T(INT64, int64, int64); + T(INT32, int32, int32); + T(FIXED64, fixed64, uint64); + T(FIXED32, fixed32, uint32); + T(BOOL, bool, bool); + T(UINT32, uint32, uint32); + T(UINT64, uint64, uint64); + T(ENUM, enum, int32); + T(SFIXED32, sfixed32, int32); + T(SFIXED64, sfixed64, int64); + T(SINT32, sint32, int32); + T(SINT64, sint64, int64); + case UPB_DESCRIPTOR_TYPE_STRING: + case UPB_DESCRIPTOR_TYPE_BYTES: + upb_handlers_setstartstr(h, f, encode_startstr, &attr); + upb_handlers_setendstr(h, f, encode_enddelimfield, &attr); + upb_handlers_setstring(h, f, encode_strbuf, &attr); + break; + case UPB_DESCRIPTOR_TYPE_MESSAGE: + upb_handlers_setstartsubmsg(h, f, encode_startdelimfield, &attr); + upb_handlers_setendsubmsg(h, f, encode_enddelimfield, &attr); + break; + case UPB_DESCRIPTOR_TYPE_GROUP: { + // Endgroup takes a different tag (wire_type = END_GROUP). + upb_handlerattr attr2; + new_tag(h, f, UPB_WIRE_TYPE_END_GROUP, &attr2); + + upb_handlers_setstartsubmsg(h, f, encode_startgroup, &attr); + upb_handlers_setendsubmsg(h, f, encode_endgroup, &attr2); + + upb_handlerattr_uninit(&attr2); + break; + } + } + +#undef T + + upb_handlerattr_uninit(&attr); + } +} + + +/* public API *****************************************************************/ + +const upb_handlers *upb_pb_encoder_newhandlers(const upb_msgdef *m, + const void *owner) { + return upb_handlers_newfrozen(m, owner, newhandlers_callback, NULL); +} + +#define ARRAYSIZE(x) (sizeof(x) / sizeof(x[0])) + +void upb_pb_encoder_init(upb_pb_encoder *e, const upb_handlers *h) { + e->output_ = NULL; + e->subc = NULL; + e->buf = e->initbuf; + e->ptr = e->buf; + e->limit = e->buf + ARRAYSIZE(e->initbuf); + e->segbuf = e->seginitbuf; + e->seglimit = e->segbuf + ARRAYSIZE(e->seginitbuf); + e->stacklimit = e->stack + ARRAYSIZE(e->stack); + upb_sink_reset(&e->input_, h, e); +} + +void upb_pb_encoder_uninit(upb_pb_encoder *e) { + if (e->buf != e->initbuf) { + free(e->buf); + } + + if (e->segbuf != e->seginitbuf) { + free(e->segbuf); + } +} + +void upb_pb_encoder_resetoutput(upb_pb_encoder *e, upb_bytessink *output) { + upb_pb_encoder_reset(e); + e->output_ = output; + e->subc = output->closure; +} + +void upb_pb_encoder_reset(upb_pb_encoder *e) { + e->segptr = NULL; + e->top = NULL; + e->depth = 0; +} + +upb_sink *upb_pb_encoder_input(upb_pb_encoder *e) { return &e->input_; } +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2010-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + + +#include +#include +#include + +upb_def **upb_load_defs_from_descriptor(const char *str, size_t len, int *n, + void *owner, upb_status *status) { + // Create handlers. + const upb_handlers *reader_h = upb_descreader_newhandlers(&reader_h); + upb_pbdecodermethodopts opts; + upb_pbdecodermethodopts_init(&opts, reader_h); + const upb_pbdecodermethod *decoder_m = + upb_pbdecodermethod_new(&opts, &decoder_m); + + upb_pbdecoder decoder; + upb_descreader reader; + + upb_pbdecoder_init(&decoder, decoder_m, status); + upb_descreader_init(&reader, reader_h, status); + upb_pbdecoder_resetoutput(&decoder, upb_descreader_input(&reader)); + + // Push input data. + bool ok = upb_bufsrc_putbuf(str, len, upb_pbdecoder_input(&decoder)); + + upb_def **ret = NULL; + + if (!ok) goto cleanup; + upb_def **defs = upb_descreader_getdefs(&reader, owner, n); + ret = malloc(sizeof(upb_def*) * (*n)); + memcpy(ret, defs, sizeof(upb_def*) * (*n)); + +cleanup: + upb_pbdecoder_uninit(&decoder); + upb_descreader_uninit(&reader); + upb_handlers_unref(reader_h, &reader_h); + upb_pbdecodermethod_unref(decoder_m, &decoder_m); + return ret; +} + +bool upb_load_descriptor_into_symtab(upb_symtab *s, const char *str, size_t len, + upb_status *status) { + int n; + upb_def **defs = upb_load_defs_from_descriptor(str, len, &n, &defs, status); + if (!defs) return false; + bool success = upb_symtab_add(s, defs, n, &defs, status); + free(defs); + return success; +} + +char *upb_readfile(const char *filename, size_t *len) { + FILE *f = fopen(filename, "rb"); + if(!f) return NULL; + if(fseek(f, 0, SEEK_END) != 0) goto error; + long size = ftell(f); + if(size < 0) goto error; + if(fseek(f, 0, SEEK_SET) != 0) goto error; + char *buf = malloc(size + 1); + if(size && fread(buf, size, 1, f) != 1) goto error; + fclose(f); + if (len) *len = size; + return buf; + +error: + fclose(f); + return NULL; +} + +bool upb_load_descriptor_file_into_symtab(upb_symtab *symtab, const char *fname, + upb_status *status) { + size_t len; + char *data = upb_readfile(fname, &len); + if (!data) { + if (status) upb_status_seterrf(status, "Couldn't read file: %s", fname); + return false; + } + bool success = upb_load_descriptor_into_symtab(symtab, data, len, status); + free(data); + return success; +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * OPT: This is not optimized at all. It uses printf() which parses the format + * string every time, and it allocates memory for every put. + */ + + +#include +#include +#include +#include +#include +#include + + +#define CHECK(x) if ((x) < 0) goto err; + +static const char *shortname(const char *longname) { + const char *last = strrchr(longname, '.'); + return last ? last + 1 : longname; +} + +static int indent(upb_textprinter *p) { + int i; + if (!p->single_line_) + for (i = 0; i < p->indent_depth_; i++) + upb_bytessink_putbuf(p->output_, p->subc, " ", 2, NULL); + return 0; +} + +static int endfield(upb_textprinter *p) { + const char ch = (p->single_line_ ? ' ' : '\n'); + upb_bytessink_putbuf(p->output_, p->subc, &ch, 1, NULL); + return 0; +} + +static int putescaped(upb_textprinter *p, const char *buf, size_t len, + bool preserve_utf8) { + // Based on CEscapeInternal() from Google's protobuf release. + char dstbuf[4096], *dst = dstbuf, *dstend = dstbuf + sizeof(dstbuf); + const char *end = buf + len; + + // I think hex is prettier and more useful, but proto2 uses octal; should + // investigate whether it can parse hex also. + const bool use_hex = false; + bool last_hex_escape = false; // true if last output char was \xNN + + for (; buf < end; buf++) { + if (dstend - dst < 4) { + upb_bytessink_putbuf(p->output_, p->subc, dstbuf, dst - dstbuf, NULL); + dst = dstbuf; + } + + bool is_hex_escape = false; + switch (*buf) { + case '\n': *(dst++) = '\\'; *(dst++) = 'n'; break; + case '\r': *(dst++) = '\\'; *(dst++) = 'r'; break; + case '\t': *(dst++) = '\\'; *(dst++) = 't'; break; + case '\"': *(dst++) = '\\'; *(dst++) = '\"'; break; + case '\'': *(dst++) = '\\'; *(dst++) = '\''; break; + case '\\': *(dst++) = '\\'; *(dst++) = '\\'; break; + default: + // Note that if we emit \xNN and the buf character after that is a hex + // digit then that digit must be escaped too to prevent it being + // interpreted as part of the character code by C. + if ((!preserve_utf8 || (uint8_t)*buf < 0x80) && + (!isprint(*buf) || (last_hex_escape && isxdigit(*buf)))) { + sprintf(dst, (use_hex ? "\\x%02x" : "\\%03o"), (uint8_t)*buf); + is_hex_escape = use_hex; + dst += 4; + } else { + *(dst++) = *buf; break; + } + } + last_hex_escape = is_hex_escape; + } + // Flush remaining data. + upb_bytessink_putbuf(p->output_, p->subc, dstbuf, dst - dstbuf, NULL); + return 0; +} + +bool putf(upb_textprinter *p, const char *fmt, ...) { + va_list args; + va_start(args, fmt); + + // Run once to get the length of the string. + va_list args_copy; + va_copy(args_copy, args); + int len = vsnprintf(NULL, 0, fmt, args_copy); + va_end(args_copy); + + // + 1 for NULL terminator (vsnprintf() requires it even if we don't). + char *str = malloc(len + 1); + if (!str) return false; + int written = vsnprintf(str, len + 1, fmt, args); + va_end(args); + UPB_ASSERT_VAR(written, written == len); + + bool ok = upb_bytessink_putbuf(p->output_, p->subc, str, len, NULL); + free(str); + return ok; +} + + +/* handlers *******************************************************************/ + +static bool textprinter_startmsg(void *c, const void *hd) { + UPB_UNUSED(hd); + upb_textprinter *p = c; + if (p->indent_depth_ == 0) { + upb_bytessink_start(p->output_, 0, &p->subc); + } + return true; +} + +static bool textprinter_endmsg(void *c, const void *hd, upb_status *s) { + UPB_UNUSED(hd); + UPB_UNUSED(s); + upb_textprinter *p = c; + if (p->indent_depth_ == 0) { + upb_bytessink_end(p->output_); + } + return true; +} + +#define TYPE(name, ctype, fmt) \ + static bool textprinter_put ## name(void *closure, const void *handler_data, \ + ctype val) { \ + upb_textprinter *p = closure; \ + const upb_fielddef *f = handler_data; \ + CHECK(indent(p)); \ + putf(p, "%s: " fmt, upb_fielddef_name(f), val); \ + CHECK(endfield(p)); \ + return true; \ + err: \ + return false; \ +} + +static bool textprinter_putbool(void *closure, const void *handler_data, + bool val) { + upb_textprinter *p = closure; + const upb_fielddef *f = handler_data; + CHECK(indent(p)); + putf(p, "%s: %s", upb_fielddef_name(f), val ? "true" : "false"); + CHECK(endfield(p)); + return true; +err: + return false; +} + +#define STRINGIFY_HELPER(x) #x +#define STRINGIFY_MACROVAL(x) STRINGIFY_HELPER(x) + +TYPE(int32, int32_t, "%" PRId32) +TYPE(int64, int64_t, "%" PRId64) +TYPE(uint32, uint32_t, "%" PRIu32); +TYPE(uint64, uint64_t, "%" PRIu64) +TYPE(float, float, "%." STRINGIFY_MACROVAL(FLT_DIG) "g") +TYPE(double, double, "%." STRINGIFY_MACROVAL(DBL_DIG) "g") + +#undef TYPE + +// Output a symbolic value from the enum if found, else just print as int32. +static bool textprinter_putenum(void *closure, const void *handler_data, + int32_t val) { + upb_textprinter *p = closure; + const upb_fielddef *f = handler_data; + const upb_enumdef *enum_def = upb_downcast_enumdef(upb_fielddef_subdef(f)); + const char *label = upb_enumdef_iton(enum_def, val); + if (label) { + indent(p); + putf(p, "%s: %s", upb_fielddef_name(f), label); + endfield(p); + } else { + if (!textprinter_putint32(closure, handler_data, val)) + return false; + } + return true; +} + +static void *textprinter_startstr(void *closure, const void *handler_data, + size_t size_hint) { + const upb_fielddef *f = handler_data; + UPB_UNUSED(size_hint); + upb_textprinter *p = closure; + indent(p); + putf(p, "%s: \"", upb_fielddef_name(f)); + return p; +} + +static bool textprinter_endstr(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_textprinter *p = closure; + putf(p, "\""); + endfield(p); + return true; +} + +static size_t textprinter_putstr(void *closure, const void *hd, const char *buf, + size_t len, const upb_bufhandle *handle) { + UPB_UNUSED(handle); + upb_textprinter *p = closure; + const upb_fielddef *f = hd; + CHECK(putescaped(p, buf, len, upb_fielddef_type(f) == UPB_TYPE_STRING)); + return len; +err: + return 0; +} + +static void *textprinter_startsubmsg(void *closure, const void *handler_data) { + upb_textprinter *p = closure; + const char *name = handler_data; + CHECK(indent(p)); + putf(p, "%s {%c", name, p->single_line_ ? ' ' : '\n'); + p->indent_depth_++; + return p; +err: + return UPB_BREAK; +} + +static bool textprinter_endsubmsg(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_textprinter *p = closure; + p->indent_depth_--; + CHECK(indent(p)); + upb_bytessink_putbuf(p->output_, p->subc, "}", 1, NULL); + CHECK(endfield(p)); + return true; +err: + return false; +} + + +/* Public API *****************************************************************/ + +void upb_textprinter_init(upb_textprinter *p, const upb_handlers *h) { + p->single_line_ = false; + p->indent_depth_ = 0; + upb_sink_reset(&p->input_, h, p); +} + +void upb_textprinter_uninit(upb_textprinter *p) { + UPB_UNUSED(p); +} + +void upb_textprinter_reset(upb_textprinter *p, bool single_line) { + p->single_line_ = single_line; + p->indent_depth_ = 0; +} + +static void onmreg(const void *c, upb_handlers *h) { + UPB_UNUSED(c); + const upb_msgdef *m = upb_handlers_msgdef(h); + + upb_handlers_setstartmsg(h, textprinter_startmsg, NULL); + upb_handlers_setendmsg(h, textprinter_endmsg, NULL); + + upb_msg_field_iter i; + for(upb_msg_field_begin(&i, m); + !upb_msg_field_done(&i); + upb_msg_field_next(&i)) { + upb_fielddef *f = upb_msg_iter_field(&i); + upb_handlerattr attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&attr, f); + switch (upb_fielddef_type(f)) { + case UPB_TYPE_INT32: + upb_handlers_setint32(h, f, textprinter_putint32, &attr); + break; + case UPB_TYPE_INT64: + upb_handlers_setint64(h, f, textprinter_putint64, &attr); + break; + case UPB_TYPE_UINT32: + upb_handlers_setuint32(h, f, textprinter_putuint32, &attr); + break; + case UPB_TYPE_UINT64: + upb_handlers_setuint64(h, f, textprinter_putuint64, &attr); + break; + case UPB_TYPE_FLOAT: + upb_handlers_setfloat(h, f, textprinter_putfloat, &attr); + break; + case UPB_TYPE_DOUBLE: + upb_handlers_setdouble(h, f, textprinter_putdouble, &attr); + break; + case UPB_TYPE_BOOL: + upb_handlers_setbool(h, f, textprinter_putbool, &attr); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: + upb_handlers_setstartstr(h, f, textprinter_startstr, &attr); + upb_handlers_setstring(h, f, textprinter_putstr, &attr); + upb_handlers_setendstr(h, f, textprinter_endstr, &attr); + break; + case UPB_TYPE_MESSAGE: { + const char *name = + upb_fielddef_istagdelim(f) + ? shortname(upb_msgdef_fullname(upb_fielddef_msgsubdef(f))) + : upb_fielddef_name(f); + upb_handlerattr_sethandlerdata(&attr, name); + upb_handlers_setstartsubmsg(h, f, textprinter_startsubmsg, &attr); + upb_handlers_setendsubmsg(h, f, textprinter_endsubmsg, &attr); + break; + } + case UPB_TYPE_ENUM: + upb_handlers_setint32(h, f, textprinter_putenum, &attr); + break; + } + } +} + +const upb_handlers *upb_textprinter_newhandlers(const upb_msgdef *m, + const void *owner) { + return upb_handlers_newfrozen(m, owner, &onmreg, NULL); +} + +upb_sink *upb_textprinter_input(upb_textprinter *p) { return &p->input_; } + +bool upb_textprinter_resetoutput(upb_textprinter *p, upb_bytessink *output) { + p->output_ = output; + return true; +} + +void upb_textprinter_setsingleline(upb_textprinter *p, bool single_line) { + p->single_line_ = single_line; +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2011 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + + +// Index is descriptor type. +const uint8_t upb_pb_native_wire_types[] = { + UPB_WIRE_TYPE_END_GROUP, // ENDGROUP + UPB_WIRE_TYPE_64BIT, // DOUBLE + UPB_WIRE_TYPE_32BIT, // FLOAT + UPB_WIRE_TYPE_VARINT, // INT64 + UPB_WIRE_TYPE_VARINT, // UINT64 + UPB_WIRE_TYPE_VARINT, // INT32 + UPB_WIRE_TYPE_64BIT, // FIXED64 + UPB_WIRE_TYPE_32BIT, // FIXED32 + UPB_WIRE_TYPE_VARINT, // BOOL + UPB_WIRE_TYPE_DELIMITED, // STRING + UPB_WIRE_TYPE_START_GROUP, // GROUP + UPB_WIRE_TYPE_DELIMITED, // MESSAGE + UPB_WIRE_TYPE_DELIMITED, // BYTES + UPB_WIRE_TYPE_VARINT, // UINT32 + UPB_WIRE_TYPE_VARINT, // ENUM + UPB_WIRE_TYPE_32BIT, // SFIXED32 + UPB_WIRE_TYPE_64BIT, // SFIXED64 + UPB_WIRE_TYPE_VARINT, // SINT32 + UPB_WIRE_TYPE_VARINT, // SINT64 +}; + +// A basic branch-based decoder, uses 32-bit values to get good performance +// on 32-bit architectures (but performs well on 64-bits also). +// This scheme comes from the original Google Protobuf implementation (proto2). +upb_decoderet upb_vdecode_max8_branch32(upb_decoderet r) { + upb_decoderet err = {NULL, 0}; + const char *p = r.p; + uint32_t low = (uint32_t)r.val; + uint32_t high = 0; + uint32_t b; + b = *(p++); low |= (b & 0x7fU) << 14; if (!(b & 0x80)) goto done; + b = *(p++); low |= (b & 0x7fU) << 21; if (!(b & 0x80)) goto done; + b = *(p++); low |= (b & 0x7fU) << 28; + high = (b & 0x7fU) >> 4; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 3; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 10; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 17; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 24; if (!(b & 0x80)) goto done; + b = *(p++); high |= (b & 0x7fU) << 31; if (!(b & 0x80)) goto done; + return err; + +done: + r.val = ((uint64_t)high << 32) | low; + r.p = p; + return r; +} + +// Like the previous, but uses 64-bit values. +upb_decoderet upb_vdecode_max8_branch64(upb_decoderet r) { + const char *p = r.p; + uint64_t val = r.val; + uint64_t b; + upb_decoderet err = {NULL, 0}; + b = *(p++); val |= (b & 0x7fU) << 14; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 21; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 28; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 35; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 42; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 49; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 56; if (!(b & 0x80)) goto done; + b = *(p++); val |= (b & 0x7fU) << 63; if (!(b & 0x80)) goto done; + return err; + +done: + r.val = val; + r.p = p; + return r; +} + +// Given an encoded varint v, returns an integer with a single bit set that +// indicates the end of the varint. Subtracting one from this value will +// yield a mask that leaves only bits that are part of the varint. Returns +// 0 if the varint is unterminated. +static uint64_t upb_get_vstopbit(uint64_t v) { + uint64_t cbits = v | 0x7f7f7f7f7f7f7f7fULL; + return ~cbits & (cbits+1); +} + +// A branchless decoder. Credit to Pascal Massimino for the bit-twiddling. +upb_decoderet upb_vdecode_max8_massimino(upb_decoderet r) { + uint64_t b; + memcpy(&b, r.p, sizeof(b)); + uint64_t stop_bit = upb_get_vstopbit(b); + b = (b & 0x7f7f7f7f7f7f7f7fULL) & (stop_bit - 1); + b += b & 0x007f007f007f007fULL; + b += 3 * (b & 0x0000ffff0000ffffULL); + b += 15 * (b & 0x00000000ffffffffULL); + if (stop_bit == 0) { + // Error: unterminated varint. + upb_decoderet err_r = {(void*)0, 0}; + return err_r; + } + upb_decoderet my_r = {r.p + ((__builtin_ctzll(stop_bit) + 1) / 8), + r.val | (b << 7)}; + return my_r; +} + +// A branchless decoder. Credit to Daniel Wright for the bit-twiddling. +upb_decoderet upb_vdecode_max8_wright(upb_decoderet r) { + uint64_t b; + memcpy(&b, r.p, sizeof(b)); + uint64_t stop_bit = upb_get_vstopbit(b); + b &= (stop_bit - 1); + b = ((b & 0x7f007f007f007f00ULL) >> 1) | (b & 0x007f007f007f007fULL); + b = ((b & 0xffff0000ffff0000ULL) >> 2) | (b & 0x0000ffff0000ffffULL); + b = ((b & 0xffffffff00000000ULL) >> 4) | (b & 0x00000000ffffffffULL); + if (stop_bit == 0) { + // Error: unterminated varint. + upb_decoderet err_r = {(void*)0, 0}; + return err_r; + } + upb_decoderet my_r = {r.p + ((__builtin_ctzll(stop_bit) + 1) / 8), + r.val | (b << 14)}; + return my_r; +} + +#line 1 "upb/json/parser.rl" +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2014 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * A parser that uses the Ragel State Machine Compiler to generate + * the finite automata. + * + * Ragel only natively handles regular languages, but we can manually + * program it a bit to handle context-free languages like JSON, by using + * the "fcall" and "fret" constructs. + * + * This parser can handle the basics, but needs several things to be fleshed + * out: + * + * - handling of unicode escape sequences (including high surrogate pairs). + * - properly check and report errors for unknown fields, stack overflow, + * improper array nesting (or lack of nesting). + * - handling of base64 sequences with padding characters. + * - handling of push-back (non-success returns from sink functions). + * - handling of keys/escape-sequences/etc that span input buffers. + */ + +#include +#include +#include +#include +#include +#include + + +#define PARSER_CHECK_RETURN(x) if (!(x)) return false + +// Used to signal that a capture has been suspended. +static char suspend_capture; + +static upb_selector_t getsel_for_handlertype(upb_json_parser *p, + upb_handlertype_t type) { + upb_selector_t sel; + bool ok = upb_handlers_getselector(p->top->f, type, &sel); + UPB_ASSERT_VAR(ok, ok); + return sel; +} + +static upb_selector_t parser_getsel(upb_json_parser *p) { + return getsel_for_handlertype( + p, upb_handlers_getprimitivehandlertype(p->top->f)); +} + +static bool check_stack(upb_json_parser *p) { + if ((p->top + 1) == p->limit) { + upb_status_seterrmsg(p->status, "Nesting too deep"); + return false; + } + + return true; +} + +// There are GCC/Clang built-ins for overflow checking which we could start +// using if there was any performance benefit to it. + +static bool checked_add(size_t a, size_t b, size_t *c) { + if (SIZE_MAX - a < b) return false; + *c = a + b; + return true; +} + +static size_t saturating_multiply(size_t a, size_t b) { + // size_t is unsigned, so this is defined behavior even on overflow. + size_t ret = a * b; + if (b != 0 && ret / b != a) { + ret = SIZE_MAX; + } + return ret; +} + + +/* Base64 decoding ************************************************************/ + +// TODO(haberman): make this streaming. + +static const signed char b64table[] = { + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, 62/*+*/, -1, -1, -1, 63/*/ */, + 52/*0*/, 53/*1*/, 54/*2*/, 55/*3*/, 56/*4*/, 57/*5*/, 58/*6*/, 59/*7*/, + 60/*8*/, 61/*9*/, -1, -1, -1, -1, -1, -1, + -1, 0/*A*/, 1/*B*/, 2/*C*/, 3/*D*/, 4/*E*/, 5/*F*/, 6/*G*/, + 07/*H*/, 8/*I*/, 9/*J*/, 10/*K*/, 11/*L*/, 12/*M*/, 13/*N*/, 14/*O*/, + 15/*P*/, 16/*Q*/, 17/*R*/, 18/*S*/, 19/*T*/, 20/*U*/, 21/*V*/, 22/*W*/, + 23/*X*/, 24/*Y*/, 25/*Z*/, -1, -1, -1, -1, -1, + -1, 26/*a*/, 27/*b*/, 28/*c*/, 29/*d*/, 30/*e*/, 31/*f*/, 32/*g*/, + 33/*h*/, 34/*i*/, 35/*j*/, 36/*k*/, 37/*l*/, 38/*m*/, 39/*n*/, 40/*o*/, + 41/*p*/, 42/*q*/, 43/*r*/, 44/*s*/, 45/*t*/, 46/*u*/, 47/*v*/, 48/*w*/, + 49/*x*/, 50/*y*/, 51/*z*/, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1, + -1, -1, -1, -1, -1, -1, -1, -1 +}; + +// Returns the table value sign-extended to 32 bits. Knowing that the upper +// bits will be 1 for unrecognized characters makes it easier to check for +// this error condition later (see below). +int32_t b64lookup(unsigned char ch) { return b64table[ch]; } + +// Returns true if the given character is not a valid base64 character or +// padding. +bool nonbase64(unsigned char ch) { return b64lookup(ch) == -1 && ch != '='; } + +static bool base64_push(upb_json_parser *p, upb_selector_t sel, const char *ptr, + size_t len) { + const char *limit = ptr + len; + for (; ptr < limit; ptr += 4) { + if (limit - ptr < 4) { + upb_status_seterrf(p->status, + "Base64 input for bytes field not a multiple of 4: %s", + upb_fielddef_name(p->top->f)); + return false; + } + + uint32_t val = b64lookup(ptr[0]) << 18 | + b64lookup(ptr[1]) << 12 | + b64lookup(ptr[2]) << 6 | + b64lookup(ptr[3]); + + // Test the upper bit; returns true if any of the characters returned -1. + if (val & 0x80000000) { + goto otherchar; + } + + char output[3]; + output[0] = val >> 16; + output[1] = (val >> 8) & 0xff; + output[2] = val & 0xff; + upb_sink_putstring(&p->top->sink, sel, output, 3, NULL); + } + return true; + +otherchar: + if (nonbase64(ptr[0]) || nonbase64(ptr[1]) || nonbase64(ptr[2]) || + nonbase64(ptr[3]) ) { + upb_status_seterrf(p->status, + "Non-base64 characters in bytes field: %s", + upb_fielddef_name(p->top->f)); + return false; + } if (ptr[2] == '=') { + // Last group contains only two input bytes, one output byte. + if (ptr[0] == '=' || ptr[1] == '=' || ptr[3] != '=') { + goto badpadding; + } + + uint32_t val = b64lookup(ptr[0]) << 18 | + b64lookup(ptr[1]) << 12; + + assert(!(val & 0x80000000)); + char output = val >> 16; + upb_sink_putstring(&p->top->sink, sel, &output, 1, NULL); + return true; + } else { + // Last group contains only three input bytes, two output bytes. + if (ptr[0] == '=' || ptr[1] == '=' || ptr[2] == '=') { + goto badpadding; + } + + uint32_t val = b64lookup(ptr[0]) << 18 | + b64lookup(ptr[1]) << 12 | + b64lookup(ptr[2]) << 6; + + char output[2]; + output[0] = val >> 16; + output[1] = (val >> 8) & 0xff; + upb_sink_putstring(&p->top->sink, sel, output, 2, NULL); + return true; + } + +badpadding: + upb_status_seterrf(p->status, + "Incorrect base64 padding for field: %s (%.*s)", + upb_fielddef_name(p->top->f), + 4, ptr); + return false; +} + + +/* Accumulate buffer **********************************************************/ + +// Functionality for accumulating a buffer. +// +// Some parts of the parser need an entire value as a contiguous string. For +// example, to look up a member name in a hash table, or to turn a string into +// a number, the relevant library routines need the input string to be in +// contiguous memory, even if the value spanned two or more buffers in the +// input. These routines handle that. +// +// In the common case we can just point to the input buffer to get this +// contiguous string and avoid any actual copy. So we optimistically begin +// this way. But there are a few cases where we must instead copy into a +// separate buffer: +// +// 1. The string was not contiguous in the input (it spanned buffers). +// +// 2. The string included escape sequences that need to be interpreted to get +// the true value in a contiguous buffer. + +static void assert_accumulate_empty(upb_json_parser *p) { + UPB_UNUSED(p); + assert(p->accumulated == NULL); + assert(p->accumulated_len == 0); +} + +static void accumulate_clear(upb_json_parser *p) { + p->accumulated = NULL; + p->accumulated_len = 0; +} + +// Used internally by accumulate_append(). +static bool accumulate_realloc(upb_json_parser *p, size_t need) { + size_t new_size = UPB_MAX(p->accumulate_buf_size, 128); + while (new_size < need) { + new_size = saturating_multiply(new_size, 2); + } + + void *mem = realloc(p->accumulate_buf, new_size); + if (!mem) { + upb_status_seterrmsg(p->status, "Out of memory allocating buffer."); + return false; + } + + p->accumulate_buf = mem; + p->accumulate_buf_size = new_size; + return true; +} + +// Logically appends the given data to the append buffer. +// If "can_alias" is true, we will try to avoid actually copying, but the buffer +// must be valid until the next accumulate_append() call (if any). +static bool accumulate_append(upb_json_parser *p, const char *buf, size_t len, + bool can_alias) { + if (!p->accumulated && can_alias) { + p->accumulated = buf; + p->accumulated_len = len; + return true; + } + + size_t need; + if (!checked_add(p->accumulated_len, len, &need)) { + upb_status_seterrmsg(p->status, "Integer overflow."); + return false; + } + + if (need > p->accumulate_buf_size && !accumulate_realloc(p, need)) { + return false; + } + + if (p->accumulated != p->accumulate_buf) { + memcpy(p->accumulate_buf, p->accumulated, p->accumulated_len); + p->accumulated = p->accumulate_buf; + } + + memcpy(p->accumulate_buf + p->accumulated_len, buf, len); + p->accumulated_len += len; + return true; +} + +// Returns a pointer to the data accumulated since the last accumulate_clear() +// call, and writes the length to *len. This with point either to the input +// buffer or a temporary accumulate buffer. +static const char *accumulate_getptr(upb_json_parser *p, size_t *len) { + assert(p->accumulated); + *len = p->accumulated_len; + return p->accumulated; +} + + +/* Mult-part text data ********************************************************/ + +// When we have text data in the input, it can often come in multiple segments. +// For example, there may be some raw string data followed by an escape +// sequence. The two segments are processed with different logic. Also buffer +// seams in the input can cause multiple segments. +// +// As we see segments, there are two main cases for how we want to process them: +// +// 1. we want to push the captured input directly to string handlers. +// +// 2. we need to accumulate all the parts into a contiguous buffer for further +// processing (field name lookup, string->number conversion, etc). + +// This is the set of states for p->multipart_state. +enum { + // We are not currently processing multipart data. + MULTIPART_INACTIVE = 0, + + // We are processing multipart data by accumulating it into a contiguous + // buffer. + MULTIPART_ACCUMULATE = 1, + + // We are processing multipart data by pushing each part directly to the + // current string handlers. + MULTIPART_PUSHEAGERLY = 2 +}; + +// Start a multi-part text value where we accumulate the data for processing at +// the end. +static void multipart_startaccum(upb_json_parser *p) { + assert_accumulate_empty(p); + assert(p->multipart_state == MULTIPART_INACTIVE); + p->multipart_state = MULTIPART_ACCUMULATE; +} + +// Start a multi-part text value where we immediately push text data to a string +// value with the given selector. +static void multipart_start(upb_json_parser *p, upb_selector_t sel) { + assert_accumulate_empty(p); + assert(p->multipart_state == MULTIPART_INACTIVE); + p->multipart_state = MULTIPART_PUSHEAGERLY; + p->string_selector = sel; +} + +static bool multipart_text(upb_json_parser *p, const char *buf, size_t len, + bool can_alias) { + switch (p->multipart_state) { + case MULTIPART_INACTIVE: + upb_status_seterrmsg( + p->status, "Internal error: unexpected state MULTIPART_INACTIVE"); + return false; + + case MULTIPART_ACCUMULATE: + if (!accumulate_append(p, buf, len, can_alias)) { + return false; + } + break; + + case MULTIPART_PUSHEAGERLY: { + const upb_bufhandle *handle = can_alias ? p->handle : NULL; + upb_sink_putstring(&p->top->sink, p->string_selector, buf, len, handle); + break; + } + } + + return true; +} + +// Note: this invalidates the accumulate buffer! Call only after reading its +// contents. +static void multipart_end(upb_json_parser *p) { + assert(p->multipart_state != MULTIPART_INACTIVE); + p->multipart_state = MULTIPART_INACTIVE; + accumulate_clear(p); +} + + +/* Input capture **************************************************************/ + +// Functionality for capturing a region of the input as text. Gracefully +// handles the case where a buffer seam occurs in the middle of the captured +// region. + +static void capture_begin(upb_json_parser *p, const char *ptr) { + assert(p->multipart_state != MULTIPART_INACTIVE); + assert(p->capture == NULL); + p->capture = ptr; +} + +static bool capture_end(upb_json_parser *p, const char *ptr) { + assert(p->capture); + if (multipart_text(p, p->capture, ptr - p->capture, true)) { + p->capture = NULL; + return true; + } else { + return false; + } +} + +// This is called at the end of each input buffer (ie. when we have hit a +// buffer seam). If we are in the middle of capturing the input, this +// processes the unprocessed capture region. +static void capture_suspend(upb_json_parser *p, const char **ptr) { + if (!p->capture) return; + + if (multipart_text(p, p->capture, *ptr - p->capture, false)) { + // We use this as a signal that we were in the middle of capturing, and + // that capturing should resume at the beginning of the next buffer. + // + // We can't use *ptr here, because we have no guarantee that this pointer + // will be valid when we resume (if the underlying memory is freed, then + // using the pointer at all, even to compare to NULL, is likely undefined + // behavior). + p->capture = &suspend_capture; + } else { + // Need to back up the pointer to the beginning of the capture, since + // we were not able to actually preserve it. + *ptr = p->capture; + } +} + +static void capture_resume(upb_json_parser *p, const char *ptr) { + if (p->capture) { + assert(p->capture == &suspend_capture); + p->capture = ptr; + } +} + + +/* Callbacks from the parser **************************************************/ + +// These are the functions called directly from the parser itself. +// We define these in the same order as their declarations in the parser. + +static char escape_char(char in) { + switch (in) { + case 'r': return '\r'; + case 't': return '\t'; + case 'n': return '\n'; + case 'f': return '\f'; + case 'b': return '\b'; + case '/': return '/'; + case '"': return '"'; + case '\\': return '\\'; + default: + assert(0); + return 'x'; + } +} + +static bool escape(upb_json_parser *p, const char *ptr) { + char ch = escape_char(*ptr); + return multipart_text(p, &ch, 1, false); +} + +static void start_hex(upb_json_parser *p) { + p->digit = 0; +} + +static void hexdigit(upb_json_parser *p, const char *ptr) { + char ch = *ptr; + + p->digit <<= 4; + + if (ch >= '0' && ch <= '9') { + p->digit += (ch - '0'); + } else if (ch >= 'a' && ch <= 'f') { + p->digit += ((ch - 'a') + 10); + } else { + assert(ch >= 'A' && ch <= 'F'); + p->digit += ((ch - 'A') + 10); + } +} + +static bool end_hex(upb_json_parser *p) { + uint32_t codepoint = p->digit; + + // emit the codepoint as UTF-8. + char utf8[3]; // support \u0000 -- \uFFFF -- need only three bytes. + int length = 0; + if (codepoint <= 0x7F) { + utf8[0] = codepoint; + length = 1; + } else if (codepoint <= 0x07FF) { + utf8[1] = (codepoint & 0x3F) | 0x80; + codepoint >>= 6; + utf8[0] = (codepoint & 0x1F) | 0xC0; + length = 2; + } else /* codepoint <= 0xFFFF */ { + utf8[2] = (codepoint & 0x3F) | 0x80; + codepoint >>= 6; + utf8[1] = (codepoint & 0x3F) | 0x80; + codepoint >>= 6; + utf8[0] = (codepoint & 0x0F) | 0xE0; + length = 3; + } + // TODO(haberman): Handle high surrogates: if codepoint is a high surrogate + // we have to wait for the next escape to get the full code point). + + return multipart_text(p, utf8, length, false); +} + +static void start_text(upb_json_parser *p, const char *ptr) { + capture_begin(p, ptr); +} + +static bool end_text(upb_json_parser *p, const char *ptr) { + return capture_end(p, ptr); +} + +static void start_number(upb_json_parser *p, const char *ptr) { + multipart_startaccum(p); + capture_begin(p, ptr); +} + +static bool parse_number(upb_json_parser *p); + +static bool end_number(upb_json_parser *p, const char *ptr) { + if (!capture_end(p, ptr)) { + return false; + } + + return parse_number(p); +} + +static bool parse_number(upb_json_parser *p) { + // strtol() and friends unfortunately do not support specifying the length of + // the input string, so we need to force a copy into a NULL-terminated buffer. + if (!multipart_text(p, "\0", 1, false)) { + return false; + } + + size_t len; + const char *buf = accumulate_getptr(p, &len); + const char *myend = buf + len - 1; // One for NULL. + + char *end; + switch (upb_fielddef_type(p->top->f)) { + case UPB_TYPE_ENUM: + case UPB_TYPE_INT32: { + long val = strtol(p->accumulated, &end, 0); + if (val > INT32_MAX || val < INT32_MIN || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putint32(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_INT64: { + long long val = strtoll(p->accumulated, &end, 0); + if (val > INT64_MAX || val < INT64_MIN || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putint64(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_UINT32: { + unsigned long val = strtoul(p->accumulated, &end, 0); + if (val > UINT32_MAX || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putuint32(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_UINT64: { + unsigned long long val = strtoull(p->accumulated, &end, 0); + if (val > UINT64_MAX || errno == ERANGE || end != myend) + goto err; + else + upb_sink_putuint64(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_DOUBLE: { + double val = strtod(p->accumulated, &end); + if (errno == ERANGE || end != myend) + goto err; + else + upb_sink_putdouble(&p->top->sink, parser_getsel(p), val); + break; + } + case UPB_TYPE_FLOAT: { + float val = strtof(p->accumulated, &end); + if (errno == ERANGE || end != myend) + goto err; + else + upb_sink_putfloat(&p->top->sink, parser_getsel(p), val); + break; + } + default: + assert(false); + } + + multipart_end(p); + + return true; + +err: + upb_status_seterrf(p->status, "error parsing number: %s", buf); + multipart_end(p); + return false; +} + +static bool parser_putbool(upb_json_parser *p, bool val) { + if (upb_fielddef_type(p->top->f) != UPB_TYPE_BOOL) { + upb_status_seterrf(p->status, + "Boolean value specified for non-bool field: %s", + upb_fielddef_name(p->top->f)); + return false; + } + + bool ok = upb_sink_putbool(&p->top->sink, parser_getsel(p), val); + UPB_ASSERT_VAR(ok, ok); + + return true; +} + +static bool start_stringval(upb_json_parser *p) { + assert(p->top->f); + + if (upb_fielddef_isstring(p->top->f)) { + if (!check_stack(p)) return false; + + // Start a new parser frame: parser frames correspond one-to-one with + // handler frames, and string events occur in a sub-frame. + upb_jsonparser_frame *inner = p->top + 1; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSTR); + upb_sink_startstr(&p->top->sink, sel, 0, &inner->sink); + inner->m = p->top->m; + inner->f = p->top->f; + inner->is_map = false; + inner->is_mapentry = false; + p->top = inner; + + if (upb_fielddef_type(p->top->f) == UPB_TYPE_STRING) { + // For STRING fields we push data directly to the handlers as it is + // parsed. We don't do this yet for BYTES fields, because our base64 + // decoder is not streaming. + // + // TODO(haberman): make base64 decoding streaming also. + multipart_start(p, getsel_for_handlertype(p, UPB_HANDLER_STRING)); + return true; + } else { + multipart_startaccum(p); + return true; + } + } else if (upb_fielddef_type(p->top->f) == UPB_TYPE_ENUM) { + // No need to push a frame -- symbolic enum names in quotes remain in the + // current parser frame. + // + // Enum string values must accumulate so we can look up the value in a table + // once it is complete. + multipart_startaccum(p); + return true; + } else { + upb_status_seterrf(p->status, + "String specified for non-string/non-enum field: %s", + upb_fielddef_name(p->top->f)); + return false; + } +} + +static bool end_stringval(upb_json_parser *p) { + bool ok = true; + + switch (upb_fielddef_type(p->top->f)) { + case UPB_TYPE_BYTES: + if (!base64_push(p, getsel_for_handlertype(p, UPB_HANDLER_STRING), + p->accumulated, p->accumulated_len)) { + return false; + } + // Fall through. + + case UPB_TYPE_STRING: { + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSTR); + upb_sink_endstr(&p->top->sink, sel); + p->top--; + break; + } + + case UPB_TYPE_ENUM: { + // Resolve enum symbolic name to integer value. + const upb_enumdef *enumdef = + (const upb_enumdef*)upb_fielddef_subdef(p->top->f); + + size_t len; + const char *buf = accumulate_getptr(p, &len); + + int32_t int_val = 0; + ok = upb_enumdef_ntoi(enumdef, buf, len, &int_val); + + if (ok) { + upb_selector_t sel = parser_getsel(p); + upb_sink_putint32(&p->top->sink, sel, int_val); + } else { + upb_status_seterrf(p->status, "Enum value unknown: '%.*s'", len, buf); + } + + break; + } + + default: + assert(false); + upb_status_seterrmsg(p->status, "Internal error in JSON decoder"); + ok = false; + break; + } + + multipart_end(p); + + return ok; +} + +static void start_member(upb_json_parser *p) { + assert(!p->top->f); + multipart_startaccum(p); +} + +// Helper: invoked during parse_mapentry() to emit the mapentry message's key +// field based on the current contents of the accumulate buffer. +static bool parse_mapentry_key(upb_json_parser *p) { + + size_t len; + const char *buf = accumulate_getptr(p, &len); + + // Emit the key field. We do a bit of ad-hoc parsing here because the + // parser state machine has already decided that this is a string field + // name, and we are reinterpreting it as some arbitrary key type. In + // particular, integer and bool keys are quoted, so we need to parse the + // quoted string contents here. + + p->top->f = upb_msgdef_itof(p->top->m, UPB_MAPENTRY_KEY); + if (p->top->f == NULL) { + upb_status_seterrmsg(p->status, "mapentry message has no key"); + return false; + } + switch (upb_fielddef_type(p->top->f)) { + case UPB_TYPE_INT32: + case UPB_TYPE_INT64: + case UPB_TYPE_UINT32: + case UPB_TYPE_UINT64: + // Invoke end_number. The accum buffer has the number's text already. + if (!parse_number(p)) { + return false; + } + break; + case UPB_TYPE_BOOL: + if (len == 4 && !strncmp(buf, "true", 4)) { + if (!parser_putbool(p, true)) { + return false; + } + } else if (len == 5 && !strncmp(buf, "false", 5)) { + if (!parser_putbool(p, false)) { + return false; + } + } else { + upb_status_seterrmsg(p->status, + "Map bool key not 'true' or 'false'"); + return false; + } + multipart_end(p); + break; + case UPB_TYPE_STRING: + case UPB_TYPE_BYTES: { + upb_sink subsink; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSTR); + upb_sink_startstr(&p->top->sink, sel, len, &subsink); + sel = getsel_for_handlertype(p, UPB_HANDLER_STRING); + upb_sink_putstring(&subsink, sel, buf, len, NULL); + sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSTR); + upb_sink_endstr(&subsink, sel); + multipart_end(p); + break; + } + default: + upb_status_seterrmsg(p->status, "Invalid field type for map key"); + return false; + } + + return true; +} + +// Helper: emit one map entry (as a submessage in the map field sequence). This +// is invoked from end_membername(), at the end of the map entry's key string, +// with the map key in the accumulate buffer. It parses the key from that +// buffer, emits the handler calls to start the mapentry submessage (setting up +// its subframe in the process), and sets up state in the subframe so that the +// value parser (invoked next) will emit the mapentry's value field and then +// end the mapentry message. + +static bool handle_mapentry(upb_json_parser *p) { + // Map entry: p->top->sink is the seq frame, so we need to start a frame + // for the mapentry itself, and then set |f| in that frame so that the map + // value field is parsed, and also set a flag to end the frame after the + // map-entry value is parsed. + if (!check_stack(p)) return false; + + const upb_fielddef *mapfield = p->top->mapfield; + const upb_msgdef *mapentrymsg = upb_fielddef_msgsubdef(mapfield); + + upb_jsonparser_frame *inner = p->top + 1; + p->top->f = mapfield; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSUBMSG); + upb_sink_startsubmsg(&p->top->sink, sel, &inner->sink); + inner->m = mapentrymsg; + inner->mapfield = mapfield; + inner->is_map = false; + + // Don't set this to true *yet* -- we reuse parsing handlers below to push + // the key field value to the sink, and these handlers will pop the frame + // if they see is_mapentry (when invoked by the parser state machine, they + // would have just seen the map-entry value, not key). + inner->is_mapentry = false; + p->top = inner; + + // send STARTMSG in submsg frame. + upb_sink_startmsg(&p->top->sink); + + parse_mapentry_key(p); + + // Set up the value field to receive the map-entry value. + p->top->f = upb_msgdef_itof(p->top->m, UPB_MAPENTRY_VALUE); + p->top->is_mapentry = true; // set up to pop frame after value is parsed. + p->top->mapfield = mapfield; + if (p->top->f == NULL) { + upb_status_seterrmsg(p->status, "mapentry message has no value"); + return false; + } + + return true; +} + +static bool end_membername(upb_json_parser *p) { + assert(!p->top->f); + + if (p->top->is_map) { + return handle_mapentry(p); + } else { + size_t len; + const char *buf = accumulate_getptr(p, &len); + const upb_fielddef *f = upb_msgdef_ntof(p->top->m, buf, len); + + if (!f) { + // TODO(haberman): Ignore unknown fields if requested/configured to do so. + upb_status_seterrf(p->status, "No such field: %.*s\n", (int)len, buf); + return false; + } + + p->top->f = f; + multipart_end(p); + + return true; + } +} + +static void end_member(upb_json_parser *p) { + // If we just parsed a map-entry value, end that frame too. + if (p->top->is_mapentry) { + assert(p->top > p->stack); + // send ENDMSG on submsg. + upb_status s = UPB_STATUS_INIT; + upb_sink_endmsg(&p->top->sink, &s); + const upb_fielddef* mapfield = p->top->mapfield; + + // send ENDSUBMSG in repeated-field-of-mapentries frame. + p->top--; + upb_selector_t sel; + bool ok = upb_handlers_getselector(mapfield, + UPB_HANDLER_ENDSUBMSG, &sel); + UPB_ASSERT_VAR(ok, ok); + upb_sink_endsubmsg(&p->top->sink, sel); + } + + p->top->f = NULL; +} + +static bool start_subobject(upb_json_parser *p) { + assert(p->top->f); + + if (upb_fielddef_ismap(p->top->f)) { + // Beginning of a map. Start a new parser frame in a repeated-field + // context. + if (!check_stack(p)) return false; + + upb_jsonparser_frame *inner = p->top + 1; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSEQ); + upb_sink_startseq(&p->top->sink, sel, &inner->sink); + inner->m = upb_fielddef_msgsubdef(p->top->f); + inner->mapfield = p->top->f; + inner->f = NULL; + inner->is_map = true; + inner->is_mapentry = false; + p->top = inner; + + return true; + } else if (upb_fielddef_issubmsg(p->top->f)) { + // Beginning of a subobject. Start a new parser frame in the submsg + // context. + if (!check_stack(p)) return false; + + upb_jsonparser_frame *inner = p->top + 1; + + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSUBMSG); + upb_sink_startsubmsg(&p->top->sink, sel, &inner->sink); + inner->m = upb_fielddef_msgsubdef(p->top->f); + inner->f = NULL; + inner->is_map = false; + inner->is_mapentry = false; + p->top = inner; + + return true; + } else { + upb_status_seterrf(p->status, + "Object specified for non-message/group field: %s", + upb_fielddef_name(p->top->f)); + return false; + } +} + +static void end_subobject(upb_json_parser *p) { + if (p->top->is_map) { + p->top--; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSEQ); + upb_sink_endseq(&p->top->sink, sel); + } else { + p->top--; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSUBMSG); + upb_sink_endsubmsg(&p->top->sink, sel); + } +} + +static bool start_array(upb_json_parser *p) { + assert(p->top->f); + + if (!upb_fielddef_isseq(p->top->f)) { + upb_status_seterrf(p->status, + "Array specified for non-repeated field: %s", + upb_fielddef_name(p->top->f)); + return false; + } + + if (!check_stack(p)) return false; + + upb_jsonparser_frame *inner = p->top + 1; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_STARTSEQ); + upb_sink_startseq(&p->top->sink, sel, &inner->sink); + inner->m = p->top->m; + inner->f = p->top->f; + inner->is_map = false; + inner->is_mapentry = false; + p->top = inner; + + return true; +} + +static void end_array(upb_json_parser *p) { + assert(p->top > p->stack); + + p->top--; + upb_selector_t sel = getsel_for_handlertype(p, UPB_HANDLER_ENDSEQ); + upb_sink_endseq(&p->top->sink, sel); +} + +static void start_object(upb_json_parser *p) { + if (!p->top->is_map) { + upb_sink_startmsg(&p->top->sink); + } +} + +static void end_object(upb_json_parser *p) { + if (!p->top->is_map) { + upb_status status; + upb_sink_endmsg(&p->top->sink, &status); + } +} + + +#define CHECK_RETURN_TOP(x) if (!(x)) goto error + + +/* The actual parser **********************************************************/ + +// What follows is the Ragel parser itself. The language is specified in Ragel +// and the actions call our C functions above. +// +// Ragel has an extensive set of functionality, and we use only a small part of +// it. There are many action types but we only use a few: +// +// ">" -- transition into a machine +// "%" -- transition out of a machine +// "@" -- transition into a final state of a machine. +// +// "@" transitions are tricky because a machine can transition into a final +// state repeatedly. But in some cases we know this can't happen, for example +// a string which is delimited by a final '"' can only transition into its +// final state once, when the closing '"' is seen. + + +#line 1085 "upb/json/parser.rl" + + + +#line 997 "upb/json/parser.c" +static const char _json_actions[] = { + 0, 1, 0, 1, 2, 1, 3, 1, + 5, 1, 6, 1, 7, 1, 8, 1, + 10, 1, 12, 1, 13, 1, 14, 1, + 15, 1, 16, 1, 17, 1, 21, 1, + 25, 1, 27, 2, 3, 8, 2, 4, + 5, 2, 6, 2, 2, 6, 8, 2, + 11, 9, 2, 13, 15, 2, 14, 15, + 2, 18, 1, 2, 19, 27, 2, 20, + 9, 2, 22, 27, 2, 23, 27, 2, + 24, 27, 2, 26, 27, 3, 14, 11, + 9 +}; + +static const unsigned char _json_key_offsets[] = { + 0, 0, 4, 9, 14, 15, 19, 24, + 29, 34, 38, 42, 45, 48, 50, 54, + 58, 60, 62, 67, 69, 71, 80, 86, + 92, 98, 104, 106, 115, 116, 116, 116, + 121, 126, 131, 132, 133, 134, 135, 135, + 136, 137, 138, 138, 139, 140, 141, 141, + 146, 151, 152, 156, 161, 166, 171, 175, + 175, 178, 178, 178 +}; + +static const char _json_trans_keys[] = { + 32, 123, 9, 13, 32, 34, 125, 9, + 13, 32, 34, 125, 9, 13, 34, 32, + 58, 9, 13, 32, 93, 125, 9, 13, + 32, 44, 125, 9, 13, 32, 44, 125, + 9, 13, 32, 34, 9, 13, 45, 48, + 49, 57, 48, 49, 57, 46, 69, 101, + 48, 57, 69, 101, 48, 57, 43, 45, + 48, 57, 48, 57, 48, 57, 46, 69, + 101, 48, 57, 34, 92, 34, 92, 34, + 47, 92, 98, 102, 110, 114, 116, 117, + 48, 57, 65, 70, 97, 102, 48, 57, + 65, 70, 97, 102, 48, 57, 65, 70, + 97, 102, 48, 57, 65, 70, 97, 102, + 34, 92, 34, 45, 91, 102, 110, 116, + 123, 48, 57, 34, 32, 93, 125, 9, + 13, 32, 44, 93, 9, 13, 32, 93, + 125, 9, 13, 97, 108, 115, 101, 117, + 108, 108, 114, 117, 101, 32, 34, 125, + 9, 13, 32, 34, 125, 9, 13, 34, + 32, 58, 9, 13, 32, 93, 125, 9, + 13, 32, 44, 125, 9, 13, 32, 44, + 125, 9, 13, 32, 34, 9, 13, 32, + 9, 13, 0 +}; + +static const char _json_single_lengths[] = { + 0, 2, 3, 3, 1, 2, 3, 3, + 3, 2, 2, 1, 3, 0, 2, 2, + 0, 0, 3, 2, 2, 9, 0, 0, + 0, 0, 2, 7, 1, 0, 0, 3, + 3, 3, 1, 1, 1, 1, 0, 1, + 1, 1, 0, 1, 1, 1, 0, 3, + 3, 1, 2, 3, 3, 3, 2, 0, + 1, 0, 0, 0 +}; + +static const char _json_range_lengths[] = { + 0, 1, 1, 1, 0, 1, 1, 1, + 1, 1, 1, 1, 0, 1, 1, 1, + 1, 1, 1, 0, 0, 0, 3, 3, + 3, 3, 0, 1, 0, 0, 0, 1, + 1, 1, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 1, + 1, 0, 1, 1, 1, 1, 1, 0, + 1, 0, 0, 0 +}; + +static const short _json_index_offsets[] = { + 0, 0, 4, 9, 14, 16, 20, 25, + 30, 35, 39, 43, 46, 50, 52, 56, + 60, 62, 64, 69, 72, 75, 85, 89, + 93, 97, 101, 104, 113, 115, 116, 117, + 122, 127, 132, 134, 136, 138, 140, 141, + 143, 145, 147, 148, 150, 152, 154, 155, + 160, 165, 167, 171, 176, 181, 186, 190, + 191, 194, 195, 196 +}; + +static const char _json_indicies[] = { + 0, 2, 0, 1, 3, 4, 5, 3, + 1, 6, 7, 8, 6, 1, 9, 1, + 10, 11, 10, 1, 11, 1, 1, 11, + 12, 13, 14, 15, 13, 1, 16, 17, + 8, 16, 1, 17, 7, 17, 1, 18, + 19, 20, 1, 19, 20, 1, 22, 23, + 23, 21, 24, 1, 23, 23, 24, 21, + 25, 25, 26, 1, 26, 1, 26, 21, + 22, 23, 23, 20, 21, 28, 29, 27, + 31, 32, 30, 33, 33, 33, 33, 33, + 33, 33, 33, 34, 1, 35, 35, 35, + 1, 36, 36, 36, 1, 37, 37, 37, + 1, 38, 38, 38, 1, 40, 41, 39, + 42, 43, 44, 45, 46, 47, 48, 43, + 1, 49, 1, 50, 51, 53, 54, 1, + 53, 52, 55, 56, 54, 55, 1, 56, + 1, 1, 56, 52, 57, 1, 58, 1, + 59, 1, 60, 1, 61, 62, 1, 63, + 1, 64, 1, 65, 66, 1, 67, 1, + 68, 1, 69, 70, 71, 72, 70, 1, + 73, 74, 75, 73, 1, 76, 1, 77, + 78, 77, 1, 78, 1, 1, 78, 79, + 80, 81, 82, 80, 1, 83, 84, 75, + 83, 1, 84, 74, 84, 1, 85, 86, + 86, 1, 1, 1, 1, 0 +}; + +static const char _json_trans_targs[] = { + 1, 0, 2, 3, 4, 56, 3, 4, + 56, 5, 5, 6, 7, 8, 9, 56, + 8, 9, 11, 12, 18, 57, 13, 15, + 14, 16, 17, 20, 58, 21, 20, 58, + 21, 19, 22, 23, 24, 25, 26, 20, + 58, 21, 28, 30, 31, 34, 39, 43, + 47, 29, 59, 59, 32, 31, 29, 32, + 33, 35, 36, 37, 38, 59, 40, 41, + 42, 59, 44, 45, 46, 59, 48, 49, + 55, 48, 49, 55, 50, 50, 51, 52, + 53, 54, 55, 53, 54, 59, 56 +}; + +static const char _json_trans_actions[] = { + 0, 0, 0, 21, 77, 53, 0, 47, + 23, 17, 0, 0, 15, 19, 19, 50, + 0, 0, 0, 0, 0, 1, 0, 0, + 0, 0, 0, 3, 13, 0, 0, 35, + 5, 11, 0, 38, 7, 7, 7, 41, + 44, 9, 62, 56, 25, 0, 0, 0, + 31, 29, 33, 59, 15, 0, 27, 0, + 0, 0, 0, 0, 0, 68, 0, 0, + 0, 71, 0, 0, 0, 65, 21, 77, + 53, 0, 47, 23, 17, 0, 0, 15, + 19, 19, 50, 0, 0, 74, 0 +}; + +static const int json_start = 1; +static const int json_first_final = 56; +static const int json_error = 0; + +static const int json_en_number_machine = 10; +static const int json_en_string_machine = 19; +static const int json_en_value_machine = 27; +static const int json_en_main = 1; + + +#line 1088 "upb/json/parser.rl" + +size_t parse(void *closure, const void *hd, const char *buf, size_t size, + const upb_bufhandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + upb_json_parser *parser = closure; + parser->handle = handle; + + // Variables used by Ragel's generated code. + int cs = parser->current_state; + int *stack = parser->parser_stack; + int top = parser->parser_top; + + const char *p = buf; + const char *pe = buf + size; + + capture_resume(parser, buf); + + +#line 1168 "upb/json/parser.c" + { + int _klen; + unsigned int _trans; + const char *_acts; + unsigned int _nacts; + const char *_keys; + + if ( p == pe ) + goto _test_eof; + if ( cs == 0 ) + goto _out; +_resume: + _keys = _json_trans_keys + _json_key_offsets[cs]; + _trans = _json_index_offsets[cs]; + + _klen = _json_single_lengths[cs]; + if ( _klen > 0 ) { + const char *_lower = _keys; + const char *_mid; + const char *_upper = _keys + _klen - 1; + while (1) { + if ( _upper < _lower ) + break; + + _mid = _lower + ((_upper-_lower) >> 1); + if ( (*p) < *_mid ) + _upper = _mid - 1; + else if ( (*p) > *_mid ) + _lower = _mid + 1; + else { + _trans += (unsigned int)(_mid - _keys); + goto _match; + } + } + _keys += _klen; + _trans += _klen; + } + + _klen = _json_range_lengths[cs]; + if ( _klen > 0 ) { + const char *_lower = _keys; + const char *_mid; + const char *_upper = _keys + (_klen<<1) - 2; + while (1) { + if ( _upper < _lower ) + break; + + _mid = _lower + (((_upper-_lower) >> 1) & ~1); + if ( (*p) < _mid[0] ) + _upper = _mid - 2; + else if ( (*p) > _mid[1] ) + _lower = _mid + 2; + else { + _trans += (unsigned int)((_mid - _keys)>>1); + goto _match; + } + } + _trans += _klen; + } + +_match: + _trans = _json_indicies[_trans]; + cs = _json_trans_targs[_trans]; + + if ( _json_trans_actions[_trans] == 0 ) + goto _again; + + _acts = _json_actions + _json_trans_actions[_trans]; + _nacts = (unsigned int) *_acts++; + while ( _nacts-- > 0 ) + { + switch ( *_acts++ ) + { + case 0: +#line 1000 "upb/json/parser.rl" + { p--; {cs = stack[--top]; goto _again;} } + break; + case 1: +#line 1001 "upb/json/parser.rl" + { p--; {stack[top++] = cs; cs = 10; goto _again;} } + break; + case 2: +#line 1005 "upb/json/parser.rl" + { start_text(parser, p); } + break; + case 3: +#line 1006 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_text(parser, p)); } + break; + case 4: +#line 1012 "upb/json/parser.rl" + { start_hex(parser); } + break; + case 5: +#line 1013 "upb/json/parser.rl" + { hexdigit(parser, p); } + break; + case 6: +#line 1014 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_hex(parser)); } + break; + case 7: +#line 1020 "upb/json/parser.rl" + { CHECK_RETURN_TOP(escape(parser, p)); } + break; + case 8: +#line 1026 "upb/json/parser.rl" + { p--; {cs = stack[--top]; goto _again;} } + break; + case 9: +#line 1029 "upb/json/parser.rl" + { {stack[top++] = cs; cs = 19; goto _again;} } + break; + case 10: +#line 1031 "upb/json/parser.rl" + { p--; {stack[top++] = cs; cs = 27; goto _again;} } + break; + case 11: +#line 1036 "upb/json/parser.rl" + { start_member(parser); } + break; + case 12: +#line 1037 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_membername(parser)); } + break; + case 13: +#line 1040 "upb/json/parser.rl" + { end_member(parser); } + break; + case 14: +#line 1046 "upb/json/parser.rl" + { start_object(parser); } + break; + case 15: +#line 1049 "upb/json/parser.rl" + { end_object(parser); } + break; + case 16: +#line 1055 "upb/json/parser.rl" + { CHECK_RETURN_TOP(start_array(parser)); } + break; + case 17: +#line 1059 "upb/json/parser.rl" + { end_array(parser); } + break; + case 18: +#line 1064 "upb/json/parser.rl" + { start_number(parser, p); } + break; + case 19: +#line 1065 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_number(parser, p)); } + break; + case 20: +#line 1067 "upb/json/parser.rl" + { CHECK_RETURN_TOP(start_stringval(parser)); } + break; + case 21: +#line 1068 "upb/json/parser.rl" + { CHECK_RETURN_TOP(end_stringval(parser)); } + break; + case 22: +#line 1070 "upb/json/parser.rl" + { CHECK_RETURN_TOP(parser_putbool(parser, true)); } + break; + case 23: +#line 1072 "upb/json/parser.rl" + { CHECK_RETURN_TOP(parser_putbool(parser, false)); } + break; + case 24: +#line 1074 "upb/json/parser.rl" + { /* null value */ } + break; + case 25: +#line 1076 "upb/json/parser.rl" + { CHECK_RETURN_TOP(start_subobject(parser)); } + break; + case 26: +#line 1077 "upb/json/parser.rl" + { end_subobject(parser); } + break; + case 27: +#line 1082 "upb/json/parser.rl" + { p--; {cs = stack[--top]; goto _again;} } + break; +#line 1354 "upb/json/parser.c" + } + } + +_again: + if ( cs == 0 ) + goto _out; + if ( ++p != pe ) + goto _resume; + _test_eof: {} + _out: {} + } + +#line 1107 "upb/json/parser.rl" + + if (p != pe) { + upb_status_seterrf(parser->status, "Parse error at %s\n", p); + } else { + capture_suspend(parser, &p); + } + +error: + // Save parsing state back to parser. + parser->current_state = cs; + parser->parser_top = top; + + return p - buf; +} + +bool end(void *closure, const void *hd) { + UPB_UNUSED(closure); + UPB_UNUSED(hd); + return true; +} + + +/* Public API *****************************************************************/ + +void upb_json_parser_init(upb_json_parser *p, upb_status *status) { + p->limit = p->stack + UPB_JSON_MAX_DEPTH; + p->accumulate_buf = NULL; + p->accumulate_buf_size = 0; + upb_byteshandler_init(&p->input_handler_); + upb_byteshandler_setstring(&p->input_handler_, parse, NULL); + upb_byteshandler_setendstr(&p->input_handler_, end, NULL); + upb_bytessink_reset(&p->input_, &p->input_handler_, p); + p->status = status; +} + +void upb_json_parser_uninit(upb_json_parser *p) { + upb_byteshandler_uninit(&p->input_handler_); + free(p->accumulate_buf); +} + +void upb_json_parser_reset(upb_json_parser *p) { + p->top = p->stack; + p->top->f = NULL; + p->top->is_map = false; + p->top->is_mapentry = false; + + int cs; + int top; + // Emit Ragel initialization of the parser. + +#line 1418 "upb/json/parser.c" + { + cs = json_start; + top = 0; + } + +#line 1157 "upb/json/parser.rl" + p->current_state = cs; + p->parser_top = top; + accumulate_clear(p); + p->multipart_state = MULTIPART_INACTIVE; + p->capture = NULL; +} + +void upb_json_parser_resetoutput(upb_json_parser *p, upb_sink *sink) { + upb_json_parser_reset(p); + upb_sink_reset(&p->top->sink, sink->handlers, sink->closure); + p->top->m = upb_handlers_msgdef(sink->handlers); + p->accumulated = NULL; +} + +upb_bytessink *upb_json_parser_input(upb_json_parser *p) { + return &p->input_; +} +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2014 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * This currently uses snprintf() to format primitives, and could be optimized + * further. + */ + + +#include +#include +#include +#include + +// StringPiece; a pointer plus a length. +typedef struct { + const char *ptr; + size_t len; +} strpc; + +strpc *newstrpc(upb_handlers *h, const upb_fielddef *f) { + strpc *ret = malloc(sizeof(*ret)); + ret->ptr = upb_fielddef_name(f); + ret->len = strlen(ret->ptr); + upb_handlers_addcleanup(h, ret, free); + return ret; +} + +// ------------ JSON string printing: values, maps, arrays -------------------- + +static void print_data( + upb_json_printer *p, const char *buf, unsigned int len) { + // TODO: Will need to change if we support pushback from the sink. + size_t n = upb_bytessink_putbuf(p->output_, p->subc_, buf, len, NULL); + UPB_ASSERT_VAR(n, n == len); +} + +static void print_comma(upb_json_printer *p) { + if (!p->first_elem_[p->depth_]) { + print_data(p, ",", 1); + } + p->first_elem_[p->depth_] = false; +} + +// Helpers that print properly formatted elements to the JSON output stream. + +// Used for escaping control chars in strings. +static const char kControlCharLimit = 0x20; + +static inline bool is_json_escaped(char c) { + // See RFC 4627. + unsigned char uc = (unsigned char)c; + return uc < kControlCharLimit || uc == '"' || uc == '\\'; +} + +static inline char* json_nice_escape(char c) { + switch (c) { + case '"': return "\\\""; + case '\\': return "\\\\"; + case '\b': return "\\b"; + case '\f': return "\\f"; + case '\n': return "\\n"; + case '\r': return "\\r"; + case '\t': return "\\t"; + default: return NULL; + } +} + +// Write a properly escaped string chunk. The surrounding quotes are *not* +// printed; this is so that the caller has the option of emitting the string +// content in chunks. +static void putstring(upb_json_printer *p, const char *buf, unsigned int len) { + const char* unescaped_run = NULL; + for (unsigned int i = 0; i < len; i++) { + char c = buf[i]; + // Handle escaping. + if (is_json_escaped(c)) { + // Use a "nice" escape, like \n, if one exists for this character. + const char* escape = json_nice_escape(c); + // If we don't have a specific 'nice' escape code, use a \uXXXX-style + // escape. + char escape_buf[8]; + if (!escape) { + unsigned char byte = (unsigned char)c; + snprintf(escape_buf, sizeof(escape_buf), "\\u%04x", (int)byte); + escape = escape_buf; + } + + // N.B. that we assume that the input encoding is equal to the output + // encoding (both UTF-8 for now), so for chars >= 0x20 and != \, ", we + // can simply pass the bytes through. + + // If there's a current run of unescaped chars, print that run first. + if (unescaped_run) { + print_data(p, unescaped_run, &buf[i] - unescaped_run); + unescaped_run = NULL; + } + // Then print the escape code. + print_data(p, escape, strlen(escape)); + } else { + // Add to the current unescaped run of characters. + if (unescaped_run == NULL) { + unescaped_run = &buf[i]; + } + } + } + + // If the string ended in a run of unescaped characters, print that last run. + if (unescaped_run) { + print_data(p, unescaped_run, &buf[len] - unescaped_run); + } +} + +#define CHKLENGTH(x) if (!(x)) return -1; + +// Helpers that format floating point values according to our custom formats. +// Right now we use %.8g and %.17g for float/double, respectively, to match +// proto2::util::JsonFormat's defaults. May want to change this later. + +static size_t fmt_double(double val, char* buf, size_t length) { + size_t n = snprintf(buf, length, "%.17g", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_float(float val, char* buf, size_t length) { + size_t n = snprintf(buf, length, "%.8g", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_bool(bool val, char* buf, size_t length) { + size_t n = snprintf(buf, length, "%s", (val ? "true" : "false")); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_int64(long val, char* buf, size_t length) { + size_t n = snprintf(buf, length, "%ld", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +static size_t fmt_uint64(unsigned long long val, char* buf, size_t length) { + size_t n = snprintf(buf, length, "%llu", val); + CHKLENGTH(n > 0 && n < length); + return n; +} + +// Print a map key given a field name. Called by scalar field handlers and by +// startseq for repeated fields. +static bool putkey(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + const strpc *key = handler_data; + print_comma(p); + print_data(p, "\"", 1); + putstring(p, key->ptr, key->len); + print_data(p, "\":", 2); + return true; +} + +#define CHKFMT(val) if ((val) == -1) return false; +#define CHK(val) if (!(val)) return false; + +#define TYPE_HANDLERS(type, fmt_func) \ + static bool put##type(void *closure, const void *handler_data, type val) { \ + upb_json_printer *p = closure; \ + UPB_UNUSED(handler_data); \ + char data[64]; \ + size_t length = fmt_func(val, data, sizeof(data)); \ + CHKFMT(length); \ + print_data(p, data, length); \ + return true; \ + } \ + static bool scalar_##type(void *closure, const void *handler_data, \ + type val) { \ + CHK(putkey(closure, handler_data)); \ + CHK(put##type(closure, handler_data, val)); \ + return true; \ + } \ + static bool repeated_##type(void *closure, const void *handler_data, \ + type val) { \ + upb_json_printer *p = closure; \ + print_comma(p); \ + CHK(put##type(closure, handler_data, val)); \ + return true; \ + } + +#define TYPE_HANDLERS_MAPKEY(type, fmt_func) \ + static bool putmapkey_##type(void *closure, const void *handler_data, \ + type val) { \ + upb_json_printer *p = closure; \ + print_data(p, "\"", 1); \ + CHK(put##type(closure, handler_data, val)); \ + print_data(p, "\":", 2); \ + return true; \ + } + +TYPE_HANDLERS(double, fmt_double); +TYPE_HANDLERS(float, fmt_float); +TYPE_HANDLERS(bool, fmt_bool); +TYPE_HANDLERS(int32_t, fmt_int64); +TYPE_HANDLERS(uint32_t, fmt_int64); +TYPE_HANDLERS(int64_t, fmt_int64); +TYPE_HANDLERS(uint64_t, fmt_uint64); + +// double and float are not allowed to be map keys. +TYPE_HANDLERS_MAPKEY(bool, fmt_bool); +TYPE_HANDLERS_MAPKEY(int32_t, fmt_int64); +TYPE_HANDLERS_MAPKEY(uint32_t, fmt_int64); +TYPE_HANDLERS_MAPKEY(int64_t, fmt_int64); +TYPE_HANDLERS_MAPKEY(uint64_t, fmt_uint64); + +#undef TYPE_HANDLERS +#undef TYPE_HANDLERS_MAPKEY + +typedef struct { + void *keyname; + const upb_enumdef *enumdef; +} EnumHandlerData; + +static bool scalar_enum(void *closure, const void *handler_data, + int32_t val) { + const EnumHandlerData *hd = handler_data; + upb_json_printer *p = closure; + CHK(putkey(closure, hd->keyname)); + + const char *symbolic_name = upb_enumdef_iton(hd->enumdef, val); + if (symbolic_name) { + print_data(p, "\"", 1); + putstring(p, symbolic_name, strlen(symbolic_name)); + print_data(p, "\"", 1); + } else { + putint32_t(closure, NULL, val); + } + + return true; +} + +static void print_enum_symbolic_name(upb_json_printer *p, + const upb_enumdef *def, + int32_t val) { + const char *symbolic_name = upb_enumdef_iton(def, val); + if (symbolic_name) { + print_data(p, "\"", 1); + putstring(p, symbolic_name, strlen(symbolic_name)); + print_data(p, "\"", 1); + } else { + putint32_t(p, NULL, val); + } +} + +static bool repeated_enum(void *closure, const void *handler_data, + int32_t val) { + const EnumHandlerData *hd = handler_data; + upb_json_printer *p = closure; + print_comma(p); + + print_enum_symbolic_name(p, hd->enumdef, val); + + return true; +} + +static bool mapvalue_enum(void *closure, const void *handler_data, + int32_t val) { + const EnumHandlerData *hd = handler_data; + upb_json_printer *p = closure; + + print_enum_symbolic_name(p, hd->enumdef, val); + + return true; +} + +static void *scalar_startsubmsg(void *closure, const void *handler_data) { + return putkey(closure, handler_data) ? closure : UPB_BREAK; +} + +static void *repeated_startsubmsg(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + print_comma(p); + return closure; +} + +static void start_frame(upb_json_printer *p) { + p->depth_++; + p->first_elem_[p->depth_] = true; + print_data(p, "{", 1); +} + +static void end_frame(upb_json_printer *p) { + print_data(p, "}", 1); + p->depth_--; +} + +static bool printer_startmsg(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + if (p->depth_ == 0) { + upb_bytessink_start(p->output_, 0, &p->subc_); + } + start_frame(p); + return true; +} + +static bool printer_endmsg(void *closure, const void *handler_data, upb_status *s) { + UPB_UNUSED(handler_data); + UPB_UNUSED(s); + upb_json_printer *p = closure; + end_frame(p); + if (p->depth_ == 0) { + upb_bytessink_end(p->output_); + } + return true; +} + +static void *startseq(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + CHK(putkey(closure, handler_data)); + p->depth_++; + p->first_elem_[p->depth_] = true; + print_data(p, "[", 1); + return closure; +} + +static bool endseq(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + print_data(p, "]", 1); + p->depth_--; + return true; +} + +static void *startmap(void *closure, const void *handler_data) { + upb_json_printer *p = closure; + CHK(putkey(closure, handler_data)); + p->depth_++; + p->first_elem_[p->depth_] = true; + print_data(p, "{", 1); + return closure; +} + +static bool endmap(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + print_data(p, "}", 1); + p->depth_--; + return true; +} + +static size_t putstr(void *closure, const void *handler_data, const char *str, + size_t len, const upb_bufhandle *handle) { + UPB_UNUSED(handler_data); + UPB_UNUSED(handle); + upb_json_printer *p = closure; + putstring(p, str, len); + return len; +} + +// This has to Base64 encode the bytes, because JSON has no "bytes" type. +static size_t putbytes(void *closure, const void *handler_data, const char *str, + size_t len, const upb_bufhandle *handle) { + UPB_UNUSED(handler_data); + UPB_UNUSED(handle); + upb_json_printer *p = closure; + + // This is the regular base64, not the "web-safe" version. + static const char base64[] = + "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/"; + + // Base64-encode. + char data[16000]; + const char *limit = data + sizeof(data); + const unsigned char *from = (const unsigned char*)str; + char *to = data; + size_t remaining = len; + while (remaining > 2) { + // TODO(haberman): handle encoded lengths > sizeof(data) + UPB_ASSERT_VAR(limit, (limit - to) >= 4); + + to[0] = base64[from[0] >> 2]; + to[1] = base64[((from[0] & 0x3) << 4) | (from[1] >> 4)]; + to[2] = base64[((from[1] & 0xf) << 2) | (from[2] >> 6)]; + to[3] = base64[from[2] & 0x3f]; + + remaining -= 3; + to += 4; + from += 3; + } + + switch (remaining) { + case 2: + to[0] = base64[from[0] >> 2]; + to[1] = base64[((from[0] & 0x3) << 4) | (from[1] >> 4)]; + to[2] = base64[(from[1] & 0xf) << 2]; + to[3] = '='; + to += 4; + from += 2; + break; + case 1: + to[0] = base64[from[0] >> 2]; + to[1] = base64[((from[0] & 0x3) << 4)]; + to[2] = '='; + to[3] = '='; + to += 4; + from += 1; + break; + } + + size_t bytes = to - data; + print_data(p, "\"", 1); + putstring(p, data, bytes); + print_data(p, "\"", 1); + return len; +} + +static void *scalar_startstr(void *closure, const void *handler_data, + size_t size_hint) { + UPB_UNUSED(handler_data); + UPB_UNUSED(size_hint); + upb_json_printer *p = closure; + CHK(putkey(closure, handler_data)); + print_data(p, "\"", 1); + return p; +} + +static size_t scalar_str(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putstr(closure, handler_data, str, len, handle)); + return len; +} + +static bool scalar_endstr(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + print_data(p, "\"", 1); + return true; +} + +static void *repeated_startstr(void *closure, const void *handler_data, + size_t size_hint) { + UPB_UNUSED(handler_data); + UPB_UNUSED(size_hint); + upb_json_printer *p = closure; + print_comma(p); + print_data(p, "\"", 1); + return p; +} + +static size_t repeated_str(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putstr(closure, handler_data, str, len, handle)); + return len; +} + +static bool repeated_endstr(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + print_data(p, "\"", 1); + return true; +} + +static void *mapkeyval_startstr(void *closure, const void *handler_data, + size_t size_hint) { + UPB_UNUSED(handler_data); + UPB_UNUSED(size_hint); + upb_json_printer *p = closure; + print_data(p, "\"", 1); + return p; +} + +static size_t mapkey_str(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putstr(closure, handler_data, str, len, handle)); + return len; +} + +static bool mapkey_endstr(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + print_data(p, "\":", 2); + return true; +} + +static bool mapvalue_endstr(void *closure, const void *handler_data) { + UPB_UNUSED(handler_data); + upb_json_printer *p = closure; + print_data(p, "\"", 1); + return true; +} + +static size_t scalar_bytes(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + CHK(putkey(closure, handler_data)); + CHK(putbytes(closure, handler_data, str, len, handle)); + return len; +} + +static size_t repeated_bytes(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + upb_json_printer *p = closure; + print_comma(p); + CHK(putbytes(closure, handler_data, str, len, handle)); + return len; +} + +static size_t mapkey_bytes(void *closure, const void *handler_data, + const char *str, size_t len, + const upb_bufhandle *handle) { + upb_json_printer *p = closure; + CHK(putbytes(closure, handler_data, str, len, handle)); + print_data(p, ":", 1); + return len; +} + +static void set_enum_hd(upb_handlers *h, + const upb_fielddef *f, + upb_handlerattr *attr) { + EnumHandlerData *hd = malloc(sizeof(EnumHandlerData)); + hd->enumdef = (const upb_enumdef *)upb_fielddef_subdef(f); + hd->keyname = newstrpc(h, f); + upb_handlers_addcleanup(h, hd, free); + upb_handlerattr_sethandlerdata(attr, hd); +} + +// Set up handlers for a mapentry submessage (i.e., an individual key/value pair +// in a map). +// +// TODO: Handle missing key, missing value, out-of-order key/value, or repeated +// key or value cases properly. The right way to do this is to allocate a +// temporary structure at the start of a mapentry submessage, store key and +// value data in it as key and value handlers are called, and then print the +// key/value pair once at the end of the submessage. If we don't do this, we +// should at least detect the case and throw an error. However, so far all of +// our sources that emit mapentry messages do so canonically (with one key +// field, and then one value field), so this is not a pressing concern at the +// moment. +void printer_sethandlers_mapentry(const void *closure, upb_handlers *h) { + UPB_UNUSED(closure); + const upb_msgdef *md = upb_handlers_msgdef(h); + + // A mapentry message is printed simply as '"key": value'. Rather than + // special-case key and value for every type below, we just handle both + // fields explicitly here. + const upb_fielddef* key_field = upb_msgdef_itof(md, UPB_MAPENTRY_KEY); + const upb_fielddef* value_field = upb_msgdef_itof(md, UPB_MAPENTRY_VALUE); + + upb_handlerattr empty_attr = UPB_HANDLERATTR_INITIALIZER; + + switch (upb_fielddef_type(key_field)) { + case UPB_TYPE_INT32: + upb_handlers_setint32(h, key_field, putmapkey_int32_t, &empty_attr); + break; + case UPB_TYPE_INT64: + upb_handlers_setint64(h, key_field, putmapkey_int64_t, &empty_attr); + break; + case UPB_TYPE_UINT32: + upb_handlers_setuint32(h, key_field, putmapkey_uint32_t, &empty_attr); + break; + case UPB_TYPE_UINT64: + upb_handlers_setuint64(h, key_field, putmapkey_uint64_t, &empty_attr); + break; + case UPB_TYPE_BOOL: + upb_handlers_setbool(h, key_field, putmapkey_bool, &empty_attr); + break; + case UPB_TYPE_STRING: + upb_handlers_setstartstr(h, key_field, mapkeyval_startstr, &empty_attr); + upb_handlers_setstring(h, key_field, mapkey_str, &empty_attr); + upb_handlers_setendstr(h, key_field, mapkey_endstr, &empty_attr); + break; + case UPB_TYPE_BYTES: + upb_handlers_setstring(h, key_field, mapkey_bytes, &empty_attr); + break; + default: + assert(false); + break; + } + + switch (upb_fielddef_type(value_field)) { + case UPB_TYPE_INT32: + upb_handlers_setint32(h, value_field, putint32_t, &empty_attr); + break; + case UPB_TYPE_INT64: + upb_handlers_setint64(h, value_field, putint64_t, &empty_attr); + break; + case UPB_TYPE_UINT32: + upb_handlers_setuint32(h, value_field, putuint32_t, &empty_attr); + break; + case UPB_TYPE_UINT64: + upb_handlers_setuint64(h, value_field, putuint64_t, &empty_attr); + break; + case UPB_TYPE_BOOL: + upb_handlers_setbool(h, value_field, putbool, &empty_attr); + break; + case UPB_TYPE_FLOAT: + upb_handlers_setfloat(h, value_field, putfloat, &empty_attr); + break; + case UPB_TYPE_DOUBLE: + upb_handlers_setdouble(h, value_field, putdouble, &empty_attr); + break; + case UPB_TYPE_STRING: + upb_handlers_setstartstr(h, value_field, mapkeyval_startstr, &empty_attr); + upb_handlers_setstring(h, value_field, putstr, &empty_attr); + upb_handlers_setendstr(h, value_field, mapvalue_endstr, &empty_attr); + break; + case UPB_TYPE_BYTES: + upb_handlers_setstring(h, value_field, putbytes, &empty_attr); + break; + case UPB_TYPE_ENUM: { + upb_handlerattr enum_attr = UPB_HANDLERATTR_INITIALIZER; + set_enum_hd(h, value_field, &enum_attr); + upb_handlers_setint32(h, value_field, mapvalue_enum, &enum_attr); + upb_handlerattr_uninit(&enum_attr); + break; + } + case UPB_TYPE_MESSAGE: + // No handler necessary -- the submsg handlers will print the message + // as appropriate. + break; + } + + upb_handlerattr_uninit(&empty_attr); +} + +void printer_sethandlers(const void *closure, upb_handlers *h) { + UPB_UNUSED(closure); + const upb_msgdef *md = upb_handlers_msgdef(h); + bool is_mapentry = upb_msgdef_mapentry(md); + upb_handlerattr empty_attr = UPB_HANDLERATTR_INITIALIZER; + + if (is_mapentry) { + // mapentry messages are sufficiently different that we handle them + // separately. + printer_sethandlers_mapentry(closure, h); + return; + } + + upb_handlers_setstartmsg(h, printer_startmsg, &empty_attr); + upb_handlers_setendmsg(h, printer_endmsg, &empty_attr); + +#define TYPE(type, name, ctype) \ + case type: \ + if (upb_fielddef_isseq(f)) { \ + upb_handlers_set##name(h, f, repeated_##ctype, &empty_attr); \ + } else { \ + upb_handlers_set##name(h, f, scalar_##ctype, &name_attr); \ + } \ + break; + + upb_msg_field_iter i; + upb_msg_field_begin(&i, md); + for(; !upb_msg_field_done(&i); upb_msg_field_next(&i)) { + const upb_fielddef *f = upb_msg_iter_field(&i); + + upb_handlerattr name_attr = UPB_HANDLERATTR_INITIALIZER; + upb_handlerattr_sethandlerdata(&name_attr, newstrpc(h, f)); + + if (upb_fielddef_ismap(f)) { + upb_handlers_setstartseq(h, f, startmap, &name_attr); + upb_handlers_setendseq(h, f, endmap, &name_attr); + } else if (upb_fielddef_isseq(f)) { + upb_handlers_setstartseq(h, f, startseq, &name_attr); + upb_handlers_setendseq(h, f, endseq, &empty_attr); + } + + switch (upb_fielddef_type(f)) { + TYPE(UPB_TYPE_FLOAT, float, float); + TYPE(UPB_TYPE_DOUBLE, double, double); + TYPE(UPB_TYPE_BOOL, bool, bool); + TYPE(UPB_TYPE_INT32, int32, int32_t); + TYPE(UPB_TYPE_UINT32, uint32, uint32_t); + TYPE(UPB_TYPE_INT64, int64, int64_t); + TYPE(UPB_TYPE_UINT64, uint64, uint64_t); + case UPB_TYPE_ENUM: { + // For now, we always emit symbolic names for enums. We may want an + // option later to control this behavior, but we will wait for a real + // need first. + upb_handlerattr enum_attr = UPB_HANDLERATTR_INITIALIZER; + set_enum_hd(h, f, &enum_attr); + + if (upb_fielddef_isseq(f)) { + upb_handlers_setint32(h, f, repeated_enum, &enum_attr); + } else { + upb_handlers_setint32(h, f, scalar_enum, &enum_attr); + } + + upb_handlerattr_uninit(&enum_attr); + break; + } + case UPB_TYPE_STRING: + if (upb_fielddef_isseq(f)) { + upb_handlers_setstartstr(h, f, repeated_startstr, &empty_attr); + upb_handlers_setstring(h, f, repeated_str, &empty_attr); + upb_handlers_setendstr(h, f, repeated_endstr, &empty_attr); + } else { + upb_handlers_setstartstr(h, f, scalar_startstr, &name_attr); + upb_handlers_setstring(h, f, scalar_str, &empty_attr); + upb_handlers_setendstr(h, f, scalar_endstr, &empty_attr); + } + break; + case UPB_TYPE_BYTES: + // XXX: this doesn't support strings that span buffers yet. The base64 + // encoder will need to be made resumable for this to work properly. + if (upb_fielddef_isseq(f)) { + upb_handlers_setstring(h, f, repeated_bytes, &empty_attr); + } else { + upb_handlers_setstring(h, f, scalar_bytes, &name_attr); + } + break; + case UPB_TYPE_MESSAGE: + if (upb_fielddef_isseq(f)) { + upb_handlers_setstartsubmsg(h, f, repeated_startsubmsg, &name_attr); + } else { + upb_handlers_setstartsubmsg(h, f, scalar_startsubmsg, &name_attr); + } + break; + } + + upb_handlerattr_uninit(&name_attr); + } + + upb_handlerattr_uninit(&empty_attr); +#undef TYPE +} + +/* Public API *****************************************************************/ + +void upb_json_printer_init(upb_json_printer *p, const upb_handlers *h) { + p->output_ = NULL; + p->depth_ = 0; + upb_sink_reset(&p->input_, h, p); +} + +void upb_json_printer_uninit(upb_json_printer *p) { + UPB_UNUSED(p); +} + +void upb_json_printer_reset(upb_json_printer *p) { + p->depth_ = 0; +} + +void upb_json_printer_resetoutput(upb_json_printer *p, upb_bytessink *output) { + upb_json_printer_reset(p); + p->output_ = output; +} + +upb_sink *upb_json_printer_input(upb_json_printer *p) { + return &p->input_; +} + +const upb_handlers *upb_json_printer_newhandlers(const upb_msgdef *md, + const void *owner) { + return upb_handlers_newfrozen(md, owner, printer_sethandlers, NULL); +} diff --git a/ruby/ext/google/protobuf_c/upb.h b/ruby/ext/google/protobuf_c/upb.h new file mode 100644 index 0000000000..8f6d36432d --- /dev/null +++ b/ruby/ext/google/protobuf_c/upb.h @@ -0,0 +1,8050 @@ +// Amalgamated source file +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Defs are upb's internal representation of the constructs that can appear + * in a .proto file: + * + * - upb_msgdef: describes a "message" construct. + * - upb_fielddef: describes a message field. + * - upb_enumdef: describes an enum. + * (TODO: definitions of services). + * + * Like upb_refcounted objects, defs are mutable only until frozen, and are + * only thread-safe once frozen. + * + * This is a mixed C/C++ interface that offers a full API to both languages. + * See the top-level README for more information. + */ + +#ifndef UPB_DEF_H_ +#define UPB_DEF_H_ + +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * A refcounting scheme that supports circular refs. It accomplishes this by + * partitioning the set of objects into groups such that no cycle spans groups; + * we can then reference-count the group as a whole and ignore refs within the + * group. When objects are mutable, these groups are computed very + * conservatively; we group any objects that have ever had a link between them. + * When objects are frozen, we compute strongly-connected components which + * allows us to be precise and only group objects that are actually cyclic. + * + * This is a mixed C/C++ interface that offers a full API to both languages. + * See the top-level README for more information. + */ + +#ifndef UPB_REFCOUNTED_H_ +#define UPB_REFCOUNTED_H_ + +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * This header is INTERNAL-ONLY! Its interfaces are not public or stable! + * This file defines very fast int->upb_value (inttable) and string->upb_value + * (strtable) hash tables. + * + * The table uses chained scatter with Brent's variation (inspired by the Lua + * implementation of hash tables). The hash function for strings is Austin + * Appleby's "MurmurHash." + * + * The inttable uses uintptr_t as its key, which guarantees it can be used to + * store pointers or integers of at least 32 bits (upb isn't really useful on + * systems where sizeof(void*) < 4). + * + * The table must be homogenous (all values of the same type). In debug + * mode, we check this on insert and lookup. + */ + +#ifndef UPB_TABLE_H_ +#define UPB_TABLE_H_ + +#include +#include +#include +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * This file contains shared definitions that are widely used across upb. + * + * This is a mixed C/C++ interface that offers a full API to both languages. + * See the top-level README for more information. + */ + +#ifndef UPB_H_ +#define UPB_H_ + +#include +#include +#include +#include + +// inline if possible, emit standalone code if required. +#ifdef __cplusplus +#define UPB_INLINE inline +#else +#define UPB_INLINE static inline +#endif + +#if __STDC_VERSION__ >= 199901L +#define UPB_C99 +#endif + +#if ((defined(__cplusplus) && __cplusplus >= 201103L) || \ + defined(__GXX_EXPERIMENTAL_CXX0X__)) && !defined(UPB_NO_CXX11) +#define UPB_CXX11 +#endif + +#ifdef UPB_CXX11 +#include +#define UPB_DISALLOW_COPY_AND_ASSIGN(class_name) \ + class_name(const class_name&) = delete; \ + void operator=(const class_name&) = delete; +#define UPB_DISALLOW_POD_OPS(class_name, full_class_name) \ + class_name() = delete; \ + ~class_name() = delete; \ + /* Friend Pointer so it can access base class. */ \ + friend class Pointer; \ + friend class Pointer; \ + UPB_DISALLOW_COPY_AND_ASSIGN(class_name) +#define UPB_ASSERT_STDLAYOUT(type) \ + static_assert(std::is_standard_layout::value, \ + #type " must be standard layout"); +#else // !defined(UPB_CXX11) +#define UPB_DISALLOW_COPY_AND_ASSIGN(class_name) \ + class_name(const class_name&); \ + void operator=(const class_name&); +#define UPB_DISALLOW_POD_OPS(class_name, full_class_name) \ + class_name(); \ + ~class_name(); \ + /* Friend Pointer so it can access base class. */ \ + friend class Pointer; \ + friend class Pointer; \ + UPB_DISALLOW_COPY_AND_ASSIGN(class_name) +#define UPB_ASSERT_STDLAYOUT(type) +#endif + + +#ifdef __cplusplus + +#define UPB_PRIVATE_FOR_CPP private: +#define UPB_DECLARE_TYPE(cppname, cname) typedef cppname cname; +#define UPB_BEGIN_EXTERN_C extern "C" { +#define UPB_END_EXTERN_C } +#define UPB_DEFINE_STRUCT0(cname, members) members; +#define UPB_DEFINE_STRUCT(cname, cbase, members) \ + public: \ + cbase* base() { return &base_; } \ + const cbase* base() const { return &base_; } \ + \ + private: \ + cbase base_; \ + members; +#define UPB_DEFINE_CLASS0(cppname, cppmethods, members) \ + class cppname { \ + cppmethods \ + members \ + }; \ + UPB_ASSERT_STDLAYOUT(cppname); +#define UPB_DEFINE_CLASS1(cppname, cppbase, cppmethods, members) \ + UPB_DEFINE_CLASS0(cppname, cppmethods, members) \ + namespace upb { \ + template <> \ + class Pointer : public PointerBase { \ + public: \ + explicit Pointer(cppname* ptr) : PointerBase(ptr) {} \ + }; \ + template <> \ + class Pointer \ + : public PointerBase { \ + public: \ + explicit Pointer(const cppname* ptr) : PointerBase(ptr) {} \ + }; \ + } +#define UPB_DEFINE_CLASS2(cppname, cppbase, cppbase2, cppmethods, members) \ + UPB_DEFINE_CLASS0(cppname, UPB_QUOTE(cppmethods), members) \ + namespace upb { \ + template <> \ + class Pointer : public PointerBase2 { \ + public: \ + explicit Pointer(cppname* ptr) : PointerBase2(ptr) {} \ + }; \ + template <> \ + class Pointer \ + : public PointerBase2 { \ + public: \ + explicit Pointer(const cppname* ptr) : PointerBase2(ptr) {} \ + }; \ + } + +#else // !defined(__cplusplus) + +#define UPB_PRIVATE_FOR_CPP +#define UPB_DECLARE_TYPE(cppname, cname) \ + struct cname; \ + typedef struct cname cname; +#define UPB_BEGIN_EXTERN_C +#define UPB_END_EXTERN_C +#define UPB_DEFINE_STRUCT0(cname, members) \ + struct cname { \ + members; \ + }; +#define UPB_DEFINE_STRUCT(cname, cbase, members) \ + struct cname { \ + cbase base; \ + members; \ + }; +#define UPB_DEFINE_CLASS0(cppname, cppmethods, members) members +#define UPB_DEFINE_CLASS1(cppname, cppbase, cppmethods, members) members +#define UPB_DEFINE_CLASS2(cppname, cppbase, cppbase2, cppmethods, members) \ + members + +#endif // defined(__cplusplus) + +#ifdef __GNUC__ +#define UPB_NORETURN __attribute__((__noreturn__)) +#else +#define UPB_NORETURN +#endif + +#define UPB_MAX(x, y) ((x) > (y) ? (x) : (y)) +#define UPB_MIN(x, y) ((x) < (y) ? (x) : (y)) + +#define UPB_UNUSED(var) (void)var + +// Code with commas confuses the preprocessor when passed as arguments, whether +// C++ type names with commas (eg. Foo) or code blocks that declare +// variables (ie. int foo, bar). +#define UPB_QUOTE(...) __VA_ARGS__ + +// For asserting something about a variable when the variable is not used for +// anything else. This prevents "unused variable" warnings when compiling in +// debug mode. +#define UPB_ASSERT_VAR(var, predicate) UPB_UNUSED(var); assert(predicate) + +// Generic function type. +typedef void upb_func(); + +/* Casts **********************************************************************/ + +// Upcasts for C. For downcasts see the definitions of the subtypes. +#define UPB_UPCAST(obj) (&(obj)->base) +#define UPB_UPCAST2(obj) UPB_UPCAST(UPB_UPCAST(obj)) + +#ifdef __cplusplus + +// Downcasts for C++. We can't use C++ inheritance directly and maintain +// compatibility with C. So our inheritance is undeclared in C++. +// Specializations of these casting functions are defined for appropriate type +// pairs, and perform the necessary checks. +// +// Example: +// upb::Def* def = <...>; +// upb::MessageDef* = upb::dyn_cast(def); + +namespace upb { + +// Casts to a direct subclass. The caller must know that cast is correct; an +// incorrect cast will throw an assertion failure in debug mode. +template To down_cast(From* f); + +// Casts to a direct subclass. If the class does not actually match the given +// subtype, returns NULL. +template To dyn_cast(From* f); + +// Pointer is a simple wrapper around a T*. It is only constructed for +// upcast() below, and its sole purpose is to be implicitly convertable to T* or +// pointers to base classes, just as a pointer would be in regular C++ if the +// inheritance were directly expressed as C++ inheritance. +template class Pointer; + +// Casts to any base class, or the type itself (ie. can be a no-op). +template inline Pointer upcast(T *f) { return Pointer(f); } + +template +class PointerBase { + public: + explicit PointerBase(T* ptr) : ptr_(ptr) {} + operator T*() { return ptr_; } + operator Base*() { return ptr_->base(); } + + private: + T* ptr_; +}; + +template +class PointerBase2 : public PointerBase { + public: + explicit PointerBase2(T* ptr) : PointerBase(ptr) {} + operator Base2*() { return Pointer(*this); } +}; + +} + +#endif + + +/* upb::reffed_ptr ************************************************************/ + +#ifdef __cplusplus + +#include // For std::swap(). + +namespace upb { + +// Provides RAII semantics for upb refcounted objects. Each reffed_ptr owns a +// ref on whatever object it points to (if any). +template class reffed_ptr { + public: + reffed_ptr() : ptr_(NULL) {} + + // If ref_donor is NULL, takes a new ref, otherwise adopts from ref_donor. + template + reffed_ptr(U* val, const void* ref_donor = NULL) + : ptr_(upb::upcast(val)) { + if (ref_donor) { + assert(ptr_); + ptr_->DonateRef(ref_donor, this); + } else if (ptr_) { + ptr_->Ref(this); + } + } + + template + reffed_ptr(const reffed_ptr& other) + : ptr_(upb::upcast(other.get())) { + if (ptr_) ptr_->Ref(this); + } + + ~reffed_ptr() { if (ptr_) ptr_->Unref(this); } + + template + reffed_ptr& operator=(const reffed_ptr& other) { + reset(other.get()); + return *this; + } + + reffed_ptr& operator=(const reffed_ptr& other) { + reset(other.get()); + return *this; + } + + // TODO(haberman): add C++11 move construction/assignment for greater + // efficiency. + + void swap(reffed_ptr& other) { + if (ptr_ == other.ptr_) { + return; + } + + if (ptr_) ptr_->DonateRef(this, &other); + if (other.ptr_) other.ptr_->DonateRef(&other, this); + std::swap(ptr_, other.ptr_); + } + + T& operator*() const { + assert(ptr_); + return *ptr_; + } + + T* operator->() const { + assert(ptr_); + return ptr_; + } + + T* get() const { return ptr_; } + + // If ref_donor is NULL, takes a new ref, otherwise adopts from ref_donor. + template + void reset(U* ptr = NULL, const void* ref_donor = NULL) { + reffed_ptr(ptr, ref_donor).swap(*this); + } + + template + reffed_ptr down_cast() { + return reffed_ptr(upb::down_cast(get())); + } + + template + reffed_ptr dyn_cast() { + return reffed_ptr(upb::dyn_cast(get())); + } + + // Plain release() is unsafe; if we were the only owner, it would leak the + // object. Instead we provide this: + T* ReleaseTo(const void* new_owner) { + T* ret = NULL; + ptr_->DonateRef(this, new_owner); + std::swap(ret, ptr_); + return ret; + } + + private: + T* ptr_; +}; + +} // namespace upb + +#endif // __cplusplus + + +/* upb::Status ****************************************************************/ + +#ifdef __cplusplus +namespace upb { +class ErrorSpace; +class Status; +} +#endif + +UPB_DECLARE_TYPE(upb::ErrorSpace, upb_errorspace); +UPB_DECLARE_TYPE(upb::Status, upb_status); + +// The maximum length of an error message before it will get truncated. +#define UPB_STATUS_MAX_MESSAGE 128 + +// An error callback function is used to report errors from some component. +// The function can return "true" to indicate that the component should try +// to recover and proceed, but this is not always possible. +typedef bool upb_errcb_t(void *closure, const upb_status* status); + +UPB_DEFINE_CLASS0(upb::ErrorSpace, +, +UPB_DEFINE_STRUCT0(upb_errorspace, + const char *name; + // Should the error message in the status object according to this code. + void (*set_message)(upb_status* status, int code); +)); + +// Object representing a success or failure status. +// It owns no resources and allocates no memory, so it should work +// even in OOM situations. +UPB_DEFINE_CLASS0(upb::Status, + public: + Status(); + + // Returns true if there is no error. + bool ok() const; + + // Optional error space and code, useful if the caller wants to + // programmatically check the specific kind of error. + ErrorSpace* error_space(); + int code() const; + + const char *error_message() const; + + // The error message will be truncated if it is longer than + // UPB_STATUS_MAX_MESSAGE-4. + void SetErrorMessage(const char* msg); + void SetFormattedErrorMessage(const char* fmt, ...); + + // If there is no error message already, this will use the ErrorSpace to + // populate the error message for this code. The caller can still call + // SetErrorMessage() to give a more specific message. + void SetErrorCode(ErrorSpace* space, int code); + + // Resets the status to a successful state with no message. + void Clear(); + + void CopyFrom(const Status& other); + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(Status); +, +UPB_DEFINE_STRUCT0(upb_status, + bool ok_; + + // Specific status code defined by some error space (optional). + int code_; + upb_errorspace *error_space_; + + // Error message; NULL-terminated. + char msg[UPB_STATUS_MAX_MESSAGE]; +)); + +#define UPB_STATUS_INIT {true, 0, NULL, {0}} + +#ifdef __cplusplus +extern "C" { +#endif + +// The returned string is invalidated by any other call into the status. +const char *upb_status_errmsg(const upb_status *status); +bool upb_ok(const upb_status *status); +upb_errorspace *upb_status_errspace(const upb_status *status); +int upb_status_errcode(const upb_status *status); + +// Any of the functions that write to a status object allow status to be NULL, +// to support use cases where the function's caller does not care about the +// status message. +void upb_status_clear(upb_status *status); +void upb_status_seterrmsg(upb_status *status, const char *msg); +void upb_status_seterrf(upb_status *status, const char *fmt, ...); +void upb_status_vseterrf(upb_status *status, const char *fmt, va_list args); +void upb_status_seterrcode(upb_status *status, upb_errorspace *space, int code); +void upb_status_copy(upb_status *to, const upb_status *from); + +#ifdef __cplusplus +} // extern "C" + +namespace upb { + +// C++ Wrappers +inline Status::Status() { Clear(); } +inline bool Status::ok() const { return upb_ok(this); } +inline const char* Status::error_message() const { + return upb_status_errmsg(this); +} +inline void Status::SetErrorMessage(const char* msg) { + upb_status_seterrmsg(this, msg); +} +inline void Status::SetFormattedErrorMessage(const char* fmt, ...) { + va_list args; + va_start(args, fmt); + upb_status_vseterrf(this, fmt, args); + va_end(args); +} +inline void Status::SetErrorCode(ErrorSpace* space, int code) { + upb_status_seterrcode(this, space, code); +} +inline void Status::Clear() { upb_status_clear(this); } +inline void Status::CopyFrom(const Status& other) { + upb_status_copy(this, &other); +} + +} // namespace upb + +#endif + +#endif /* UPB_H_ */ + +#ifdef __cplusplus +extern "C" { +#endif + + +/* upb_value ******************************************************************/ + +// A tagged union (stored untagged inside the table) so that we can check that +// clients calling table accessors are correctly typed without having to have +// an explosion of accessors. +typedef enum { + UPB_CTYPE_INT32 = 1, + UPB_CTYPE_INT64 = 2, + UPB_CTYPE_UINT32 = 3, + UPB_CTYPE_UINT64 = 4, + UPB_CTYPE_BOOL = 5, + UPB_CTYPE_CSTR = 6, + UPB_CTYPE_PTR = 7, + UPB_CTYPE_CONSTPTR = 8, + UPB_CTYPE_FPTR = 9, +} upb_ctype_t; + +typedef union { + int32_t int32; + int64_t int64; + uint64_t uint64; + uint32_t uint32; + bool _bool; + char *cstr; + void *ptr; + const void *constptr; + upb_func *fptr; +} _upb_value; + +typedef struct { + _upb_value val; +#ifndef NDEBUG + // In debug mode we carry the value type around also so we can check accesses + // to be sure the right member is being read. + upb_ctype_t ctype; +#endif +} upb_value; + +#ifdef UPB_C99 +#define UPB_VALUE_INIT(v, member) {.member = v} +#endif +#define UPB__VALUE_INIT_NONE UPB_VALUE_INIT(NULL, ptr) + +#ifdef NDEBUG +#define SET_TYPE(dest, val) UPB_UNUSED(val) +#define UPB_VALUE_INIT_NONE {UPB__VALUE_INIT_NONE} +#else +#define SET_TYPE(dest, val) dest = val +// Non-existent type, all reads will fail. +#define UPB_VALUE_INIT_NONE {UPB__VALUE_INIT_NONE, -1} +#endif + +#define UPB_VALUE_INIT_INT32(v) UPB_VALUE_INIT(v, int32) +#define UPB_VALUE_INIT_INT64(v) UPB_VALUE_INIT(v, int64) +#define UPB_VALUE_INIT_UINT32(v) UPB_VALUE_INIT(v, uint32) +#define UPB_VALUE_INIT_UINT64(v) UPB_VALUE_INIT(v, uint64) +#define UPB_VALUE_INIT_BOOL(v) UPB_VALUE_INIT(v, _bool) +#define UPB_VALUE_INIT_CSTR(v) UPB_VALUE_INIT(v, cstr) +#define UPB_VALUE_INIT_PTR(v) UPB_VALUE_INIT(v, ptr) +#define UPB_VALUE_INIT_CONSTPTR(v) UPB_VALUE_INIT(v, constptr) +#define UPB_VALUE_INIT_FPTR(v) UPB_VALUE_INIT(v, fptr) + +// Like strdup(), which isn't always available since it's not ANSI C. +char *upb_strdup(const char *s); +// Variant that works with a length-delimited rather than NULL-delimited string, +// as supported by strtable. +char *upb_strdup2(const char *s, size_t len); + +UPB_INLINE void _upb_value_setval(upb_value *v, _upb_value val, + upb_ctype_t ctype) { + v->val = val; + SET_TYPE(v->ctype, ctype); +} + +UPB_INLINE upb_value _upb_value_val(_upb_value val, upb_ctype_t ctype) { + upb_value ret; + _upb_value_setval(&ret, val, ctype); + return ret; +} + +// For each value ctype, define the following set of functions: +// +// // Get/set an int32 from a upb_value. +// int32_t upb_value_getint32(upb_value val); +// void upb_value_setint32(upb_value *val, int32_t cval); +// +// // Construct a new upb_value from an int32. +// upb_value upb_value_int32(int32_t val); +#define FUNCS(name, membername, type_t, proto_type) \ + UPB_INLINE void upb_value_set ## name(upb_value *val, type_t cval) { \ + val->val.uint64 = 0; \ + SET_TYPE(val->ctype, proto_type); \ + val->val.membername = cval; \ + } \ + UPB_INLINE upb_value upb_value_ ## name(type_t val) { \ + upb_value ret; \ + upb_value_set ## name(&ret, val); \ + return ret; \ + } \ + UPB_INLINE type_t upb_value_get ## name(upb_value val) { \ + assert(val.ctype == proto_type); \ + return val.val.membername; \ + } + +FUNCS(int32, int32, int32_t, UPB_CTYPE_INT32); +FUNCS(int64, int64, int64_t, UPB_CTYPE_INT64); +FUNCS(uint32, uint32, uint32_t, UPB_CTYPE_UINT32); +FUNCS(uint64, uint64, uint64_t, UPB_CTYPE_UINT64); +FUNCS(bool, _bool, bool, UPB_CTYPE_BOOL); +FUNCS(cstr, cstr, char*, UPB_CTYPE_CSTR); +FUNCS(ptr, ptr, void*, UPB_CTYPE_PTR); +FUNCS(constptr, constptr, const void*, UPB_CTYPE_CONSTPTR); +FUNCS(fptr, fptr, upb_func*, UPB_CTYPE_FPTR); + +#undef FUNCS + + +/* upb_table ******************************************************************/ + +typedef union { + uintptr_t num; + struct { + // We own this. NULL-terminated but may also contain binary data; see + // explicit length below. + // TODO: move the length to the start of the string in order to reduce + // tabkey's size (to one machine word) in a way that supports static + // initialization. + const char *str; + size_t length; + } s; +} upb_tabkey; + +#define UPB_TABKEY_NUM(n) {n} +#ifdef UPB_C99 +// Given that |s| is a string literal, sizeof(s) gives us a +// compile-time-constant strlen(). We must ensure that this works for static +// data initializers. +#define UPB_TABKEY_STR(strval) { .s = { .str = strval, \ + .length = sizeof(strval) - 1 } } +#endif +// TODO(haberman): C++ +#define UPB_TABKEY_NONE {0} + +typedef struct _upb_tabent { + upb_tabkey key; + _upb_value val; + // Internal chaining. This is const so we can create static initializers for + // tables. We cast away const sometimes, but *only* when the containing + // upb_table is known to be non-const. This requires a bit of care, but + // the subtlety is confined to table.c. + const struct _upb_tabent *next; +} upb_tabent; + +typedef struct { + size_t count; // Number of entries in the hash part. + size_t mask; // Mask to turn hash value -> bucket. + upb_ctype_t ctype; // Type of all values. + uint8_t size_lg2; // Size of the hash table part is 2^size_lg2 entries. + + // Hash table entries. + // Making this const isn't entirely accurate; what we really want is for it to + // have the same const-ness as the table it's inside. But there's no way to + // declare that in C. So we have to make it const so that we can statically + // initialize const hash tables. Then we cast away const when we have to. + const upb_tabent *entries; +} upb_table; + +typedef struct { + upb_table t; +} upb_strtable; + +#define UPB_STRTABLE_INIT(count, mask, ctype, size_lg2, entries) \ + {{count, mask, ctype, size_lg2, entries}} + +#define UPB_EMPTY_STRTABLE_INIT(ctype) \ + UPB_STRTABLE_INIT(0, 0, ctype, 0, NULL) + +typedef struct { + upb_table t; // For entries that don't fit in the array part. + const _upb_value *array; // Array part of the table. See const note above. + size_t array_size; // Array part size. + size_t array_count; // Array part number of elements. +} upb_inttable; + +#define UPB_INTTABLE_INIT(count, mask, ctype, size_lg2, ent, a, asize, acount) \ + {{count, mask, ctype, size_lg2, ent}, a, asize, acount} + +#define UPB_EMPTY_INTTABLE_INIT(ctype) \ + UPB_INTTABLE_INIT(0, 0, ctype, 0, NULL, NULL, 0, 0) + +#define UPB_ARRAY_EMPTYVAL -1 +#define UPB_ARRAY_EMPTYENT UPB_VALUE_INIT_INT64(UPB_ARRAY_EMPTYVAL) + +UPB_INLINE size_t upb_table_size(const upb_table *t) { + if (t->size_lg2 == 0) + return 0; + else + return 1 << t->size_lg2; +} + +// Internal-only functions, in .h file only out of necessity. +UPB_INLINE bool upb_tabent_isempty(const upb_tabent *e) { + return e->key.num == 0; +} + +// Used by some of the unit tests for generic hashing functionality. +uint32_t MurmurHash2(const void * key, size_t len, uint32_t seed); + +UPB_INLINE upb_tabkey upb_intkey(uintptr_t key) { + upb_tabkey k; + k.num = key; + return k; +} + +UPB_INLINE uint32_t upb_inthash(uintptr_t key) { + return (uint32_t)key; +} + +static const upb_tabent *upb_getentry(const upb_table *t, uint32_t hash) { + return t->entries + (hash & t->mask); +} + +UPB_INLINE bool upb_arrhas(_upb_value v) { + return v.uint64 != (uint64_t)UPB_ARRAY_EMPTYVAL; +} + +// Initialize and uninitialize a table, respectively. If memory allocation +// failed, false is returned that the table is uninitialized. +bool upb_inttable_init(upb_inttable *table, upb_ctype_t ctype); +bool upb_strtable_init(upb_strtable *table, upb_ctype_t ctype); +void upb_inttable_uninit(upb_inttable *table); +void upb_strtable_uninit(upb_strtable *table); + +// Returns the number of values in the table. +size_t upb_inttable_count(const upb_inttable *t); +UPB_INLINE size_t upb_strtable_count(const upb_strtable *t) { + return t->t.count; +} + +// Inserts the given key into the hashtable with the given value. The key must +// not already exist in the hash table. For string tables, the key must be +// NULL-terminated, and the table will make an internal copy of the key. +// Inttables must not insert a value of UINTPTR_MAX. +// +// If a table resize was required but memory allocation failed, false is +// returned and the table is unchanged. +bool upb_inttable_insert(upb_inttable *t, uintptr_t key, upb_value val); +bool upb_strtable_insert2(upb_strtable *t, const char *key, size_t len, + upb_value val); + +// For NULL-terminated strings. +UPB_INLINE bool upb_strtable_insert(upb_strtable *t, const char *key, + upb_value val) { + return upb_strtable_insert2(t, key, strlen(key), val); +} + +// Looks up key in this table, returning "true" if the key was found. +// If v is non-NULL, copies the value for this key into *v. +bool upb_inttable_lookup(const upb_inttable *t, uintptr_t key, upb_value *v); +bool upb_strtable_lookup2(const upb_strtable *t, const char *key, size_t len, + upb_value *v); + +// For NULL-terminated strings. +UPB_INLINE bool upb_strtable_lookup(const upb_strtable *t, const char *key, + upb_value *v) { + return upb_strtable_lookup2(t, key, strlen(key), v); +} + +// Removes an item from the table. Returns true if the remove was successful, +// and stores the removed item in *val if non-NULL. +bool upb_inttable_remove(upb_inttable *t, uintptr_t key, upb_value *val); +bool upb_strtable_remove2(upb_strtable *t, const char *key, size_t len, + upb_value *val); + +// For NULL-terminated strings. +UPB_INLINE bool upb_strtable_remove(upb_strtable *t, const char *key, + upb_value *v) { + return upb_strtable_remove2(t, key, strlen(key), v); +} + +// Updates an existing entry in an inttable. If the entry does not exist, +// returns false and does nothing. Unlike insert/remove, this does not +// invalidate iterators. +bool upb_inttable_replace(upb_inttable *t, uintptr_t key, upb_value val); + +// Handy routines for treating an inttable like a stack. May not be mixed with +// other insert/remove calls. +bool upb_inttable_push(upb_inttable *t, upb_value val); +upb_value upb_inttable_pop(upb_inttable *t); + +// Convenience routines for inttables with pointer keys. +bool upb_inttable_insertptr(upb_inttable *t, const void *key, upb_value val); +bool upb_inttable_removeptr(upb_inttable *t, const void *key, upb_value *val); +bool upb_inttable_lookupptr( + const upb_inttable *t, const void *key, upb_value *val); + +// Optimizes the table for the current set of entries, for both memory use and +// lookup time. Client should call this after all entries have been inserted; +// inserting more entries is legal, but will likely require a table resize. +void upb_inttable_compact(upb_inttable *t); + +// A special-case inlinable version of the lookup routine for 32-bit integers. +UPB_INLINE bool upb_inttable_lookup32(const upb_inttable *t, uint32_t key, + upb_value *v) { + *v = upb_value_int32(0); // Silence compiler warnings. + if (key < t->array_size) { + _upb_value arrval = t->array[key]; + if (upb_arrhas(arrval)) { + _upb_value_setval(v, arrval, t->t.ctype); + return true; + } else { + return false; + } + } else { + const upb_tabent *e; + if (t->t.entries == NULL) return false; + for (e = upb_getentry(&t->t, upb_inthash(key)); true; e = e->next) { + if ((uint32_t)e->key.num == key) { + _upb_value_setval(v, e->val, t->t.ctype); + return true; + } + if (e->next == NULL) return false; + } + } +} + +// Exposed for testing only. +bool upb_strtable_resize(upb_strtable *t, size_t size_lg2); + +/* Iterators ******************************************************************/ + +// Iterators for int and string tables. We are subject to some kind of unusual +// design constraints: +// +// For high-level languages: +// - we must be able to guarantee that we don't crash or corrupt memory even if +// the program accesses an invalidated iterator. +// +// For C++11 range-based for: +// - iterators must be copyable +// - iterators must be comparable +// - it must be possible to construct an "end" value. +// +// Iteration order is undefined. +// +// Modifying the table invalidates iterators. upb_{str,int}table_done() is +// guaranteed to work even on an invalidated iterator, as long as the table it +// is iterating over has not been freed. Calling next() or accessing data from +// an invalidated iterator yields unspecified elements from the table, but it is +// guaranteed not to crash and to return real table elements (except when done() +// is true). + + +/* upb_strtable_iter **********************************************************/ + +// upb_strtable_iter i; +// upb_strtable_begin(&i, t); +// for(; !upb_strtable_done(&i); upb_strtable_next(&i)) { +// const char *key = upb_strtable_iter_key(&i); +// const upb_value val = upb_strtable_iter_value(&i); +// // ... +// } + +typedef struct { + const upb_strtable *t; + size_t index; +} upb_strtable_iter; + +void upb_strtable_begin(upb_strtable_iter *i, const upb_strtable *t); +void upb_strtable_next(upb_strtable_iter *i); +bool upb_strtable_done(const upb_strtable_iter *i); +const char *upb_strtable_iter_key(upb_strtable_iter *i); +size_t upb_strtable_iter_keylength(upb_strtable_iter *i); +upb_value upb_strtable_iter_value(const upb_strtable_iter *i); +void upb_strtable_iter_setdone(upb_strtable_iter *i); +bool upb_strtable_iter_isequal(const upb_strtable_iter *i1, + const upb_strtable_iter *i2); + + +/* upb_inttable_iter **********************************************************/ + +// upb_inttable_iter i; +// upb_inttable_begin(&i, t); +// for(; !upb_inttable_done(&i); upb_inttable_next(&i)) { +// uintptr_t key = upb_inttable_iter_key(&i); +// upb_value val = upb_inttable_iter_value(&i); +// // ... +// } + +typedef struct { + const upb_inttable *t; + size_t index; + bool array_part; +} upb_inttable_iter; + +void upb_inttable_begin(upb_inttable_iter *i, const upb_inttable *t); +void upb_inttable_next(upb_inttable_iter *i); +bool upb_inttable_done(const upb_inttable_iter *i); +uintptr_t upb_inttable_iter_key(const upb_inttable_iter *i); +upb_value upb_inttable_iter_value(const upb_inttable_iter *i); +void upb_inttable_iter_setdone(upb_inttable_iter *i); +bool upb_inttable_iter_isequal(const upb_inttable_iter *i1, + const upb_inttable_iter *i2); + + +#ifdef __cplusplus +} /* extern "C" */ +#endif + +#endif /* UPB_TABLE_H_ */ + +// Reference tracking will check ref()/unref() operations to make sure the +// ref ownership is correct. Where possible it will also make tools like +// Valgrind attribute ref leaks to the code that took the leaked ref, not +// the code that originally created the object. +// +// Enabling this requires the application to define upb_lock()/upb_unlock() +// functions that acquire/release a global mutex (or #define UPB_THREAD_UNSAFE). +#ifndef NDEBUG +#define UPB_DEBUG_REFS +#endif + +#ifdef __cplusplus +namespace upb { class RefCounted; } +#endif + +UPB_DECLARE_TYPE(upb::RefCounted, upb_refcounted); + +struct upb_refcounted_vtbl; + +UPB_DEFINE_CLASS0(upb::RefCounted, + public: + // Returns true if the given object is frozen. + bool IsFrozen() const; + + // Increases the ref count, the new ref is owned by "owner" which must not + // already own a ref (and should not itself be a refcounted object if the ref + // could possibly be circular; see below). + // Thread-safe iff "this" is frozen. + void Ref(const void *owner) const; + + // Release a ref that was acquired from upb_refcounted_ref() and collects any + // objects it can. + void Unref(const void *owner) const; + + // Moves an existing ref from "from" to "to", without changing the overall + // ref count. DonateRef(foo, NULL, owner) is the same as Ref(foo, owner), + // but "to" may not be NULL. + void DonateRef(const void *from, const void *to) const; + + // Verifies that a ref to the given object is currently held by the given + // owner. Only effective in UPB_DEBUG_REFS builds. + void CheckRef(const void *owner) const; + + private: + UPB_DISALLOW_POD_OPS(RefCounted, upb::RefCounted); +, +UPB_DEFINE_STRUCT0(upb_refcounted, + // A single reference count shared by all objects in the group. + uint32_t *group; + + // A singly-linked list of all objects in the group. + upb_refcounted *next; + + // Table of function pointers for this type. + const struct upb_refcounted_vtbl *vtbl; + + // Maintained only when mutable, this tracks the number of refs (but not + // ref2's) to this object. *group should be the sum of all individual_count + // in the group. + uint32_t individual_count; + + bool is_frozen; + +#ifdef UPB_DEBUG_REFS + upb_inttable *refs; // Maps owner -> trackedref for incoming refs. + upb_inttable *ref2s; // Set of targets for outgoing ref2s. +#endif +)); + +UPB_BEGIN_EXTERN_C // { + +// It is better to use tracked refs when possible, for the extra debugging +// capability. But if this is not possible (because you don't have easy access +// to a stable pointer value that is associated with the ref), you can pass +// UPB_UNTRACKED_REF instead. +extern const void *UPB_UNTRACKED_REF; + +// Native C API. +bool upb_refcounted_isfrozen(const upb_refcounted *r); +void upb_refcounted_ref(const upb_refcounted *r, const void *owner); +void upb_refcounted_unref(const upb_refcounted *r, const void *owner); +void upb_refcounted_donateref( + const upb_refcounted *r, const void *from, const void *to); +void upb_refcounted_checkref(const upb_refcounted *r, const void *owner); + + +// Internal-to-upb Interface /////////////////////////////////////////////////// + +typedef void upb_refcounted_visit(const upb_refcounted *r, + const upb_refcounted *subobj, + void *closure); + +struct upb_refcounted_vtbl { + // Must visit all subobjects that are currently ref'd via upb_refcounted_ref2. + // Must be longjmp()-safe. + void (*visit)(const upb_refcounted *r, upb_refcounted_visit *visit, void *c); + + // Must free the object and release all references to other objects. + void (*free)(upb_refcounted *r); +}; + +// Initializes the refcounted with a single ref for the given owner. Returns +// false if memory could not be allocated. +bool upb_refcounted_init(upb_refcounted *r, + const struct upb_refcounted_vtbl *vtbl, + const void *owner); + +// Adds a ref from one refcounted object to another ("from" must not already +// own a ref). These refs may be circular; cycles will be collected correctly +// (if conservatively). These refs do not need to be freed in from's free() +// function. +void upb_refcounted_ref2(const upb_refcounted *r, upb_refcounted *from); + +// Removes a ref that was acquired from upb_refcounted_ref2(), and collects any +// object it can. This is only necessary when "from" no longer points to "r", +// and not from from's "free" function. +void upb_refcounted_unref2(const upb_refcounted *r, upb_refcounted *from); + +#define upb_ref2(r, from) \ + upb_refcounted_ref2((const upb_refcounted*)r, (upb_refcounted*)from) +#define upb_unref2(r, from) \ + upb_refcounted_unref2((const upb_refcounted*)r, (upb_refcounted*)from) + +// Freezes all mutable object reachable by ref2() refs from the given roots. +// This will split refcounting groups into precise SCC groups, so that +// refcounting of frozen objects can be more aggressive. If memory allocation +// fails, or if more than 2**31 mutable objects are reachable from "roots", or +// if the maximum depth of the graph exceeds "maxdepth", false is returned and +// the objects are unchanged. +// +// After this operation succeeds, the objects are frozen/const, and may not be +// used through non-const pointers. In particular, they may not be passed as +// the second parameter of upb_refcounted_{ref,unref}2(). On the upside, all +// operations on frozen refcounteds are threadsafe, and objects will be freed +// at the precise moment that they become unreachable. +// +// Caller must own refs on each object in the "roots" list. +bool upb_refcounted_freeze(upb_refcounted *const*roots, int n, upb_status *s, + int maxdepth); + +// Shared by all compiled-in refcounted objects. +extern uint32_t static_refcount; + +UPB_END_EXTERN_C // } + +#ifdef UPB_DEBUG_REFS +#define UPB_REFCOUNT_INIT(refs, ref2s) \ + {&static_refcount, NULL, NULL, 0, true, refs, ref2s} +#else +#define UPB_REFCOUNT_INIT(refs, ref2s) {&static_refcount, NULL, NULL, 0, true} +#endif + +#ifdef __cplusplus +// C++ Wrappers. +namespace upb { +inline bool RefCounted::IsFrozen() const { + return upb_refcounted_isfrozen(this); +} +inline void RefCounted::Ref(const void *owner) const { + upb_refcounted_ref(this, owner); +} +inline void RefCounted::Unref(const void *owner) const { + upb_refcounted_unref(this, owner); +} +inline void RefCounted::DonateRef(const void *from, const void *to) const { + upb_refcounted_donateref(this, from, to); +} +inline void RefCounted::CheckRef(const void *owner) const { + upb_refcounted_checkref(this, owner); +} +} // namespace upb +#endif + +#endif // UPB_REFCOUNT_H_ + +#ifdef __cplusplus +#include +#include +#include + +namespace upb { +class Def; +class EnumDef; +class FieldDef; +class MessageDef; +class OneofDef; +} +#endif + +UPB_DECLARE_TYPE(upb::Def, upb_def); +UPB_DECLARE_TYPE(upb::EnumDef, upb_enumdef); +UPB_DECLARE_TYPE(upb::FieldDef, upb_fielddef); +UPB_DECLARE_TYPE(upb::MessageDef, upb_msgdef); +UPB_DECLARE_TYPE(upb::OneofDef, upb_oneofdef); + +// Maximum field number allowed for FieldDefs. This is an inherent limit of the +// protobuf wire format. +#define UPB_MAX_FIELDNUMBER ((1 << 29) - 1) + +// The maximum message depth that the type graph can have. This is a resource +// limit for the C stack since we sometimes need to recursively traverse the +// graph. Cycles are ok; the traversal will stop when it detects a cycle, but +// we must hit the cycle before the maximum depth is reached. +// +// If having a single static limit is too inflexible, we can add another variant +// of Def::Freeze that allows specifying this as a parameter. +#define UPB_MAX_MESSAGE_DEPTH 64 + + +/* upb::Def: base class for defs *********************************************/ + +// All the different kind of defs we support. These correspond 1:1 with +// declarations in a .proto file. +typedef enum { + UPB_DEF_MSG, + UPB_DEF_FIELD, + UPB_DEF_ENUM, + UPB_DEF_ONEOF, + UPB_DEF_SERVICE, // Not yet implemented. + UPB_DEF_ANY = -1, // Wildcard for upb_symtab_get*() +} upb_deftype_t; + +// The base class of all defs. Its base is upb::RefCounted (use upb::upcast() +// to convert). +UPB_DEFINE_CLASS1(upb::Def, upb::RefCounted, + public: + typedef upb_deftype_t Type; + + Def* Dup(const void *owner) const; + + // Functionality from upb::RefCounted. + bool IsFrozen() const; + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void* from, const void* to) const; + void CheckRef(const void* owner) const; + + Type def_type() const; + + // "fullname" is the def's fully-qualified name (eg. foo.bar.Message). + const char *full_name() const; + + // The def must be mutable. Caller retains ownership of fullname. Defs are + // not required to have a name; if a def has no name when it is frozen, it + // will remain an anonymous def. On failure, returns false and details in "s" + // if non-NULL. + bool set_full_name(const char* fullname, upb::Status* s); + bool set_full_name(const std::string &fullname, upb::Status* s); + + // Freezes the given defs; this validates all constraints and marks the defs + // as frozen (read-only). "defs" may not contain any fielddefs, but fields + // of any msgdefs will be frozen. + // + // Symbolic references to sub-types and enum defaults must have already been + // resolved. Any mutable defs reachable from any of "defs" must also be in + // the list; more formally, "defs" must be a transitive closure of mutable + // defs. + // + // After this operation succeeds, the finalized defs must only be accessed + // through a const pointer! + static bool Freeze(Def* const* defs, int n, Status* status); + static bool Freeze(const std::vector& defs, Status* status); + + private: + UPB_DISALLOW_POD_OPS(Def, upb::Def); +, +UPB_DEFINE_STRUCT(upb_def, upb_refcounted, + const char *fullname; + upb_deftype_t type : 8; + // Used as a flag during the def's mutable stage. Must be false unless + // it is currently being used by a function on the stack. This allows + // us to easily determine which defs were passed into the function's + // current invocation. + bool came_from_user; +)); + +#define UPB_DEF_INIT(name, type, refs, ref2s) \ + { UPB_REFCOUNT_INIT(refs, ref2s), name, type, false } + +UPB_BEGIN_EXTERN_C // { + +// Native C API. +upb_def *upb_def_dup(const upb_def *def, const void *owner); + +// From upb_refcounted. +bool upb_def_isfrozen(const upb_def *def); +void upb_def_ref(const upb_def *def, const void *owner); +void upb_def_unref(const upb_def *def, const void *owner); +void upb_def_donateref(const upb_def *def, const void *from, const void *to); +void upb_def_checkref(const upb_def *def, const void *owner); + +upb_deftype_t upb_def_type(const upb_def *d); +const char *upb_def_fullname(const upb_def *d); +bool upb_def_setfullname(upb_def *def, const char *fullname, upb_status *s); +bool upb_def_freeze(upb_def *const *defs, int n, upb_status *s); + +UPB_END_EXTERN_C // } + + +/* upb::Def casts *************************************************************/ + +#ifdef __cplusplus +#define UPB_CPP_CASTS(cname, cpptype) \ + namespace upb { \ + template <> \ + inline cpptype *down_cast(Def * def) { \ + return upb_downcast_##cname##_mutable(def); \ + } \ + template <> \ + inline cpptype *dyn_cast(Def * def) { \ + return upb_dyncast_##cname##_mutable(def); \ + } \ + template <> \ + inline const cpptype *down_cast( \ + const Def *def) { \ + return upb_downcast_##cname(def); \ + } \ + template <> \ + inline const cpptype *dyn_cast(const Def *def) { \ + return upb_dyncast_##cname(def); \ + } \ + template <> \ + inline const cpptype *down_cast(Def * def) { \ + return upb_downcast_##cname(def); \ + } \ + template <> \ + inline const cpptype *dyn_cast(Def * def) { \ + return upb_dyncast_##cname(def); \ + } \ + } // namespace upb +#else +#define UPB_CPP_CASTS(cname, cpptype) +#endif + +// Dynamic casts, for determining if a def is of a particular type at runtime. +// Downcasts, for when some wants to assert that a def is of a particular type. +// These are only checked if we are building debug. +#define UPB_DEF_CASTS(lower, upper, cpptype) \ + UPB_INLINE const upb_##lower *upb_dyncast_##lower(const upb_def *def) { \ + if (upb_def_type(def) != UPB_DEF_##upper) return NULL; \ + return (upb_##lower *)def; \ + } \ + UPB_INLINE const upb_##lower *upb_downcast_##lower(const upb_def *def) { \ + assert(upb_def_type(def) == UPB_DEF_##upper); \ + return (const upb_##lower *)def; \ + } \ + UPB_INLINE upb_##lower *upb_dyncast_##lower##_mutable(upb_def *def) { \ + return (upb_##lower *)upb_dyncast_##lower(def); \ + } \ + UPB_INLINE upb_##lower *upb_downcast_##lower##_mutable(upb_def *def) { \ + return (upb_##lower *)upb_downcast_##lower(def); \ + } \ + UPB_CPP_CASTS(lower, cpptype) + +#define UPB_DEFINE_DEF(cppname, lower, upper, cppmethods, members) \ + UPB_DEFINE_CLASS2(cppname, upb::Def, upb::RefCounted, UPB_QUOTE(cppmethods), \ + members) \ + UPB_DEF_CASTS(lower, upper, cppname) + + +/* upb::FieldDef **************************************************************/ + +// The types a field can have. Note that this list is not identical to the +// types defined in descriptor.proto, which gives INT32 and SINT32 separate +// types (we distinguish the two with the "integer encoding" enum below). +typedef enum { + UPB_TYPE_FLOAT = 1, + UPB_TYPE_DOUBLE = 2, + UPB_TYPE_BOOL = 3, + UPB_TYPE_STRING = 4, + UPB_TYPE_BYTES = 5, + UPB_TYPE_MESSAGE = 6, + UPB_TYPE_ENUM = 7, // Enum values are int32. + UPB_TYPE_INT32 = 8, + UPB_TYPE_UINT32 = 9, + UPB_TYPE_INT64 = 10, + UPB_TYPE_UINT64 = 11, +} upb_fieldtype_t; + +// The repeated-ness of each field; this matches descriptor.proto. +typedef enum { + UPB_LABEL_OPTIONAL = 1, + UPB_LABEL_REQUIRED = 2, + UPB_LABEL_REPEATED = 3, +} upb_label_t; + +// How integers should be encoded in serializations that offer multiple +// integer encoding methods. +typedef enum { + UPB_INTFMT_VARIABLE = 1, + UPB_INTFMT_FIXED = 2, + UPB_INTFMT_ZIGZAG = 3, // Only for signed types (INT32/INT64). +} upb_intfmt_t; + +// Descriptor types, as defined in descriptor.proto. +typedef enum { + UPB_DESCRIPTOR_TYPE_DOUBLE = 1, + UPB_DESCRIPTOR_TYPE_FLOAT = 2, + UPB_DESCRIPTOR_TYPE_INT64 = 3, + UPB_DESCRIPTOR_TYPE_UINT64 = 4, + UPB_DESCRIPTOR_TYPE_INT32 = 5, + UPB_DESCRIPTOR_TYPE_FIXED64 = 6, + UPB_DESCRIPTOR_TYPE_FIXED32 = 7, + UPB_DESCRIPTOR_TYPE_BOOL = 8, + UPB_DESCRIPTOR_TYPE_STRING = 9, + UPB_DESCRIPTOR_TYPE_GROUP = 10, + UPB_DESCRIPTOR_TYPE_MESSAGE = 11, + UPB_DESCRIPTOR_TYPE_BYTES = 12, + UPB_DESCRIPTOR_TYPE_UINT32 = 13, + UPB_DESCRIPTOR_TYPE_ENUM = 14, + UPB_DESCRIPTOR_TYPE_SFIXED32 = 15, + UPB_DESCRIPTOR_TYPE_SFIXED64 = 16, + UPB_DESCRIPTOR_TYPE_SINT32 = 17, + UPB_DESCRIPTOR_TYPE_SINT64 = 18, +} upb_descriptortype_t; + + +// A upb_fielddef describes a single field in a message. It is most often +// found as a part of a upb_msgdef, but can also stand alone to represent +// an extension. +// +// Its base class is upb::Def (use upb::upcast() to convert). +UPB_DEFINE_DEF(upb::FieldDef, fielddef, FIELD, + public: + typedef upb_fieldtype_t Type; + typedef upb_label_t Label; + typedef upb_intfmt_t IntegerFormat; + typedef upb_descriptortype_t DescriptorType; + + // These return true if the given value is a valid member of the enumeration. + static bool CheckType(int32_t val); + static bool CheckLabel(int32_t val); + static bool CheckDescriptorType(int32_t val); + static bool CheckIntegerFormat(int32_t val); + + // These convert to the given enumeration; they require that the value is + // valid. + static Type ConvertType(int32_t val); + static Label ConvertLabel(int32_t val); + static DescriptorType ConvertDescriptorType(int32_t val); + static IntegerFormat ConvertIntegerFormat(int32_t val); + + // Returns NULL if memory allocation failed. + static reffed_ptr New(); + + // Duplicates the given field, returning NULL if memory allocation failed. + // When a fielddef is duplicated, the subdef (if any) is made symbolic if it + // wasn't already. If the subdef is set but has no name (which is possible + // since msgdefs are not required to have a name) the new fielddef's subdef + // will be unset. + FieldDef* Dup(const void* owner) const; + + // Functionality from upb::RefCounted. + bool IsFrozen() const; + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void* from, const void* to) const; + void CheckRef(const void* owner) const; + + // Functionality from upb::Def. + const char* full_name() const; + + bool type_is_set() const; // Whether set_[descriptor_]type() has been called. + Type type() const; // Requires that type_is_set() == true. + Label label() const; // Defaults to UPB_LABEL_OPTIONAL. + const char* name() const; // NULL if uninitialized. + uint32_t number() const; // Returns 0 if uninitialized. + bool is_extension() const; + + // For UPB_TYPE_MESSAGE fields only where is_tag_delimited() == false, + // indicates whether this field should have lazy parsing handlers that yield + // the unparsed string for the submessage. + // + // TODO(haberman): I think we want to move this into a FieldOptions container + // when we add support for custom options (the FieldOptions struct will + // contain both regular FieldOptions like "lazy" *and* custom options). + bool lazy() const; + + // For non-string, non-submessage fields, this indicates whether binary + // protobufs are encoded in packed or non-packed format. + // + // TODO(haberman): see note above about putting options like this into a + // FieldOptions container. + bool packed() const; + + // An integer that can be used as an index into an array of fields for + // whatever message this field belongs to. Guaranteed to be less than + // f->containing_type()->field_count(). May only be accessed once the def has + // been finalized. + int index() const; + + // The MessageDef to which this field belongs. + // + // If this field has been added to a MessageDef, that message can be retrieved + // directly (this is always the case for frozen FieldDefs). + // + // If the field has not yet been added to a MessageDef, you can set the name + // of the containing type symbolically instead. This is mostly useful for + // extensions, where the extension is declared separately from the message. + const MessageDef* containing_type() const; + const char* containing_type_name(); + + // The OneofDef to which this field belongs, or NULL if this field is not part + // of a oneof. + const OneofDef* containing_oneof() const; + + // The field's type according to the enum in descriptor.proto. This is not + // the same as UPB_TYPE_*, because it distinguishes between (for example) + // INT32 and SINT32, whereas our "type" enum does not. This return of + // descriptor_type() is a function of type(), integer_format(), and + // is_tag_delimited(). Likewise set_descriptor_type() sets all three + // appropriately. + DescriptorType descriptor_type() const; + + // Convenient field type tests. + bool IsSubMessage() const; + bool IsString() const; + bool IsSequence() const; + bool IsPrimitive() const; + bool IsMap() const; + + // How integers are encoded. Only meaningful for integer types. + // Defaults to UPB_INTFMT_VARIABLE, and is reset when "type" changes. + IntegerFormat integer_format() const; + + // Whether a submessage field is tag-delimited or not (if false, then + // length-delimited). May only be set when type() == UPB_TYPE_MESSAGE. + bool is_tag_delimited() const; + + // Returns the non-string default value for this fielddef, which may either + // be something the client set explicitly or the "default default" (0 for + // numbers, empty for strings). The field's type indicates the type of the + // returned value, except for enum fields that are still mutable. + // + // Requires that the given function matches the field's current type. + int64_t default_int64() const; + int32_t default_int32() const; + uint64_t default_uint64() const; + uint32_t default_uint32() const; + bool default_bool() const; + float default_float() const; + double default_double() const; + + // The resulting string is always NULL-terminated. If non-NULL, the length + // will be stored in *len. + const char *default_string(size_t* len) const; + + // For frozen UPB_TYPE_ENUM fields, enum defaults can always be read as either + // string or int32, and both of these methods will always return true. + // + // For mutable UPB_TYPE_ENUM fields, the story is a bit more complicated. + // Enum defaults are unusual. They can be specified either as string or int32, + // but to be valid the enum must have that value as a member. And if no + // default is specified, the "default default" comes from the EnumDef. + // + // We allow reading the default as either an int32 or a string, but only if + // we have a meaningful value to report. We have a meaningful value if it was + // set explicitly, or if we could get the "default default" from the EnumDef. + // Also if you explicitly set the name and we find the number in the EnumDef + bool EnumHasStringDefault() const; + bool EnumHasInt32Default() const; + + // Submessage and enum fields must reference a "subdef", which is the + // upb::MessageDef or upb::EnumDef that defines their type. Note that when + // the FieldDef is mutable it may not have a subdef *yet*, but this function + // still returns true to indicate that the field's type requires a subdef. + bool HasSubDef() const; + + // Returns the enum or submessage def for this field, if any. The field's + // type must match (ie. you may only call enum_subdef() for fields where + // type() == UPB_TYPE_ENUM). Returns NULL if the subdef has not been set or + // is currently set symbolically. + const EnumDef* enum_subdef() const; + const MessageDef* message_subdef() const; + + // Returns the generic subdef for this field. Requires that HasSubDef() (ie. + // only works for UPB_TYPE_ENUM and UPB_TYPE_MESSAGE fields). + const Def* subdef() const; + + // Returns the symbolic name of the subdef. If the subdef is currently set + // unresolved (ie. set symbolically) returns the symbolic name. If it has + // been resolved to a specific subdef, returns the name from that subdef. + const char* subdef_name() const; + + ////////////////////////////////////////////////////////////////////////////// + // Setters (non-const methods), only valid for mutable FieldDefs! + ////////////////////////////////////////////////////////////////////////////// + + bool set_full_name(const char* fullname, upb::Status* s); + bool set_full_name(const std::string& fullname, upb::Status* s); + + // This may only be called if containing_type() == NULL (ie. the field has not + // been added to a message yet). + bool set_containing_type_name(const char *name, Status* status); + bool set_containing_type_name(const std::string& name, Status* status); + + // Defaults to false. When we freeze, we ensure that this can only be true + // for length-delimited message fields. Prior to freezing this can be true or + // false with no restrictions. + void set_lazy(bool lazy); + + // Defaults to true. Sets whether this field is encoded in packed format. + void set_packed(bool packed); + + // "type" or "descriptor_type" MUST be set explicitly before the fielddef is + // finalized. These setters require that the enum value is valid; if the + // value did not come directly from an enum constant, the caller should + // validate it first with the functions above (CheckFieldType(), etc). + void set_type(Type type); + void set_label(Label label); + void set_descriptor_type(DescriptorType type); + void set_is_extension(bool is_extension); + + // "number" and "name" must be set before the FieldDef is added to a + // MessageDef, and may not be set after that. + // + // "name" is the same as full_name()/set_full_name(), but since fielddefs + // most often use simple, non-qualified names, we provide this accessor + // also. Generally only extensions will want to think of this name as + // fully-qualified. + bool set_number(uint32_t number, upb::Status* s); + bool set_name(const char* name, upb::Status* s); + bool set_name(const std::string& name, upb::Status* s); + + void set_integer_format(IntegerFormat format); + bool set_tag_delimited(bool tag_delimited, upb::Status* s); + + // Sets default value for the field. The call must exactly match the type + // of the field. Enum fields may use either setint32 or setstring to set + // the default numerically or symbolically, respectively, but symbolic + // defaults must be resolved before finalizing (see ResolveEnumDefault()). + // + // Changing the type of a field will reset its default. + void set_default_int64(int64_t val); + void set_default_int32(int32_t val); + void set_default_uint64(uint64_t val); + void set_default_uint32(uint32_t val); + void set_default_bool(bool val); + void set_default_float(float val); + void set_default_double(double val); + bool set_default_string(const void *str, size_t len, Status *s); + bool set_default_string(const std::string &str, Status *s); + void set_default_cstr(const char *str, Status *s); + + // Before a fielddef is frozen, its subdef may be set either directly (with a + // upb::Def*) or symbolically. Symbolic refs must be resolved before the + // containing msgdef can be frozen (see upb_resolve() above). upb always + // guarantees that any def reachable from a live def will also be kept alive. + // + // Both methods require that upb_hassubdef(f) (so the type must be set prior + // to calling these methods). Returns false if this is not the case, or if + // the given subdef is not of the correct type. The subdef is reset if the + // field's type is changed. The subdef can be set to NULL to clear it. + bool set_subdef(const Def* subdef, Status* s); + bool set_enum_subdef(const EnumDef* subdef, Status* s); + bool set_message_subdef(const MessageDef* subdef, Status* s); + bool set_subdef_name(const char* name, Status* s); + bool set_subdef_name(const std::string &name, Status* s); + + private: + UPB_DISALLOW_POD_OPS(FieldDef, upb::FieldDef); +, +UPB_DEFINE_STRUCT(upb_fielddef, upb_def, + union { + int64_t sint; + uint64_t uint; + double dbl; + float flt; + void *bytes; + } defaultval; + union { + const upb_msgdef *def; // If !msg_is_symbolic. + char *name; // If msg_is_symbolic. + } msg; + union { + const upb_def *def; // If !subdef_is_symbolic. + char *name; // If subdef_is_symbolic. + } sub; // The msgdef or enumdef for this field, if upb_hassubdef(f). + bool subdef_is_symbolic; + bool msg_is_symbolic; + const upb_oneofdef *oneof; + bool default_is_string; + bool type_is_set_; // False until type is explicitly set. + bool is_extension_; + bool lazy_; + bool packed_; + upb_intfmt_t intfmt; + bool tagdelim; + upb_fieldtype_t type_; + upb_label_t label_; + uint32_t number_; + uint32_t selector_base; // Used to index into a upb::Handlers table. + uint32_t index_; +)); + +#define UPB_FIELDDEF_INIT(label, type, intfmt, tagdelim, is_extension, lazy, \ + packed, name, num, msgdef, subdef, selector_base, \ + index, defaultval, refs, ref2s) \ + { \ + UPB_DEF_INIT(name, UPB_DEF_FIELD, refs, ref2s), defaultval, {msgdef}, \ + {subdef}, NULL, false, false, \ + type == UPB_TYPE_STRING || type == UPB_TYPE_BYTES, true, is_extension, \ + lazy, packed, intfmt, tagdelim, type, label, num, selector_base, index \ + } + +UPB_BEGIN_EXTERN_C // { + +// Native C API. +upb_fielddef *upb_fielddef_new(const void *owner); +upb_fielddef *upb_fielddef_dup(const upb_fielddef *f, const void *owner); + +// From upb_refcounted. +bool upb_fielddef_isfrozen(const upb_fielddef *f); +void upb_fielddef_ref(const upb_fielddef *f, const void *owner); +void upb_fielddef_unref(const upb_fielddef *f, const void *owner); +void upb_fielddef_donateref(const upb_fielddef *f, const void *from, + const void *to); +void upb_fielddef_checkref(const upb_fielddef *f, const void *owner); + +// From upb_def. +const char *upb_fielddef_fullname(const upb_fielddef *f); +bool upb_fielddef_setfullname(upb_fielddef *f, const char *fullname, + upb_status *s); + +bool upb_fielddef_typeisset(const upb_fielddef *f); +upb_fieldtype_t upb_fielddef_type(const upb_fielddef *f); +upb_descriptortype_t upb_fielddef_descriptortype(const upb_fielddef *f); +upb_label_t upb_fielddef_label(const upb_fielddef *f); +uint32_t upb_fielddef_number(const upb_fielddef *f); +const char *upb_fielddef_name(const upb_fielddef *f); +bool upb_fielddef_isextension(const upb_fielddef *f); +bool upb_fielddef_lazy(const upb_fielddef *f); +bool upb_fielddef_packed(const upb_fielddef *f); +const upb_msgdef *upb_fielddef_containingtype(const upb_fielddef *f); +const upb_oneofdef *upb_fielddef_containingoneof(const upb_fielddef *f); +upb_msgdef *upb_fielddef_containingtype_mutable(upb_fielddef *f); +const char *upb_fielddef_containingtypename(upb_fielddef *f); +upb_intfmt_t upb_fielddef_intfmt(const upb_fielddef *f); +uint32_t upb_fielddef_index(const upb_fielddef *f); +bool upb_fielddef_istagdelim(const upb_fielddef *f); +bool upb_fielddef_issubmsg(const upb_fielddef *f); +bool upb_fielddef_isstring(const upb_fielddef *f); +bool upb_fielddef_isseq(const upb_fielddef *f); +bool upb_fielddef_isprimitive(const upb_fielddef *f); +bool upb_fielddef_ismap(const upb_fielddef *f); +int64_t upb_fielddef_defaultint64(const upb_fielddef *f); +int32_t upb_fielddef_defaultint32(const upb_fielddef *f); +uint64_t upb_fielddef_defaultuint64(const upb_fielddef *f); +uint32_t upb_fielddef_defaultuint32(const upb_fielddef *f); +bool upb_fielddef_defaultbool(const upb_fielddef *f); +float upb_fielddef_defaultfloat(const upb_fielddef *f); +double upb_fielddef_defaultdouble(const upb_fielddef *f); +const char *upb_fielddef_defaultstr(const upb_fielddef *f, size_t *len); +bool upb_fielddef_enumhasdefaultint32(const upb_fielddef *f); +bool upb_fielddef_enumhasdefaultstr(const upb_fielddef *f); +bool upb_fielddef_hassubdef(const upb_fielddef *f); +const upb_def *upb_fielddef_subdef(const upb_fielddef *f); +const upb_msgdef *upb_fielddef_msgsubdef(const upb_fielddef *f); +const upb_enumdef *upb_fielddef_enumsubdef(const upb_fielddef *f); +const char *upb_fielddef_subdefname(const upb_fielddef *f); + +void upb_fielddef_settype(upb_fielddef *f, upb_fieldtype_t type); +void upb_fielddef_setdescriptortype(upb_fielddef *f, int type); +void upb_fielddef_setlabel(upb_fielddef *f, upb_label_t label); +bool upb_fielddef_setnumber(upb_fielddef *f, uint32_t number, upb_status *s); +bool upb_fielddef_setname(upb_fielddef *f, const char *name, upb_status *s); +bool upb_fielddef_setcontainingtypename(upb_fielddef *f, const char *name, + upb_status *s); +void upb_fielddef_setisextension(upb_fielddef *f, bool is_extension); +void upb_fielddef_setlazy(upb_fielddef *f, bool lazy); +void upb_fielddef_setpacked(upb_fielddef *f, bool packed); +void upb_fielddef_setintfmt(upb_fielddef *f, upb_intfmt_t fmt); +void upb_fielddef_settagdelim(upb_fielddef *f, bool tag_delim); +void upb_fielddef_setdefaultint64(upb_fielddef *f, int64_t val); +void upb_fielddef_setdefaultint32(upb_fielddef *f, int32_t val); +void upb_fielddef_setdefaultuint64(upb_fielddef *f, uint64_t val); +void upb_fielddef_setdefaultuint32(upb_fielddef *f, uint32_t val); +void upb_fielddef_setdefaultbool(upb_fielddef *f, bool val); +void upb_fielddef_setdefaultfloat(upb_fielddef *f, float val); +void upb_fielddef_setdefaultdouble(upb_fielddef *f, double val); +bool upb_fielddef_setdefaultstr(upb_fielddef *f, const void *str, size_t len, + upb_status *s); +void upb_fielddef_setdefaultcstr(upb_fielddef *f, const char *str, + upb_status *s); +bool upb_fielddef_setsubdef(upb_fielddef *f, const upb_def *subdef, + upb_status *s); +bool upb_fielddef_setmsgsubdef(upb_fielddef *f, const upb_msgdef *subdef, + upb_status *s); +bool upb_fielddef_setenumsubdef(upb_fielddef *f, const upb_enumdef *subdef, + upb_status *s); +bool upb_fielddef_setsubdefname(upb_fielddef *f, const char *name, + upb_status *s); + +bool upb_fielddef_checklabel(int32_t label); +bool upb_fielddef_checktype(int32_t type); +bool upb_fielddef_checkdescriptortype(int32_t type); +bool upb_fielddef_checkintfmt(int32_t fmt); + +UPB_END_EXTERN_C // } + + +/* upb::MessageDef ************************************************************/ + +typedef upb_inttable_iter upb_msg_field_iter; +typedef upb_strtable_iter upb_msg_oneof_iter; + +// Structure that describes a single .proto message type. +// +// Its base class is upb::Def (use upb::upcast() to convert). +UPB_DEFINE_DEF(upb::MessageDef, msgdef, MSG, UPB_QUOTE( + public: + // Returns NULL if memory allocation failed. + static reffed_ptr New(); + + // Functionality from upb::RefCounted. + bool IsFrozen() const; + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void* from, const void* to) const; + void CheckRef(const void* owner) const; + + // Functionality from upb::Def. + const char* full_name() const; + bool set_full_name(const char* fullname, Status* s); + bool set_full_name(const std::string& fullname, Status* s); + + // Call to freeze this MessageDef. + // WARNING: this will fail if this message has any unfrozen submessages! + // Messages with cycles must be frozen as a batch using upb::Def::Freeze(). + bool Freeze(Status* s); + + // The number of fields that belong to the MessageDef. + int field_count() const; + + // The number of oneofs that belong to the MessageDef. + int oneof_count() const; + + // Adds a field (upb_fielddef object) to a msgdef. Requires that the msgdef + // and the fielddefs are mutable. The fielddef's name and number must be + // set, and the message may not already contain any field with this name or + // number, and this fielddef may not be part of another message. In error + // cases false is returned and the msgdef is unchanged. + // + // If the given field is part of a oneof, this call succeeds if and only if + // that oneof is already part of this msgdef. (Note that adding a oneof to a + // msgdef automatically adds all of its fields to the msgdef at the time that + // the oneof is added, so it is usually more idiomatic to add the oneof's + // fields first then add the oneof to the msgdef. This case is supported for + // convenience.) + // + // If |f| is already part of this MessageDef, this method performs no action + // and returns true (success). Thus, this method is idempotent. + bool AddField(FieldDef* f, Status* s); + bool AddField(const reffed_ptr& f, Status* s); + + // Adds a oneof (upb_oneofdef object) to a msgdef. Requires that the msgdef, + // oneof, and any fielddefs are mutable, that the fielddefs contained in the + // oneof do not have any name or number conflicts with existing fields in the + // msgdef, and that the oneof's name is unique among all oneofs in the msgdef. + // If the oneof is added successfully, all of its fields will be added + // directly to the msgdef as well. In error cases, false is returned and the + // msgdef is unchanged. + bool AddOneof(OneofDef* o, Status* s); + bool AddOneof(const reffed_ptr& o, Status* s); + + // These return NULL if the field is not found. + FieldDef* FindFieldByNumber(uint32_t number); + FieldDef* FindFieldByName(const char *name, size_t len); + const FieldDef* FindFieldByNumber(uint32_t number) const; + const FieldDef* FindFieldByName(const char* name, size_t len) const; + + + FieldDef* FindFieldByName(const char *name) { + return FindFieldByName(name, strlen(name)); + } + const FieldDef* FindFieldByName(const char *name) const { + return FindFieldByName(name, strlen(name)); + } + + template + FieldDef* FindFieldByName(const T& str) { + return FindFieldByName(str.c_str(), str.size()); + } + template + const FieldDef* FindFieldByName(const T& str) const { + return FindFieldByName(str.c_str(), str.size()); + } + + OneofDef* FindOneofByName(const char* name, size_t len); + const OneofDef* FindOneofByName(const char* name, size_t len) const; + + OneofDef* FindOneofByName(const char* name) { + return FindOneofByName(name, strlen(name)); + } + const OneofDef* FindOneofByName(const char* name) const { + return FindOneofByName(name, strlen(name)); + } + + template + OneofDef* FindOneofByName(const T& str) { + return FindOneofByName(str.c_str(), str.size()); + } + template + const OneofDef* FindOneofByName(const T& str) const { + return FindOneofByName(str.c_str(), str.size()); + } + + // Returns a new msgdef that is a copy of the given msgdef (and a copy of all + // the fields) but with any references to submessages broken and replaced + // with just the name of the submessage. Returns NULL if memory allocation + // failed. + // + // TODO(haberman): which is more useful, keeping fields resolved or + // unresolving them? If there's no obvious answer, Should this functionality + // just be moved into symtab.c? + MessageDef* Dup(const void* owner) const; + + // Is this message a map entry? + void setmapentry(bool map_entry); + bool mapentry() const; + + // Iteration over fields. The order is undefined. + class field_iterator + : public std::iterator { + public: + explicit field_iterator(MessageDef* md); + static field_iterator end(MessageDef* md); + + void operator++(); + FieldDef* operator*() const; + bool operator!=(const field_iterator& other) const; + bool operator==(const field_iterator& other) const; + + private: + upb_msg_field_iter iter_; + }; + + class const_field_iterator + : public std::iterator { + public: + explicit const_field_iterator(const MessageDef* md); + static const_field_iterator end(const MessageDef* md); + + void operator++(); + const FieldDef* operator*() const; + bool operator!=(const const_field_iterator& other) const; + bool operator==(const const_field_iterator& other) const; + + private: + upb_msg_field_iter iter_; + }; + + // Iteration over oneofs. The order is undefined. + class oneof_iterator + : public std::iterator { + public: + explicit oneof_iterator(MessageDef* md); + static oneof_iterator end(MessageDef* md); + + void operator++(); + OneofDef* operator*() const; + bool operator!=(const oneof_iterator& other) const; + bool operator==(const oneof_iterator& other) const; + + private: + upb_msg_oneof_iter iter_; + }; + + class const_oneof_iterator + : public std::iterator { + public: + explicit const_oneof_iterator(const MessageDef* md); + static const_oneof_iterator end(const MessageDef* md); + + void operator++(); + const OneofDef* operator*() const; + bool operator!=(const const_oneof_iterator& other) const; + bool operator==(const const_oneof_iterator& other) const; + + private: + upb_msg_oneof_iter iter_; + }; + + class FieldAccessor { + public: + explicit FieldAccessor(MessageDef* msg) : msg_(msg) {} + field_iterator begin() { return msg_->field_begin(); } + field_iterator end() { return msg_->field_end(); } + private: + MessageDef* msg_; + }; + + class ConstFieldAccessor { + public: + explicit ConstFieldAccessor(const MessageDef* msg) : msg_(msg) {} + const_field_iterator begin() { return msg_->field_begin(); } + const_field_iterator end() { return msg_->field_end(); } + private: + const MessageDef* msg_; + }; + + class OneofAccessor { + public: + explicit OneofAccessor(MessageDef* msg) : msg_(msg) {} + oneof_iterator begin() { return msg_->oneof_begin(); } + oneof_iterator end() { return msg_->oneof_end(); } + private: + MessageDef* msg_; + }; + + class ConstOneofAccessor { + public: + explicit ConstOneofAccessor(const MessageDef* msg) : msg_(msg) {} + const_oneof_iterator begin() { return msg_->oneof_begin(); } + const_oneof_iterator end() { return msg_->oneof_end(); } + private: + const MessageDef* msg_; + }; + + field_iterator field_begin(); + field_iterator field_end(); + const_field_iterator field_begin() const; + const_field_iterator field_end() const; + + oneof_iterator oneof_begin(); + oneof_iterator oneof_end(); + const_oneof_iterator oneof_begin() const; + const_oneof_iterator oneof_end() const; + + FieldAccessor fields() { return FieldAccessor(this); } + ConstFieldAccessor fields() const { return ConstFieldAccessor(this); } + OneofAccessor oneofs() { return OneofAccessor(this); } + ConstOneofAccessor oneofs() const { return ConstOneofAccessor(this); } + + private: + UPB_DISALLOW_POD_OPS(MessageDef, upb::MessageDef); +), +UPB_DEFINE_STRUCT(upb_msgdef, upb_def, + size_t selector_count; + uint32_t submsg_field_count; + + // Tables for looking up fields by number and name. + upb_inttable itof; // int to field + upb_strtable ntof; // name to field + + // Tables for looking up oneofs by name. + upb_strtable ntoo; // name to oneof + + // Is this a map-entry message? + // TODO: set this flag properly for static descriptors; regenerate + // descriptor.upb.c. + bool map_entry; + + // TODO(haberman): proper extension ranges (there can be multiple). +)); + +// TODO: also support static initialization of the oneofs table. This will be +// needed if we compile in descriptors that contain oneofs. +#define UPB_MSGDEF_INIT(name, selector_count, submsg_field_count, itof, ntof, \ + refs, ref2s) \ + { \ + UPB_DEF_INIT(name, UPB_DEF_MSG, refs, ref2s), selector_count, \ + submsg_field_count, itof, ntof, \ + UPB_EMPTY_STRTABLE_INIT(UPB_CTYPE_PTR), false \ + } + +UPB_BEGIN_EXTERN_C // { + +// Returns NULL if memory allocation failed. +upb_msgdef *upb_msgdef_new(const void *owner); + +// From upb_refcounted. +bool upb_msgdef_isfrozen(const upb_msgdef *m); +void upb_msgdef_ref(const upb_msgdef *m, const void *owner); +void upb_msgdef_unref(const upb_msgdef *m, const void *owner); +void upb_msgdef_donateref(const upb_msgdef *m, const void *from, + const void *to); +void upb_msgdef_checkref(const upb_msgdef *m, const void *owner); +bool upb_msgdef_freeze(upb_msgdef *m, upb_status *status); + +// From upb_def. +const char *upb_msgdef_fullname(const upb_msgdef *m); +bool upb_msgdef_setfullname(upb_msgdef *m, const char *fullname, upb_status *s); + +upb_msgdef *upb_msgdef_dup(const upb_msgdef *m, const void *owner); +bool upb_msgdef_addfield(upb_msgdef *m, upb_fielddef *f, const void *ref_donor, + upb_status *s); +bool upb_msgdef_addoneof(upb_msgdef *m, upb_oneofdef *o, const void *ref_donor, + upb_status *s); + +// Field lookup in a couple of different variations: +// - itof = int to field +// - ntof = name to field +// - ntofz = name to field, null-terminated string. +const upb_fielddef *upb_msgdef_itof(const upb_msgdef *m, uint32_t i); +const upb_fielddef *upb_msgdef_ntof(const upb_msgdef *m, const char *name, + size_t len); +int upb_msgdef_numfields(const upb_msgdef *m); + +UPB_INLINE const upb_fielddef *upb_msgdef_ntofz(const upb_msgdef *m, + const char *name) { + return upb_msgdef_ntof(m, name, strlen(name)); +} + +UPB_INLINE upb_fielddef *upb_msgdef_itof_mutable(upb_msgdef *m, uint32_t i) { + return (upb_fielddef*)upb_msgdef_itof(m, i); +} + +UPB_INLINE upb_fielddef *upb_msgdef_ntof_mutable(upb_msgdef *m, + const char *name, size_t len) { + return (upb_fielddef *)upb_msgdef_ntof(m, name, len); +} + +// Oneof lookup: +// - ntoo = name to oneof +// - ntooz = name to oneof, null-terminated string. +const upb_oneofdef *upb_msgdef_ntoo(const upb_msgdef *m, const char *name, + size_t len); +int upb_msgdef_numoneofs(const upb_msgdef *m); + +UPB_INLINE const upb_oneofdef *upb_msgdef_ntooz(const upb_msgdef *m, + const char *name) { + return upb_msgdef_ntoo(m, name, strlen(name)); +} + +UPB_INLINE upb_oneofdef *upb_msgdef_ntoo_mutable(upb_msgdef *m, + const char *name, size_t len) { + return (upb_oneofdef *)upb_msgdef_ntoo(m, name, len); +} + +void upb_msgdef_setmapentry(upb_msgdef *m, bool map_entry); +bool upb_msgdef_mapentry(const upb_msgdef *m); + +// Well-known field tag numbers for map-entry messages. +#define UPB_MAPENTRY_KEY 1 +#define UPB_MAPENTRY_VALUE 2 + +const upb_oneofdef *upb_msgdef_findoneof(const upb_msgdef *m, + const char *name); +int upb_msgdef_numoneofs(const upb_msgdef *m); + +// upb_msg_field_iter i; +// for(upb_msg_field_begin(&i, m); +// !upb_msg_field_done(&i); +// upb_msg_field_next(&i)) { +// upb_fielddef *f = upb_msg_iter_field(&i); +// // ... +// } +// +// For C we don't have separate iterators for const and non-const. +// It is the caller's responsibility to cast the upb_fielddef* to +// const if the upb_msgdef* is const. +void upb_msg_field_begin(upb_msg_field_iter *iter, const upb_msgdef *m); +void upb_msg_field_next(upb_msg_field_iter *iter); +bool upb_msg_field_done(const upb_msg_field_iter *iter); +upb_fielddef *upb_msg_iter_field(const upb_msg_field_iter *iter); +void upb_msg_field_iter_setdone(upb_msg_field_iter *iter); + +// Similar to above, we also support iterating through the oneofs in a msgdef. +void upb_msg_oneof_begin(upb_msg_oneof_iter *iter, const upb_msgdef *m); +void upb_msg_oneof_next(upb_msg_oneof_iter *iter); +bool upb_msg_oneof_done(const upb_msg_oneof_iter *iter); +upb_oneofdef *upb_msg_iter_oneof(const upb_msg_oneof_iter *iter); +void upb_msg_oneof_iter_setdone(upb_msg_oneof_iter *iter); + +UPB_END_EXTERN_C // } + + +/* upb::EnumDef ***************************************************************/ + +typedef upb_strtable_iter upb_enum_iter; + +// Class that represents an enum. Its base class is upb::Def (convert with +// upb::upcast()). +UPB_DEFINE_DEF(upb::EnumDef, enumdef, ENUM, + public: + // Returns NULL if memory allocation failed. + static reffed_ptr New(); + + // Functionality from upb::RefCounted. + bool IsFrozen() const; + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void* from, const void* to) const; + void CheckRef(const void* owner) const; + + // Functionality from upb::Def. + const char* full_name() const; + bool set_full_name(const char* fullname, Status* s); + bool set_full_name(const std::string& fullname, Status* s); + + // Call to freeze this EnumDef. + bool Freeze(Status* s); + + // The value that is used as the default when no field default is specified. + // If not set explicitly, the first value that was added will be used. + // The default value must be a member of the enum. + // Requires that value_count() > 0. + int32_t default_value() const; + + // Sets the default value. If this value is not valid, returns false and an + // error message in status. + bool set_default_value(int32_t val, Status* status); + + // Returns the number of values currently defined in the enum. Note that + // multiple names can refer to the same number, so this may be greater than + // the total number of unique numbers. + int value_count() const; + + // Adds a single name/number pair to the enum. Fails if this name has + // already been used by another value. + bool AddValue(const char* name, int32_t num, Status* status); + bool AddValue(const std::string& name, int32_t num, Status* status); + + // Lookups from name to integer, returning true if found. + bool FindValueByName(const char* name, int32_t* num) const; + + // Finds the name corresponding to the given number, or NULL if none was + // found. If more than one name corresponds to this number, returns the + // first one that was added. + const char* FindValueByNumber(int32_t num) const; + + // Returns a new EnumDef with all the same values. The new EnumDef will be + // owned by the given owner. + EnumDef* Dup(const void* owner) const; + + // Iteration over name/value pairs. The order is undefined. + // Adding an enum val invalidates any iterators. + // + // TODO: make compatible with range-for, with elements as pairs? + class Iterator { + public: + explicit Iterator(const EnumDef*); + + int32_t number(); + const char *name(); + bool Done(); + void Next(); + + private: + upb_enum_iter iter_; + }; + + private: + UPB_DISALLOW_POD_OPS(EnumDef, upb::EnumDef); +, +UPB_DEFINE_STRUCT(upb_enumdef, upb_def, + upb_strtable ntoi; + upb_inttable iton; + int32_t defaultval; +)); + +#define UPB_ENUMDEF_INIT(name, ntoi, iton, defaultval, refs, ref2s) \ + { UPB_DEF_INIT(name, UPB_DEF_ENUM, refs, ref2s), ntoi, iton, defaultval } + +UPB_BEGIN_EXTERN_C // { + +// Native C API. +upb_enumdef *upb_enumdef_new(const void *owner); +upb_enumdef *upb_enumdef_dup(const upb_enumdef *e, const void *owner); + +// From upb_refcounted. +void upb_enumdef_unref(const upb_enumdef *e, const void *owner); +bool upb_enumdef_isfrozen(const upb_enumdef *e); +void upb_enumdef_ref(const upb_enumdef *e, const void *owner); +void upb_enumdef_donateref(const upb_enumdef *m, const void *from, + const void *to); +void upb_enumdef_checkref(const upb_enumdef *e, const void *owner); +bool upb_enumdef_freeze(upb_enumdef *e, upb_status *status); + +// From upb_def. +const char *upb_enumdef_fullname(const upb_enumdef *e); +bool upb_enumdef_setfullname(upb_enumdef *e, const char *fullname, + upb_status *s); + +int32_t upb_enumdef_default(const upb_enumdef *e); +bool upb_enumdef_setdefault(upb_enumdef *e, int32_t val, upb_status *s); +int upb_enumdef_numvals(const upb_enumdef *e); +bool upb_enumdef_addval(upb_enumdef *e, const char *name, int32_t num, + upb_status *status); + +// Enum lookups: +// - ntoi: look up a name with specified length. +// - ntoiz: look up a name provided as a null-terminated string. +// - iton: look up an integer, returning the name as a null-terminated string. +bool upb_enumdef_ntoi(const upb_enumdef *e, const char *name, size_t len, + int32_t *num); +UPB_INLINE bool upb_enumdef_ntoiz(const upb_enumdef *e, + const char *name, int32_t *num) { + return upb_enumdef_ntoi(e, name, strlen(name), num); +} +const char *upb_enumdef_iton(const upb_enumdef *e, int32_t num); + +// upb_enum_iter i; +// for(upb_enum_begin(&i, e); !upb_enum_done(&i); upb_enum_next(&i)) { +// // ... +// } +void upb_enum_begin(upb_enum_iter *iter, const upb_enumdef *e); +void upb_enum_next(upb_enum_iter *iter); +bool upb_enum_done(upb_enum_iter *iter); +const char *upb_enum_iter_name(upb_enum_iter *iter); +int32_t upb_enum_iter_number(upb_enum_iter *iter); + +UPB_END_EXTERN_C // } + +/* upb::OneofDef **************************************************************/ + +typedef upb_inttable_iter upb_oneof_iter; + +// Class that represents a oneof. Its base class is upb::Def (convert with +// upb::upcast()). +UPB_DEFINE_DEF(upb::OneofDef, oneofdef, ONEOF, UPB_QUOTE( + public: + // Returns NULL if memory allocation failed. + static reffed_ptr New(); + + // Functionality from upb::RefCounted. + bool IsFrozen() const; + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void* from, const void* to) const; + void CheckRef(const void* owner) const; + + // Functionality from upb::Def. + const char* full_name() const; + + // Returns the MessageDef that owns this OneofDef. + const MessageDef* containing_type() const; + + // Returns the name of this oneof. This is the name used to look up the oneof + // by name once added to a message def. + const char* name() const; + bool set_name(const char* name, Status* s); + + // Returns the number of fields currently defined in the oneof. + int field_count() const; + + // Adds a field to the oneof. The field must not have been added to any other + // oneof or msgdef. If the oneof is not yet part of a msgdef, then when the + // oneof is eventually added to a msgdef, all fields added to the oneof will + // also be added to the msgdef at that time. If the oneof is already part of a + // msgdef, the field must either be a part of that msgdef already, or must not + // be a part of any msgdef; in the latter case, the field is added to the + // msgdef as a part of this operation. + // + // The field may only have an OPTIONAL label, never REQUIRED or REPEATED. + // + // If |f| is already part of this MessageDef, this method performs no action + // and returns true (success). Thus, this method is idempotent. + bool AddField(FieldDef* field, Status* s); + bool AddField(const reffed_ptr& field, Status* s); + + // Looks up by name. + const FieldDef* FindFieldByName(const char* name, size_t len) const; + FieldDef* FindFieldByName(const char* name, size_t len); + const FieldDef* FindFieldByName(const char* name) const { + return FindFieldByName(name, strlen(name)); + } + FieldDef* FindFieldByName(const char* name) { + return FindFieldByName(name, strlen(name)); + } + + template + FieldDef* FindFieldByName(const T& str) { + return FindFieldByName(str.c_str(), str.size()); + } + template + const FieldDef* FindFieldByName(const T& str) const { + return FindFieldByName(str.c_str(), str.size()); + } + + // Looks up by tag number. + const FieldDef* FindFieldByNumber(uint32_t num) const; + + // Returns a new OneofDef with all the same fields. The OneofDef will be owned + // by the given owner. + OneofDef* Dup(const void* owner) const; + + // Iteration over fields. The order is undefined. + class iterator : public std::iterator { + public: + explicit iterator(OneofDef* md); + static iterator end(OneofDef* md); + + void operator++(); + FieldDef* operator*() const; + bool operator!=(const iterator& other) const; + bool operator==(const iterator& other) const; + + private: + upb_oneof_iter iter_; + }; + + class const_iterator + : public std::iterator { + public: + explicit const_iterator(const OneofDef* md); + static const_iterator end(const OneofDef* md); + + void operator++(); + const FieldDef* operator*() const; + bool operator!=(const const_iterator& other) const; + bool operator==(const const_iterator& other) const; + + private: + upb_oneof_iter iter_; + }; + + iterator begin(); + iterator end(); + const_iterator begin() const; + const_iterator end() const; + + private: + UPB_DISALLOW_POD_OPS(OneofDef, upb::OneofDef); +), +UPB_DEFINE_STRUCT(upb_oneofdef, upb_def, + upb_strtable ntof; + upb_inttable itof; + const upb_msgdef *parent; +)); + +#define UPB_ONEOFDEF_INIT(name, ntof, itof, refs, ref2s) \ + { UPB_DEF_INIT(name, UPB_DEF_ENUM, refs, ref2s), ntof, itof } + +UPB_BEGIN_EXTERN_C // { + +// Native C API. +upb_oneofdef *upb_oneofdef_new(const void *owner); +upb_oneofdef *upb_oneofdef_dup(const upb_oneofdef *o, const void *owner); + +// From upb_refcounted. +void upb_oneofdef_unref(const upb_oneofdef *o, const void *owner); +bool upb_oneofdef_isfrozen(const upb_oneofdef *e); +void upb_oneofdef_ref(const upb_oneofdef *o, const void *owner); +void upb_oneofdef_donateref(const upb_oneofdef *m, const void *from, + const void *to); +void upb_oneofdef_checkref(const upb_oneofdef *o, const void *owner); + +const char *upb_oneofdef_name(const upb_oneofdef *o); +bool upb_oneofdef_setname(upb_oneofdef *o, const char *name, upb_status *s); + +const upb_msgdef *upb_oneofdef_containingtype(const upb_oneofdef *o); +int upb_oneofdef_numfields(const upb_oneofdef *o); +bool upb_oneofdef_addfield(upb_oneofdef *o, upb_fielddef *f, + const void *ref_donor, + upb_status *s); + +// Oneof lookups: +// - ntof: look up a field by name. +// - ntofz: look up a field by name (as a null-terminated string). +// - itof: look up a field by number. +const upb_fielddef *upb_oneofdef_ntof(const upb_oneofdef *o, + const char *name, size_t length); +UPB_INLINE const upb_fielddef *upb_oneofdef_ntofz(const upb_oneofdef *o, + const char *name) { + return upb_oneofdef_ntof(o, name, strlen(name)); +} +const upb_fielddef *upb_oneofdef_itof(const upb_oneofdef *o, uint32_t num); + +// upb_oneof_iter i; +// for(upb_oneof_begin(&i, e); !upb_oneof_done(&i); upb_oneof_next(&i)) { +// // ... +// } +void upb_oneof_begin(upb_oneof_iter *iter, const upb_oneofdef *o); +void upb_oneof_next(upb_oneof_iter *iter); +bool upb_oneof_done(upb_oneof_iter *iter); +upb_fielddef *upb_oneof_iter_field(const upb_oneof_iter *iter); +void upb_oneof_iter_setdone(upb_oneof_iter *iter); + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus + +UPB_INLINE const char* upb_safecstr(const std::string& str) { + assert(str.size() == std::strlen(str.c_str())); + return str.c_str(); +} + +// Inline C++ wrappers. +namespace upb { + +inline Def* Def::Dup(const void* owner) const { + return upb_def_dup(this, owner); +} +inline bool Def::IsFrozen() const { return upb_def_isfrozen(this); } +inline void Def::Ref(const void* owner) const { upb_def_ref(this, owner); } +inline void Def::Unref(const void* owner) const { upb_def_unref(this, owner); } +inline void Def::DonateRef(const void* from, const void* to) const { + upb_def_donateref(this, from, to); +} +inline void Def::CheckRef(const void* owner) const { + upb_def_checkref(this, owner); +} +inline Def::Type Def::def_type() const { return upb_def_type(this); } +inline const char* Def::full_name() const { return upb_def_fullname(this); } +inline bool Def::set_full_name(const char* fullname, Status* s) { + return upb_def_setfullname(this, fullname, s); +} +inline bool Def::set_full_name(const std::string& fullname, Status* s) { + return upb_def_setfullname(this, upb_safecstr(fullname), s); +} +inline bool Def::Freeze(Def* const* defs, int n, Status* status) { + return upb_def_freeze(defs, n, status); +} +inline bool Def::Freeze(const std::vector& defs, Status* status) { + return upb_def_freeze((Def* const*)&defs[0], defs.size(), status); +} + +inline bool FieldDef::CheckType(int32_t val) { + return upb_fielddef_checktype(val); +} +inline bool FieldDef::CheckLabel(int32_t val) { + return upb_fielddef_checklabel(val); +} +inline bool FieldDef::CheckDescriptorType(int32_t val) { + return upb_fielddef_checkdescriptortype(val); +} +inline bool FieldDef::CheckIntegerFormat(int32_t val) { + return upb_fielddef_checkintfmt(val); +} +inline FieldDef::Type FieldDef::ConvertType(int32_t val) { + assert(CheckType(val)); + return static_cast(val); +} +inline FieldDef::Label FieldDef::ConvertLabel(int32_t val) { + assert(CheckLabel(val)); + return static_cast(val); +} +inline FieldDef::DescriptorType FieldDef::ConvertDescriptorType(int32_t val) { + assert(CheckDescriptorType(val)); + return static_cast(val); +} +inline FieldDef::IntegerFormat FieldDef::ConvertIntegerFormat(int32_t val) { + assert(CheckIntegerFormat(val)); + return static_cast(val); +} + +inline reffed_ptr FieldDef::New() { + upb_fielddef *f = upb_fielddef_new(&f); + return reffed_ptr(f, &f); +} +inline FieldDef* FieldDef::Dup(const void* owner) const { + return upb_fielddef_dup(this, owner); +} +inline bool FieldDef::IsFrozen() const { return upb_fielddef_isfrozen(this); } +inline void FieldDef::Ref(const void* owner) const { + upb_fielddef_ref(this, owner); +} +inline void FieldDef::Unref(const void* owner) const { + upb_fielddef_unref(this, owner); +} +inline void FieldDef::DonateRef(const void* from, const void* to) const { + upb_fielddef_donateref(this, from, to); +} +inline void FieldDef::CheckRef(const void* owner) const { + upb_fielddef_checkref(this, owner); +} +inline const char* FieldDef::full_name() const { + return upb_fielddef_fullname(this); +} +inline bool FieldDef::set_full_name(const char* fullname, Status* s) { + return upb_fielddef_setfullname(this, fullname, s); +} +inline bool FieldDef::set_full_name(const std::string& fullname, Status* s) { + return upb_fielddef_setfullname(this, upb_safecstr(fullname), s); +} +inline bool FieldDef::type_is_set() const { + return upb_fielddef_typeisset(this); +} +inline FieldDef::Type FieldDef::type() const { return upb_fielddef_type(this); } +inline FieldDef::DescriptorType FieldDef::descriptor_type() const { + return upb_fielddef_descriptortype(this); +} +inline FieldDef::Label FieldDef::label() const { + return upb_fielddef_label(this); +} +inline uint32_t FieldDef::number() const { return upb_fielddef_number(this); } +inline const char* FieldDef::name() const { return upb_fielddef_name(this); } +inline bool FieldDef::is_extension() const { + return upb_fielddef_isextension(this); +} +inline bool FieldDef::lazy() const { + return upb_fielddef_lazy(this); +} +inline void FieldDef::set_lazy(bool lazy) { + upb_fielddef_setlazy(this, lazy); +} +inline bool FieldDef::packed() const { + return upb_fielddef_packed(this); +} +inline void FieldDef::set_packed(bool packed) { + upb_fielddef_setpacked(this, packed); +} +inline const MessageDef* FieldDef::containing_type() const { + return upb_fielddef_containingtype(this); +} +inline const OneofDef* FieldDef::containing_oneof() const { + return upb_fielddef_containingoneof(this); +} +inline const char* FieldDef::containing_type_name() { + return upb_fielddef_containingtypename(this); +} +inline bool FieldDef::set_number(uint32_t number, Status* s) { + return upb_fielddef_setnumber(this, number, s); +} +inline bool FieldDef::set_name(const char *name, Status* s) { + return upb_fielddef_setname(this, name, s); +} +inline bool FieldDef::set_name(const std::string& name, Status* s) { + return upb_fielddef_setname(this, upb_safecstr(name), s); +} +inline bool FieldDef::set_containing_type_name(const char *name, Status* s) { + return upb_fielddef_setcontainingtypename(this, name, s); +} +inline bool FieldDef::set_containing_type_name(const std::string &name, + Status *s) { + return upb_fielddef_setcontainingtypename(this, upb_safecstr(name), s); +} +inline void FieldDef::set_type(upb_fieldtype_t type) { + upb_fielddef_settype(this, type); +} +inline void FieldDef::set_is_extension(bool is_extension) { + upb_fielddef_setisextension(this, is_extension); +} +inline void FieldDef::set_descriptor_type(FieldDef::DescriptorType type) { + upb_fielddef_setdescriptortype(this, type); +} +inline void FieldDef::set_label(upb_label_t label) { + upb_fielddef_setlabel(this, label); +} +inline bool FieldDef::IsSubMessage() const { + return upb_fielddef_issubmsg(this); +} +inline bool FieldDef::IsString() const { return upb_fielddef_isstring(this); } +inline bool FieldDef::IsSequence() const { return upb_fielddef_isseq(this); } +inline bool FieldDef::IsMap() const { return upb_fielddef_ismap(this); } +inline int64_t FieldDef::default_int64() const { + return upb_fielddef_defaultint64(this); +} +inline int32_t FieldDef::default_int32() const { + return upb_fielddef_defaultint32(this); +} +inline uint64_t FieldDef::default_uint64() const { + return upb_fielddef_defaultuint64(this); +} +inline uint32_t FieldDef::default_uint32() const { + return upb_fielddef_defaultuint32(this); +} +inline bool FieldDef::default_bool() const { + return upb_fielddef_defaultbool(this); +} +inline float FieldDef::default_float() const { + return upb_fielddef_defaultfloat(this); +} +inline double FieldDef::default_double() const { + return upb_fielddef_defaultdouble(this); +} +inline const char* FieldDef::default_string(size_t* len) const { + return upb_fielddef_defaultstr(this, len); +} +inline void FieldDef::set_default_int64(int64_t value) { + upb_fielddef_setdefaultint64(this, value); +} +inline void FieldDef::set_default_int32(int32_t value) { + upb_fielddef_setdefaultint32(this, value); +} +inline void FieldDef::set_default_uint64(uint64_t value) { + upb_fielddef_setdefaultuint64(this, value); +} +inline void FieldDef::set_default_uint32(uint32_t value) { + upb_fielddef_setdefaultuint32(this, value); +} +inline void FieldDef::set_default_bool(bool value) { + upb_fielddef_setdefaultbool(this, value); +} +inline void FieldDef::set_default_float(float value) { + upb_fielddef_setdefaultfloat(this, value); +} +inline void FieldDef::set_default_double(double value) { + upb_fielddef_setdefaultdouble(this, value); +} +inline bool FieldDef::set_default_string(const void *str, size_t len, + Status *s) { + return upb_fielddef_setdefaultstr(this, str, len, s); +} +inline bool FieldDef::set_default_string(const std::string& str, Status* s) { + return upb_fielddef_setdefaultstr(this, str.c_str(), str.size(), s); +} +inline void FieldDef::set_default_cstr(const char* str, Status* s) { + return upb_fielddef_setdefaultcstr(this, str, s); +} +inline bool FieldDef::HasSubDef() const { return upb_fielddef_hassubdef(this); } +inline const Def* FieldDef::subdef() const { return upb_fielddef_subdef(this); } +inline const MessageDef *FieldDef::message_subdef() const { + return upb_fielddef_msgsubdef(this); +} +inline const EnumDef *FieldDef::enum_subdef() const { + return upb_fielddef_enumsubdef(this); +} +inline const char* FieldDef::subdef_name() const { + return upb_fielddef_subdefname(this); +} +inline bool FieldDef::set_subdef(const Def* subdef, Status* s) { + return upb_fielddef_setsubdef(this, subdef, s); +} +inline bool FieldDef::set_enum_subdef(const EnumDef* subdef, Status* s) { + return upb_fielddef_setenumsubdef(this, subdef, s); +} +inline bool FieldDef::set_message_subdef(const MessageDef* subdef, Status* s) { + return upb_fielddef_setmsgsubdef(this, subdef, s); +} +inline bool FieldDef::set_subdef_name(const char* name, Status* s) { + return upb_fielddef_setsubdefname(this, name, s); +} +inline bool FieldDef::set_subdef_name(const std::string& name, Status* s) { + return upb_fielddef_setsubdefname(this, upb_safecstr(name), s); +} + +inline reffed_ptr MessageDef::New() { + upb_msgdef *m = upb_msgdef_new(&m); + return reffed_ptr(m, &m); +} +inline bool MessageDef::IsFrozen() const { return upb_msgdef_isfrozen(this); } +inline void MessageDef::Ref(const void* owner) const { + return upb_msgdef_ref(this, owner); +} +inline void MessageDef::Unref(const void* owner) const { + return upb_msgdef_unref(this, owner); +} +inline void MessageDef::DonateRef(const void* from, const void* to) const { + return upb_msgdef_donateref(this, from, to); +} +inline void MessageDef::CheckRef(const void* owner) const { + return upb_msgdef_checkref(this, owner); +} +inline const char *MessageDef::full_name() const { + return upb_msgdef_fullname(this); +} +inline bool MessageDef::set_full_name(const char* fullname, Status* s) { + return upb_msgdef_setfullname(this, fullname, s); +} +inline bool MessageDef::set_full_name(const std::string& fullname, Status* s) { + return upb_msgdef_setfullname(this, upb_safecstr(fullname), s); +} +inline bool MessageDef::Freeze(Status* status) { + return upb_msgdef_freeze(this, status); +} +inline int MessageDef::field_count() const { + return upb_msgdef_numfields(this); +} +inline int MessageDef::oneof_count() const { + return upb_msgdef_numoneofs(this); +} +inline bool MessageDef::AddField(upb_fielddef* f, Status* s) { + return upb_msgdef_addfield(this, f, NULL, s); +} +inline bool MessageDef::AddField(const reffed_ptr& f, Status* s) { + return upb_msgdef_addfield(this, f.get(), NULL, s); +} +inline bool MessageDef::AddOneof(upb_oneofdef* o, Status* s) { + return upb_msgdef_addoneof(this, o, NULL, s); +} +inline bool MessageDef::AddOneof(const reffed_ptr& o, Status* s) { + return upb_msgdef_addoneof(this, o.get(), NULL, s); +} +inline FieldDef* MessageDef::FindFieldByNumber(uint32_t number) { + return upb_msgdef_itof_mutable(this, number); +} +inline FieldDef* MessageDef::FindFieldByName(const char* name, size_t len) { + return upb_msgdef_ntof_mutable(this, name, len); +} +inline const FieldDef* MessageDef::FindFieldByNumber(uint32_t number) const { + return upb_msgdef_itof(this, number); +} +inline const FieldDef *MessageDef::FindFieldByName(const char *name, + size_t len) const { + return upb_msgdef_ntof(this, name, len); +} +inline OneofDef* MessageDef::FindOneofByName(const char* name, size_t len) { + return upb_msgdef_ntoo_mutable(this, name, len); +} +inline const OneofDef* MessageDef::FindOneofByName(const char* name, + size_t len) const { + return upb_msgdef_ntoo(this, name, len); +} +inline MessageDef* MessageDef::Dup(const void *owner) const { + return upb_msgdef_dup(this, owner); +} +inline void MessageDef::setmapentry(bool map_entry) { + upb_msgdef_setmapentry(this, map_entry); +} +inline bool MessageDef::mapentry() const { + return upb_msgdef_mapentry(this); +} +inline MessageDef::field_iterator MessageDef::field_begin() { + return field_iterator(this); +} +inline MessageDef::field_iterator MessageDef::field_end() { + return field_iterator::end(this); +} +inline MessageDef::const_field_iterator MessageDef::field_begin() const { + return const_field_iterator(this); +} +inline MessageDef::const_field_iterator MessageDef::field_end() const { + return const_field_iterator::end(this); +} + +inline MessageDef::oneof_iterator MessageDef::oneof_begin() { + return oneof_iterator(this); +} +inline MessageDef::oneof_iterator MessageDef::oneof_end() { + return oneof_iterator::end(this); +} +inline MessageDef::const_oneof_iterator MessageDef::oneof_begin() const { + return const_oneof_iterator(this); +} +inline MessageDef::const_oneof_iterator MessageDef::oneof_end() const { + return const_oneof_iterator::end(this); +} + +inline MessageDef::field_iterator::field_iterator(MessageDef* md) { + upb_msg_field_begin(&iter_, md); +} +inline MessageDef::field_iterator MessageDef::field_iterator::end( + MessageDef* md) { + MessageDef::field_iterator iter(md); + upb_msg_field_iter_setdone(&iter.iter_); + return iter; +} +inline FieldDef* MessageDef::field_iterator::operator*() const { + return upb_msg_iter_field(&iter_); +} +inline void MessageDef::field_iterator::operator++() { + return upb_msg_field_next(&iter_); +} +inline bool MessageDef::field_iterator::operator==( + const field_iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::field_iterator::operator!=( + const field_iterator &other) const { + return !(*this == other); +} + +inline MessageDef::const_field_iterator::const_field_iterator( + const MessageDef* md) { + upb_msg_field_begin(&iter_, md); +} +inline MessageDef::const_field_iterator MessageDef::const_field_iterator::end( + const MessageDef *md) { + MessageDef::const_field_iterator iter(md); + upb_msg_field_iter_setdone(&iter.iter_); + return iter; +} +inline const FieldDef* MessageDef::const_field_iterator::operator*() const { + return upb_msg_iter_field(&iter_); +} +inline void MessageDef::const_field_iterator::operator++() { + return upb_msg_field_next(&iter_); +} +inline bool MessageDef::const_field_iterator::operator==( + const const_field_iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::const_field_iterator::operator!=( + const const_field_iterator &other) const { + return !(*this == other); +} + +inline MessageDef::oneof_iterator::oneof_iterator(MessageDef* md) { + upb_msg_oneof_begin(&iter_, md); +} +inline MessageDef::oneof_iterator MessageDef::oneof_iterator::end( + MessageDef* md) { + MessageDef::oneof_iterator iter(md); + upb_msg_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline OneofDef* MessageDef::oneof_iterator::operator*() const { + return upb_msg_iter_oneof(&iter_); +} +inline void MessageDef::oneof_iterator::operator++() { + return upb_msg_oneof_next(&iter_); +} +inline bool MessageDef::oneof_iterator::operator==( + const oneof_iterator &other) const { + return upb_strtable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::oneof_iterator::operator!=( + const oneof_iterator &other) const { + return !(*this == other); +} + +inline MessageDef::const_oneof_iterator::const_oneof_iterator( + const MessageDef* md) { + upb_msg_oneof_begin(&iter_, md); +} +inline MessageDef::const_oneof_iterator MessageDef::const_oneof_iterator::end( + const MessageDef *md) { + MessageDef::const_oneof_iterator iter(md); + upb_msg_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline const OneofDef* MessageDef::const_oneof_iterator::operator*() const { + return upb_msg_iter_oneof(&iter_); +} +inline void MessageDef::const_oneof_iterator::operator++() { + return upb_msg_oneof_next(&iter_); +} +inline bool MessageDef::const_oneof_iterator::operator==( + const const_oneof_iterator &other) const { + return upb_strtable_iter_isequal(&iter_, &other.iter_); +} +inline bool MessageDef::const_oneof_iterator::operator!=( + const const_oneof_iterator &other) const { + return !(*this == other); +} + +inline reffed_ptr EnumDef::New() { + upb_enumdef *e = upb_enumdef_new(&e); + return reffed_ptr(e, &e); +} +inline bool EnumDef::IsFrozen() const { return upb_enumdef_isfrozen(this); } +inline void EnumDef::Ref(const void* owner) const { + return upb_enumdef_ref(this, owner); +} +inline void EnumDef::Unref(const void* owner) const { + return upb_enumdef_unref(this, owner); +} +inline void EnumDef::DonateRef(const void* from, const void* to) const { + return upb_enumdef_donateref(this, from, to); +} +inline void EnumDef::CheckRef(const void* owner) const { + return upb_enumdef_checkref(this, owner); +} +inline const char* EnumDef::full_name() const { + return upb_enumdef_fullname(this); +} +inline bool EnumDef::set_full_name(const char* fullname, Status* s) { + return upb_enumdef_setfullname(this, fullname, s); +} +inline bool EnumDef::set_full_name(const std::string& fullname, Status* s) { + return upb_enumdef_setfullname(this, upb_safecstr(fullname), s); +} +inline bool EnumDef::Freeze(Status* status) { + return upb_enumdef_freeze(this, status); +} +inline int32_t EnumDef::default_value() const { + return upb_enumdef_default(this); +} +inline bool EnumDef::set_default_value(int32_t val, Status* status) { + return upb_enumdef_setdefault(this, val, status); +} +inline int EnumDef::value_count() const { return upb_enumdef_numvals(this); } +inline bool EnumDef::AddValue(const char* name, int32_t num, Status* status) { + return upb_enumdef_addval(this, name, num, status); +} +inline bool EnumDef::AddValue(const std::string& name, int32_t num, + Status* status) { + return upb_enumdef_addval(this, upb_safecstr(name), num, status); +} +inline bool EnumDef::FindValueByName(const char* name, int32_t *num) const { + return upb_enumdef_ntoiz(this, name, num); +} +inline const char* EnumDef::FindValueByNumber(int32_t num) const { + return upb_enumdef_iton(this, num); +} +inline EnumDef* EnumDef::Dup(const void* owner) const { + return upb_enumdef_dup(this, owner); +} + +inline EnumDef::Iterator::Iterator(const EnumDef* e) { + upb_enum_begin(&iter_, e); +} +inline int32_t EnumDef::Iterator::number() { + return upb_enum_iter_number(&iter_); +} +inline const char* EnumDef::Iterator::name() { + return upb_enum_iter_name(&iter_); +} +inline bool EnumDef::Iterator::Done() { return upb_enum_done(&iter_); } +inline void EnumDef::Iterator::Next() { return upb_enum_next(&iter_); } + +inline reffed_ptr OneofDef::New() { + upb_oneofdef *o = upb_oneofdef_new(&o); + return reffed_ptr(o, &o); +} +inline bool OneofDef::IsFrozen() const { return upb_oneofdef_isfrozen(this); } +inline void OneofDef::Ref(const void* owner) const { + return upb_oneofdef_ref(this, owner); +} +inline void OneofDef::Unref(const void* owner) const { + return upb_oneofdef_unref(this, owner); +} +inline void OneofDef::DonateRef(const void* from, const void* to) const { + return upb_oneofdef_donateref(this, from, to); +} +inline void OneofDef::CheckRef(const void* owner) const { + return upb_oneofdef_checkref(this, owner); +} +inline const char* OneofDef::full_name() const { + return upb_oneofdef_name(this); +} + +inline const MessageDef* OneofDef::containing_type() const { + return upb_oneofdef_containingtype(this); +} +inline const char* OneofDef::name() const { + return upb_oneofdef_name(this); +} +inline bool OneofDef::set_name(const char* name, Status* s) { + return upb_oneofdef_setname(this, name, s); +} +inline int OneofDef::field_count() const { + return upb_oneofdef_numfields(this); +} +inline bool OneofDef::AddField(FieldDef* field, Status* s) { + return upb_oneofdef_addfield(this, field, NULL, s); +} +inline bool OneofDef::AddField(const reffed_ptr& field, Status* s) { + return upb_oneofdef_addfield(this, field.get(), NULL, s); +} +inline const FieldDef* OneofDef::FindFieldByName(const char* name, + size_t len) const { + return upb_oneofdef_ntof(this, name, len); +} +inline const FieldDef* OneofDef::FindFieldByNumber(uint32_t num) const { + return upb_oneofdef_itof(this, num); +} +inline OneofDef::iterator OneofDef::begin() { return iterator(this); } +inline OneofDef::iterator OneofDef::end() { return iterator::end(this); } +inline OneofDef::const_iterator OneofDef::begin() const { + return const_iterator(this); +} +inline OneofDef::const_iterator OneofDef::end() const { + return const_iterator::end(this); +} + +inline OneofDef::iterator::iterator(OneofDef* o) { + upb_oneof_begin(&iter_, o); +} +inline OneofDef::iterator OneofDef::iterator::end(OneofDef* o) { + OneofDef::iterator iter(o); + upb_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline FieldDef* OneofDef::iterator::operator*() const { + return upb_oneof_iter_field(&iter_); +} +inline void OneofDef::iterator::operator++() { return upb_oneof_next(&iter_); } +inline bool OneofDef::iterator::operator==(const iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool OneofDef::iterator::operator!=(const iterator &other) const { + return !(*this == other); +} + +inline OneofDef::const_iterator::const_iterator(const OneofDef* md) { + upb_oneof_begin(&iter_, md); +} +inline OneofDef::const_iterator OneofDef::const_iterator::end( + const OneofDef *md) { + OneofDef::const_iterator iter(md); + upb_oneof_iter_setdone(&iter.iter_); + return iter; +} +inline const FieldDef* OneofDef::const_iterator::operator*() const { + return upb_msg_iter_field(&iter_); +} +inline void OneofDef::const_iterator::operator++() { + return upb_oneof_next(&iter_); +} +inline bool OneofDef::const_iterator::operator==( + const const_iterator &other) const { + return upb_inttable_iter_isequal(&iter_, &other.iter_); +} +inline bool OneofDef::const_iterator::operator!=( + const const_iterator &other) const { + return !(*this == other); +} + +} // namespace upb +#endif + +#undef UPB_DEFINE_DEF +#undef UPB_DEF_CASTS +#undef UPB_CPP_CASTS + +#endif /* UPB_DEF_H_ */ +// This file contains accessors for a set of compiled-in defs. +// Note that unlike Google's protobuf, it does *not* define +// generated classes or any other kind of data structure for +// actually storing protobufs. It only contains *defs* which +// let you reflect over a protobuf *schema*. +// +// This file was generated by upbc (the upb compiler). +// Do not edit -- your changes will be discarded when the file is +// regenerated. + +#ifndef GOOGLE_PROTOBUF_DESCRIPTOR_UPB_H_ +#define GOOGLE_PROTOBUF_DESCRIPTOR_UPB_H_ + +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * A symtab (symbol table) stores a name->def map of upb_defs. Clients could + * always create such tables themselves, but upb_symtab has logic for resolving + * symbolic references, and in particular, for keeping a whole set of consistent + * defs when replacing some subset of those defs. This logic is nontrivial. + * + * This is a mixed C/C++ interface that offers a full API to both languages. + * See the top-level README for more information. + */ + +#ifndef UPB_SYMTAB_H_ +#define UPB_SYMTAB_H_ + + +#ifdef __cplusplus +#include +namespace upb { class SymbolTable; } +#endif + +UPB_DECLARE_TYPE(upb::SymbolTable, upb_symtab); + +typedef struct { + UPB_PRIVATE_FOR_CPP + upb_strtable_iter iter; + upb_deftype_t type; +} upb_symtab_iter; + +// Non-const methods in upb::SymbolTable are NOT thread-safe. +UPB_DEFINE_CLASS1(upb::SymbolTable, upb::RefCounted, + public: + // Returns a new symbol table with a single ref owned by "owner." + // Returns NULL if memory allocation failed. + static reffed_ptr New(); + + // Functionality from upb::RefCounted. + bool IsFrozen() const; + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void *from, const void *to) const; + void CheckRef(const void *owner) const; + + // For all lookup functions, the returned pointer is not owned by the + // caller; it may be invalidated by any non-const call or unref of the + // SymbolTable! To protect against this, take a ref if desired. + + // Freezes the symbol table: prevents further modification of it. + // After the Freeze() operation is successful, the SymbolTable must only be + // accessed via a const pointer. + // + // Unlike with upb::MessageDef/upb::EnumDef/etc, freezing a SymbolTable is not + // a necessary step in using a SymbolTable. If you have no need for it to be + // immutable, there is no need to freeze it ever. However sometimes it is + // useful, and SymbolTables that are statically compiled into the binary are + // always frozen by nature. + void Freeze(); + + // Resolves the given symbol using the rules described in descriptor.proto, + // namely: + // + // If the name starts with a '.', it is fully-qualified. Otherwise, + // C++-like scoping rules are used to find the type (i.e. first the nested + // types within this message are searched, then within the parent, on up + // to the root namespace). + // + // If not found, returns NULL. + const Def* Resolve(const char* base, const char* sym) const; + + // Finds an entry in the symbol table with this exact name. If not found, + // returns NULL. + const Def* Lookup(const char *sym) const; + const MessageDef* LookupMessage(const char *sym) const; + const EnumDef* LookupEnum(const char *sym) const; + + // TODO: introduce a C++ iterator, but make it nice and templated so that if + // you ask for an iterator of MessageDef the iterated elements are strongly + // typed as MessageDef*. + + // Adds the given mutable defs to the symtab, resolving all symbols + // (including enum default values) and finalizing the defs. Only one def per + // name may be in the list, but defs can replace existing defs in the symtab. + // All defs must have a name -- anonymous defs are not allowed. Anonymous + // defs can still be frozen by calling upb_def_freeze() directly. + // + // Any existing defs that can reach defs that are being replaced will + // themselves be replaced also, so that the resulting set of defs is fully + // consistent. + // + // This logic implemented in this method is a convenience; ultimately it + // calls some combination of upb_fielddef_setsubdef(), upb_def_dup(), and + // upb_freeze(), any of which the client could call themself. However, since + // the logic for doing so is nontrivial, we provide it here. + // + // The entire operation either succeeds or fails. If the operation fails, + // the symtab is unchanged, false is returned, and status indicates the + // error. The caller passes a ref on all defs to the symtab (even if the + // operation fails). + // + // TODO(haberman): currently failure will leave the symtab unchanged, but may + // leave the defs themselves partially resolved. Does this matter? If so we + // could do a prepass that ensures that all symbols are resolvable and bail + // if not, so we don't mutate anything until we know the operation will + // succeed. + // + // TODO(haberman): since the defs must be mutable, refining a frozen def + // requires making mutable copies of the entire tree. This is wasteful if + // only a few messages are changing. We may want to add a way of adding a + // tree of frozen defs to the symtab (perhaps an alternate constructor where + // you pass the root of the tree?) + bool Add(Def*const* defs, int n, void* ref_donor, upb_status* status); + + bool Add(const std::vector& defs, void *owner, Status* status) { + return Add((Def*const*)&defs[0], defs.size(), owner, status); + } + + private: + UPB_DISALLOW_POD_OPS(SymbolTable, upb::SymbolTable); +, +UPB_DEFINE_STRUCT(upb_symtab, upb_refcounted, + upb_strtable symtab; +)); + +#define UPB_SYMTAB_INIT(symtab, refs, ref2s) \ + { UPB_REFCOUNT_INIT(refs, ref2s), symtab } + +UPB_BEGIN_EXTERN_C // { + +// Native C API. +// From upb_refcounted. +bool upb_symtab_isfrozen(const upb_symtab *s); +void upb_symtab_ref(const upb_symtab *s, const void *owner); +void upb_symtab_unref(const upb_symtab *s, const void *owner); +void upb_symtab_donateref( + const upb_symtab *s, const void *from, const void *to); +void upb_symtab_checkref(const upb_symtab *s, const void *owner); + +upb_symtab *upb_symtab_new(const void *owner); +void upb_symtab_freeze(upb_symtab *s); +const upb_def *upb_symtab_resolve(const upb_symtab *s, const char *base, + const char *sym); +const upb_def *upb_symtab_lookup(const upb_symtab *s, const char *sym); +const upb_msgdef *upb_symtab_lookupmsg(const upb_symtab *s, const char *sym); +const upb_enumdef *upb_symtab_lookupenum(const upb_symtab *s, const char *sym); +bool upb_symtab_add(upb_symtab *s, upb_def *const*defs, int n, void *ref_donor, + upb_status *status); + +// upb_symtab_iter i; +// for(upb_symtab_begin(&i, s, type); !upb_symtab_done(&i); +// upb_symtab_next(&i)) { +// const upb_def *def = upb_symtab_iter_def(&i); +// // ... +// } +// +// For C we don't have separate iterators for const and non-const. +// It is the caller's responsibility to cast the upb_fielddef* to +// const if the upb_msgdef* is const. +void upb_symtab_begin(upb_symtab_iter *iter, const upb_symtab *s, + upb_deftype_t type); +void upb_symtab_next(upb_symtab_iter *iter); +bool upb_symtab_done(const upb_symtab_iter *iter); +const upb_def *upb_symtab_iter_def(const upb_symtab_iter *iter); + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus +// C++ inline wrappers. +namespace upb { +inline reffed_ptr SymbolTable::New() { + upb_symtab *s = upb_symtab_new(&s); + return reffed_ptr(s, &s); +} + +inline bool SymbolTable::IsFrozen() const { + return upb_symtab_isfrozen(this); +} +inline void SymbolTable::Ref(const void *owner) const { + upb_symtab_ref(this, owner); +} +inline void SymbolTable::Unref(const void *owner) const { + upb_symtab_unref(this, owner); +} +inline void SymbolTable::DonateRef(const void *from, const void *to) const { + upb_symtab_donateref(this, from, to); +} +inline void SymbolTable::CheckRef(const void *owner) const { + upb_symtab_checkref(this, owner); +} + +inline void SymbolTable::Freeze() { + return upb_symtab_freeze(this); +} +inline const Def *SymbolTable::Resolve(const char *base, + const char *sym) const { + return upb_symtab_resolve(this, base, sym); +} +inline const Def* SymbolTable::Lookup(const char *sym) const { + return upb_symtab_lookup(this, sym); +} +inline const MessageDef *SymbolTable::LookupMessage(const char *sym) const { + return upb_symtab_lookupmsg(this, sym); +} +inline bool SymbolTable::Add( + Def*const* defs, int n, void* ref_donor, upb_status* status) { + return upb_symtab_add(this, (upb_def*const*)defs, n, ref_donor, status); +} +} // namespace upb +#endif + +#endif /* UPB_SYMTAB_H_ */ + +#ifdef __cplusplus +extern "C" { +#endif + +// Enums + +typedef enum { + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_OPTIONAL = 1, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_REQUIRED = 2, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_REPEATED = 3, +} google_protobuf_FieldDescriptorProto_Label; + +typedef enum { + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_DOUBLE = 1, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_FLOAT = 2, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_INT64 = 3, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_UINT64 = 4, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_INT32 = 5, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_FIXED64 = 6, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_FIXED32 = 7, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_BOOL = 8, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_STRING = 9, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_GROUP = 10, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_MESSAGE = 11, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_BYTES = 12, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_UINT32 = 13, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_ENUM = 14, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SFIXED32 = 15, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SFIXED64 = 16, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SINT32 = 17, + GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_SINT64 = 18, +} google_protobuf_FieldDescriptorProto_Type; + +typedef enum { + GOOGLE_PROTOBUF_FIELDOPTIONS_STRING = 0, + GOOGLE_PROTOBUF_FIELDOPTIONS_CORD = 1, + GOOGLE_PROTOBUF_FIELDOPTIONS_STRING_PIECE = 2, +} google_protobuf_FieldOptions_CType; + +typedef enum { + GOOGLE_PROTOBUF_FILEOPTIONS_SPEED = 1, + GOOGLE_PROTOBUF_FILEOPTIONS_CODE_SIZE = 2, + GOOGLE_PROTOBUF_FILEOPTIONS_LITE_RUNTIME = 3, +} google_protobuf_FileOptions_OptimizeMode; + +// Selectors + +// google.protobuf.DescriptorProto +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_STARTSUBMSG 3 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_STARTSUBMSG 4 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_STARTSUBMSG 5 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_STARTSUBMSG 6 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_OPTIONS_STARTSUBMSG 7 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_STARTSEQ 8 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_ENDSEQ 9 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_FIELD_ENDSUBMSG 10 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_STARTSEQ 11 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_ENDSEQ 12 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NESTED_TYPE_ENDSUBMSG 13 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_STARTSEQ 14 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_ENDSEQ 15 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_ENUM_TYPE_ENDSUBMSG 16 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_STARTSEQ 17 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_ENDSEQ 18 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_RANGE_ENDSUBMSG 19 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_STARTSEQ 20 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_ENDSEQ 21 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSION_ENDSUBMSG 22 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_OPTIONS_ENDSUBMSG 23 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NAME_STRING 24 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NAME_STARTSTR 25 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_NAME_ENDSTR 26 + +// google.protobuf.DescriptorProto.ExtensionRange +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSIONRANGE_START_INT32 2 +#define SEL_GOOGLE_PROTOBUF_DESCRIPTORPROTO_EXTENSIONRANGE_END_INT32 3 + +// google.protobuf.EnumDescriptorProto +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 3 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_STARTSEQ 4 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_ENDSEQ 5 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_VALUE_ENDSUBMSG 6 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 7 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_NAME_STRING 8 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_NAME_STARTSTR 9 +#define SEL_GOOGLE_PROTOBUF_ENUMDESCRIPTORPROTO_NAME_ENDSTR 10 + +// google.protobuf.EnumOptions +#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5 +#define SEL_GOOGLE_PROTOBUF_ENUMOPTIONS_ALLOW_ALIAS_BOOL 6 + +// google.protobuf.EnumValueDescriptorProto +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 3 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NAME_STRING 4 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NAME_STARTSTR 5 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NAME_ENDSTR 6 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEDESCRIPTORPROTO_NUMBER_INT32 7 + +// google.protobuf.EnumValueOptions +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_ENUMVALUEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5 + +// google.protobuf.FieldDescriptorProto +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 3 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NAME_STRING 4 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NAME_STARTSTR 5 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NAME_ENDSTR 6 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_EXTENDEE_STRING 7 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_EXTENDEE_STARTSTR 8 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_EXTENDEE_ENDSTR 9 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_NUMBER_INT32 10 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_LABEL_INT32 11 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_INT32 12 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_NAME_STRING 13 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_NAME_STARTSTR 14 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_TYPE_NAME_ENDSTR 15 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_DEFAULT_VALUE_STRING 16 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_DEFAULT_VALUE_STARTSTR 17 +#define SEL_GOOGLE_PROTOBUF_FIELDDESCRIPTORPROTO_DEFAULT_VALUE_ENDSTR 18 + +// google.protobuf.FieldOptions +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_CTYPE_INT32 6 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_PACKED_BOOL 7 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_DEPRECATED_BOOL 8 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_LAZY_BOOL 9 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_EXPERIMENTAL_MAP_KEY_STRING 10 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_EXPERIMENTAL_MAP_KEY_STARTSTR 11 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_EXPERIMENTAL_MAP_KEY_ENDSTR 12 +#define SEL_GOOGLE_PROTOBUF_FIELDOPTIONS_WEAK_BOOL 13 + +// google.protobuf.FileDescriptorProto +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_STARTSUBMSG 3 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_STARTSUBMSG 4 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_STARTSUBMSG 5 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 6 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SOURCE_CODE_INFO_STARTSUBMSG 7 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_STARTSEQ 8 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_ENDSEQ 9 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_MESSAGE_TYPE_ENDSUBMSG 10 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_STARTSEQ 11 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_ENDSEQ 12 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_ENUM_TYPE_ENDSUBMSG 13 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_STARTSEQ 14 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_ENDSEQ 15 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SERVICE_ENDSUBMSG 16 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_STARTSEQ 17 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_ENDSEQ 18 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_EXTENSION_ENDSUBMSG 19 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 20 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_SOURCE_CODE_INFO_ENDSUBMSG 21 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_NAME_STRING 22 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_NAME_STARTSTR 23 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_NAME_ENDSTR 24 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PACKAGE_STRING 25 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PACKAGE_STARTSTR 26 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PACKAGE_ENDSTR 27 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_STARTSEQ 28 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_ENDSEQ 29 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_STRING 30 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_STARTSTR 31 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_DEPENDENCY_ENDSTR 32 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PUBLIC_DEPENDENCY_STARTSEQ 33 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PUBLIC_DEPENDENCY_ENDSEQ 34 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_PUBLIC_DEPENDENCY_INT32 35 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_WEAK_DEPENDENCY_STARTSEQ 36 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_WEAK_DEPENDENCY_ENDSEQ 37 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORPROTO_WEAK_DEPENDENCY_INT32 38 + +// google.protobuf.FileDescriptorSet +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_FILEDESCRIPTORSET_FILE_ENDSUBMSG 5 + +// google.protobuf.FileOptions +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_PACKAGE_STRING 6 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_PACKAGE_STARTSTR 7 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_PACKAGE_ENDSTR 8 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_OUTER_CLASSNAME_STRING 9 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_OUTER_CLASSNAME_STARTSTR 10 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_OUTER_CLASSNAME_ENDSTR 11 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_OPTIMIZE_FOR_INT32 12 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_MULTIPLE_FILES_BOOL 13 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_GO_PACKAGE_STRING 14 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_GO_PACKAGE_STARTSTR 15 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_GO_PACKAGE_ENDSTR 16 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_CC_GENERIC_SERVICES_BOOL 17 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_GENERIC_SERVICES_BOOL 18 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_PY_GENERIC_SERVICES_BOOL 19 +#define SEL_GOOGLE_PROTOBUF_FILEOPTIONS_JAVA_GENERATE_EQUALS_AND_HASH_BOOL 20 + +// google.protobuf.MessageOptions +#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5 +#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_MESSAGE_SET_WIRE_FORMAT_BOOL 6 +#define SEL_GOOGLE_PROTOBUF_MESSAGEOPTIONS_NO_STANDARD_DESCRIPTOR_ACCESSOR_BOOL 7 + +// google.protobuf.MethodDescriptorProto +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 3 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_NAME_STRING 4 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_NAME_STARTSTR 5 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_NAME_ENDSTR 6 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_INPUT_TYPE_STRING 7 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_INPUT_TYPE_STARTSTR 8 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_INPUT_TYPE_ENDSTR 9 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OUTPUT_TYPE_STRING 10 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OUTPUT_TYPE_STARTSTR 11 +#define SEL_GOOGLE_PROTOBUF_METHODDESCRIPTORPROTO_OUTPUT_TYPE_ENDSTR 12 + +// google.protobuf.MethodOptions +#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_METHODOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5 + +// google.protobuf.ServiceDescriptorProto +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_OPTIONS_STARTSUBMSG 3 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_STARTSEQ 4 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_ENDSEQ 5 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_METHOD_ENDSUBMSG 6 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_OPTIONS_ENDSUBMSG 7 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_NAME_STRING 8 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_NAME_STARTSTR 9 +#define SEL_GOOGLE_PROTOBUF_SERVICEDESCRIPTORPROTO_NAME_ENDSTR 10 + +// google.protobuf.ServiceOptions +#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_SERVICEOPTIONS_UNINTERPRETED_OPTION_ENDSUBMSG 5 + +// google.protobuf.SourceCodeInfo +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_ENDSUBMSG 5 + +// google.protobuf.SourceCodeInfo.Location +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_PATH_STARTSEQ 2 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_PATH_ENDSEQ 3 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_PATH_INT32 4 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_SPAN_STARTSEQ 5 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_SPAN_ENDSEQ 6 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_SPAN_INT32 7 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_LEADING_COMMENTS_STRING 8 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_LEADING_COMMENTS_STARTSTR 9 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_LEADING_COMMENTS_ENDSTR 10 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_TRAILING_COMMENTS_STRING 11 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_TRAILING_COMMENTS_STARTSTR 12 +#define SEL_GOOGLE_PROTOBUF_SOURCECODEINFO_LOCATION_TRAILING_COMMENTS_ENDSTR 13 + +// google.protobuf.UninterpretedOption +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_STARTSUBMSG 2 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_STARTSEQ 3 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_ENDSEQ 4 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAME_ENDSUBMSG 5 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_IDENTIFIER_VALUE_STRING 6 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_IDENTIFIER_VALUE_STARTSTR 7 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_IDENTIFIER_VALUE_ENDSTR 8 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_POSITIVE_INT_VALUE_UINT64 9 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NEGATIVE_INT_VALUE_INT64 10 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_DOUBLE_VALUE_DOUBLE 11 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_STRING_VALUE_STRING 12 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_STRING_VALUE_STARTSTR 13 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_STRING_VALUE_ENDSTR 14 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_AGGREGATE_VALUE_STRING 15 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_AGGREGATE_VALUE_STARTSTR 16 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_AGGREGATE_VALUE_ENDSTR 17 + +// google.protobuf.UninterpretedOption.NamePart +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_NAME_PART_STRING 2 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_NAME_PART_STARTSTR 3 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_NAME_PART_ENDSTR 4 +#define SEL_GOOGLE_PROTOBUF_UNINTERPRETEDOPTION_NAMEPART_IS_EXTENSION_BOOL 5 + +const upb_symtab *upbdefs_google_protobuf_descriptor(const void *owner); + +// MessageDefs +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_DescriptorProto(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.DescriptorProto"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.DescriptorProto.ExtensionRange"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumDescriptorProto(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumDescriptorProto"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumOptions(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumOptions"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumValueDescriptorProto(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumValueDescriptorProto"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_EnumValueOptions(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.EnumValueOptions"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FieldDescriptorProto(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FieldDescriptorProto"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FieldOptions(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FieldOptions"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FileDescriptorProto(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FileDescriptorProto"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FileDescriptorSet(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FileDescriptorSet"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_FileOptions(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.FileOptions"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_MessageOptions(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.MessageOptions"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_MethodDescriptorProto(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.MethodDescriptorProto"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_MethodOptions(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.MethodOptions"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_ServiceDescriptorProto(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.ServiceDescriptorProto"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_ServiceOptions(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.ServiceOptions"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.SourceCodeInfo"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_SourceCodeInfo_Location(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.SourceCodeInfo.Location"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.UninterpretedOption"); + assert(m); + return m; +} +UPB_INLINE const upb_msgdef *upbdefs_google_protobuf_UninterpretedOption_NamePart(const upb_symtab *s) { + const upb_msgdef *m = upb_symtab_lookupmsg(s, "google.protobuf.UninterpretedOption.NamePart"); + assert(m); + return m; +} + + +// EnumDefs +UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Label(const upb_symtab *s) { + const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FieldDescriptorProto.Label"); + assert(e); + return e; +} +UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FieldDescriptorProto_Type(const upb_symtab *s) { + const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FieldDescriptorProto.Type"); + assert(e); + return e; +} +UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FieldOptions_CType(const upb_symtab *s) { + const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FieldOptions.CType"); + assert(e); + return e; +} +UPB_INLINE const upb_enumdef *upbdefs_google_protobuf_FileOptions_OptimizeMode(const upb_symtab *s) { + const upb_enumdef *e = upb_symtab_lookupenum(s, "google.protobuf.FileOptions.OptimizeMode"); + assert(e); + return e; +} + +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_end(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto_ExtensionRange(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_ExtensionRange_start(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto_ExtensionRange(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_enum_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_extension(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_extension_range(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_field(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_nested_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_DescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_DescriptorProto(s), 7); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumDescriptorProto(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumDescriptorProto(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumDescriptorProto_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumDescriptorProto(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumOptions_allow_alias(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumOptions(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumOptions(s), 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueDescriptorProto(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_number(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueDescriptorProto(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueDescriptorProto(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_EnumValueOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_EnumValueOptions(s), 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_default_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 7); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_extendee(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_label(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_number(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldDescriptorProto_type_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldDescriptorProto(s), 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_ctype(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_deprecated(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_experimental_map_key(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 9); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_lazy(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_packed(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FieldOptions_weak(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FieldOptions(s), 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_dependency(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_enum_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_extension(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 7); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_message_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_package(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_public_dependency(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_service(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_source_code_info(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 9); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorProto_weak_dependency(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorProto(s), 11); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileDescriptorSet_file(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileDescriptorSet(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_cc_generic_services(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 16); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_go_package(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 11); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_generate_equals_and_hash(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 20); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_generic_services(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 17); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_multiple_files(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 10); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_outer_classname(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_java_package(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_optimize_for(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 9); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_py_generic_services(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 18); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_FileOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_FileOptions(s), 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_message_set_wire_format(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MessageOptions(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_no_standard_descriptor_accessor(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MessageOptions(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MessageOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MessageOptions(s), 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_input_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodDescriptorProto_output_type(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodDescriptorProto(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_MethodOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_MethodOptions(s), 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_method(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceDescriptorProto(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceDescriptorProto(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceDescriptorProto_options(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceDescriptorProto(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_ServiceOptions_uninterpreted_option(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_ServiceOptions(s), 999); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_leading_comments(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_path(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_span(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_Location_trailing_comments(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo_Location(s), 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_SourceCodeInfo_location(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_SourceCodeInfo(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_NamePart_is_extension(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption_NamePart(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_NamePart_name_part(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption_NamePart(s), 1); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_aggregate_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 8); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_double_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 6); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_identifier_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 3); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_name(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 2); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_negative_int_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 5); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_positive_int_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 4); } +UPB_INLINE const upb_fielddef *upbdefs_google_protobuf_UninterpretedOption_string_value(const upb_symtab *s) { return upb_msgdef_itof(upbdefs_google_protobuf_UninterpretedOption(s), 7); } + +#ifdef __cplusplus +}; // extern "C" +#endif + +#ifdef __cplusplus + +namespace upbdefs { +namespace google { +namespace protobuf { +namespace descriptor { +inline upb::reffed_ptr SymbolTable() { + const upb::SymbolTable* s = upbdefs_google_protobuf_descriptor(&s); + return upb::reffed_ptr(s, &s); +} +} // namespace descriptor +} // namespace protobuf +} // namespace google + +#define RETURN_REFFED(type, func) \ + const type* obj = func(upbdefs::google::protobuf::descriptor::SymbolTable().get()); \ + return upb::reffed_ptr(obj); + +namespace google { +namespace protobuf { +namespace DescriptorProto { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_DescriptorProto) } +inline upb::reffed_ptr enum_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_enum_type) } +inline upb::reffed_ptr extension() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_extension) } +inline upb::reffed_ptr extension_range() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_extension_range) } +inline upb::reffed_ptr field() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_field) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_name) } +inline upb::reffed_ptr nested_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_nested_type) } +inline upb::reffed_ptr options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_options) } +} // namespace DescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace DescriptorProto { +namespace ExtensionRange { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_DescriptorProto_ExtensionRange) } +inline upb::reffed_ptr end() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_ExtensionRange_end) } +inline upb::reffed_ptr start() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_DescriptorProto_ExtensionRange_start) } +} // namespace ExtensionRange +} // namespace DescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace EnumDescriptorProto { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumDescriptorProto) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumDescriptorProto_name) } +inline upb::reffed_ptr options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumDescriptorProto_options) } +inline upb::reffed_ptr value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumDescriptorProto_value) } +} // namespace EnumDescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace EnumOptions { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumOptions) } +inline upb::reffed_ptr allow_alias() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumOptions_allow_alias) } +inline upb::reffed_ptr uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumOptions_uninterpreted_option) } +} // namespace EnumOptions +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace EnumValueDescriptorProto { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumValueDescriptorProto) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueDescriptorProto_name) } +inline upb::reffed_ptr number() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueDescriptorProto_number) } +inline upb::reffed_ptr options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueDescriptorProto_options) } +} // namespace EnumValueDescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace EnumValueOptions { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_EnumValueOptions) } +inline upb::reffed_ptr uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_EnumValueOptions_uninterpreted_option) } +} // namespace EnumValueOptions +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace FieldDescriptorProto { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FieldDescriptorProto) } +inline upb::reffed_ptr default_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_default_value) } +inline upb::reffed_ptr extendee() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_extendee) } +inline upb::reffed_ptr label() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_label) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_name) } +inline upb::reffed_ptr number() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_number) } +inline upb::reffed_ptr options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_options) } +inline upb::reffed_ptr type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_type) } +inline upb::reffed_ptr type_name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldDescriptorProto_type_name) } +inline upb::reffed_ptr Label() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FieldDescriptorProto_Label) } +inline upb::reffed_ptr Type() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FieldDescriptorProto_Type) } +} // namespace FieldDescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace FieldOptions { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FieldOptions) } +inline upb::reffed_ptr ctype() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_ctype) } +inline upb::reffed_ptr deprecated() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_deprecated) } +inline upb::reffed_ptr experimental_map_key() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_experimental_map_key) } +inline upb::reffed_ptr lazy() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_lazy) } +inline upb::reffed_ptr packed() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_packed) } +inline upb::reffed_ptr uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_uninterpreted_option) } +inline upb::reffed_ptr weak() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FieldOptions_weak) } +inline upb::reffed_ptr CType() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FieldOptions_CType) } +} // namespace FieldOptions +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace FileDescriptorProto { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FileDescriptorProto) } +inline upb::reffed_ptr dependency() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_dependency) } +inline upb::reffed_ptr enum_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_enum_type) } +inline upb::reffed_ptr extension() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_extension) } +inline upb::reffed_ptr message_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_message_type) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_name) } +inline upb::reffed_ptr options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_options) } +inline upb::reffed_ptr package() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_package) } +inline upb::reffed_ptr public_dependency() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_public_dependency) } +inline upb::reffed_ptr service() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_service) } +inline upb::reffed_ptr source_code_info() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_source_code_info) } +inline upb::reffed_ptr weak_dependency() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorProto_weak_dependency) } +} // namespace FileDescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace FileDescriptorSet { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FileDescriptorSet) } +inline upb::reffed_ptr file() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileDescriptorSet_file) } +} // namespace FileDescriptorSet +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace FileOptions { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_FileOptions) } +inline upb::reffed_ptr cc_generic_services() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_cc_generic_services) } +inline upb::reffed_ptr go_package() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_go_package) } +inline upb::reffed_ptr java_generate_equals_and_hash() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_generate_equals_and_hash) } +inline upb::reffed_ptr java_generic_services() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_generic_services) } +inline upb::reffed_ptr java_multiple_files() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_multiple_files) } +inline upb::reffed_ptr java_outer_classname() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_outer_classname) } +inline upb::reffed_ptr java_package() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_java_package) } +inline upb::reffed_ptr optimize_for() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_optimize_for) } +inline upb::reffed_ptr py_generic_services() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_py_generic_services) } +inline upb::reffed_ptr uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_FileOptions_uninterpreted_option) } +inline upb::reffed_ptr OptimizeMode() { RETURN_REFFED(upb::EnumDef, upbdefs_google_protobuf_FileOptions_OptimizeMode) } +} // namespace FileOptions +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace MessageOptions { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_MessageOptions) } +inline upb::reffed_ptr message_set_wire_format() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MessageOptions_message_set_wire_format) } +inline upb::reffed_ptr no_standard_descriptor_accessor() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MessageOptions_no_standard_descriptor_accessor) } +inline upb::reffed_ptr uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MessageOptions_uninterpreted_option) } +} // namespace MessageOptions +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace MethodDescriptorProto { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_MethodDescriptorProto) } +inline upb::reffed_ptr input_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_input_type) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_name) } +inline upb::reffed_ptr options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_options) } +inline upb::reffed_ptr output_type() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodDescriptorProto_output_type) } +} // namespace MethodDescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace MethodOptions { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_MethodOptions) } +inline upb::reffed_ptr uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_MethodOptions_uninterpreted_option) } +} // namespace MethodOptions +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace ServiceDescriptorProto { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_ServiceDescriptorProto) } +inline upb::reffed_ptr method() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceDescriptorProto_method) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceDescriptorProto_name) } +inline upb::reffed_ptr options() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceDescriptorProto_options) } +} // namespace ServiceDescriptorProto +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace ServiceOptions { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_ServiceOptions) } +inline upb::reffed_ptr uninterpreted_option() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_ServiceOptions_uninterpreted_option) } +} // namespace ServiceOptions +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace SourceCodeInfo { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_SourceCodeInfo) } +inline upb::reffed_ptr location() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_location) } +} // namespace SourceCodeInfo +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace SourceCodeInfo { +namespace Location { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_SourceCodeInfo_Location) } +inline upb::reffed_ptr leading_comments() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_leading_comments) } +inline upb::reffed_ptr path() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_path) } +inline upb::reffed_ptr span() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_span) } +inline upb::reffed_ptr trailing_comments() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_SourceCodeInfo_Location_trailing_comments) } +} // namespace Location +} // namespace SourceCodeInfo +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace UninterpretedOption { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_UninterpretedOption) } +inline upb::reffed_ptr aggregate_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_aggregate_value) } +inline upb::reffed_ptr double_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_double_value) } +inline upb::reffed_ptr identifier_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_identifier_value) } +inline upb::reffed_ptr name() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_name) } +inline upb::reffed_ptr negative_int_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_negative_int_value) } +inline upb::reffed_ptr positive_int_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_positive_int_value) } +inline upb::reffed_ptr string_value() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_string_value) } +} // namespace UninterpretedOption +} // namespace protobuf +} // namespace google + +namespace google { +namespace protobuf { +namespace UninterpretedOption { +namespace NamePart { +inline upb::reffed_ptr MessageDef() { RETURN_REFFED(upb::MessageDef, upbdefs_google_protobuf_UninterpretedOption_NamePart) } +inline upb::reffed_ptr is_extension() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_NamePart_is_extension) } +inline upb::reffed_ptr name_part() { RETURN_REFFED(upb::FieldDef, upbdefs_google_protobuf_UninterpretedOption_NamePart_name_part) } +} // namespace NamePart +} // namespace UninterpretedOption +} // namespace protobuf +} // namespace google + +} // namespace upbdefs + + +#undef RETURN_REFFED +#endif // __cplusplus + +#endif // GOOGLE_PROTOBUF_DESCRIPTOR_UPB_H_ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2010-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * A upb_handlers is like a virtual table for a upb_msgdef. Each field of the + * message can have associated functions that will be called when we are + * parsing or visiting a stream of data. This is similar to how handlers work + * in SAX (the Simple API for XML). + * + * The handlers have no idea where the data is coming from, so a single set of + * handlers could be used with two completely different data sources (for + * example, a parser and a visitor over in-memory objects). This decoupling is + * the most important feature of upb, because it allows parsers and serializers + * to be highly reusable. + * + * This is a mixed C/C++ interface that offers a full API to both languages. + * See the top-level README for more information. + */ + +#ifndef UPB_HANDLERS_H +#define UPB_HANDLERS_H + + +#ifdef __cplusplus +namespace upb { +class BufferHandle; +class BytesHandler; +class HandlerAttributes; +class Handlers; +template class Handler; +template struct CanonicalType; +} // namespace upb +#endif + +UPB_DECLARE_TYPE(upb::BufferHandle, upb_bufhandle); +UPB_DECLARE_TYPE(upb::BytesHandler, upb_byteshandler); +UPB_DECLARE_TYPE(upb::HandlerAttributes, upb_handlerattr); +UPB_DECLARE_TYPE(upb::Handlers, upb_handlers); + +// The maximum depth that the handler graph can have. This is a resource limit +// for the C stack since we sometimes need to recursively traverse the graph. +// Cycles are ok; the traversal will stop when it detects a cycle, but we must +// hit the cycle before the maximum depth is reached. +// +// If having a single static limit is too inflexible, we can add another variant +// of Handlers::Freeze that allows specifying this as a parameter. +#define UPB_MAX_HANDLER_DEPTH 64 + +// All the different types of handlers that can be registered. +// Only needed for the advanced functions in upb::Handlers. +typedef enum { + UPB_HANDLER_INT32, + UPB_HANDLER_INT64, + UPB_HANDLER_UINT32, + UPB_HANDLER_UINT64, + UPB_HANDLER_FLOAT, + UPB_HANDLER_DOUBLE, + UPB_HANDLER_BOOL, + UPB_HANDLER_STARTSTR, + UPB_HANDLER_STRING, + UPB_HANDLER_ENDSTR, + UPB_HANDLER_STARTSUBMSG, + UPB_HANDLER_ENDSUBMSG, + UPB_HANDLER_STARTSEQ, + UPB_HANDLER_ENDSEQ, +} upb_handlertype_t; + +#define UPB_HANDLER_MAX (UPB_HANDLER_ENDSEQ+1) + +#define UPB_BREAK NULL + +// A convenient definition for when no closure is needed. +extern char _upb_noclosure; +#define UPB_NO_CLOSURE &_upb_noclosure + +// A selector refers to a specific field handler in the Handlers object +// (for example: the STARTSUBMSG handler for field "field15"). +typedef int32_t upb_selector_t; + +UPB_BEGIN_EXTERN_C + +// Forward-declares for C inline accessors. We need to declare these here +// so we can "friend" them in the class declarations in C++. +UPB_INLINE upb_func *upb_handlers_gethandler(const upb_handlers *h, + upb_selector_t s); +UPB_INLINE const void *upb_handlerattr_handlerdata(const upb_handlerattr *attr); +UPB_INLINE const void *upb_handlers_gethandlerdata(const upb_handlers *h, + upb_selector_t s); + +UPB_INLINE void upb_bufhandle_init(upb_bufhandle *h); +UPB_INLINE void upb_bufhandle_setobj(upb_bufhandle *h, const void *obj, + const void *type); +UPB_INLINE void upb_bufhandle_setbuf(upb_bufhandle *h, const char *buf, + size_t ofs); +UPB_INLINE const void *upb_bufhandle_obj(const upb_bufhandle *h); +UPB_INLINE const void *upb_bufhandle_objtype(const upb_bufhandle *h); +UPB_INLINE const char *upb_bufhandle_buf(const upb_bufhandle *h); + +UPB_END_EXTERN_C + + +// Static selectors for upb::Handlers. +#define UPB_STARTMSG_SELECTOR 0 +#define UPB_ENDMSG_SELECTOR 1 +#define UPB_STATIC_SELECTOR_COUNT 2 + +// Static selectors for upb::BytesHandler. +#define UPB_STARTSTR_SELECTOR 0 +#define UPB_STRING_SELECTOR 1 +#define UPB_ENDSTR_SELECTOR 2 + +typedef void upb_handlerfree(void *d); + +// A set of attributes that accompanies a handler's function pointer. +UPB_DEFINE_CLASS0(upb::HandlerAttributes, + public: + HandlerAttributes(); + ~HandlerAttributes(); + + // Sets the handler data that will be passed as the second parameter of the + // handler. To free this pointer when the handlers are freed, call + // Handlers::AddCleanup(). + bool SetHandlerData(const void *handler_data); + const void* handler_data() const; + + // Use this to specify the type of the closure. This will be checked against + // all other closure types for handler that use the same closure. + // Registration will fail if this does not match all other non-NULL closure + // types. + bool SetClosureType(const void *closure_type); + const void* closure_type() const; + + // Use this to specify the type of the returned closure. Only used for + // Start*{String,SubMessage,Sequence} handlers. This must match the closure + // type of any handlers that use it (for example, the StringBuf handler must + // match the closure returned from StartString). + bool SetReturnClosureType(const void *return_closure_type); + const void* return_closure_type() const; + + // Set to indicate that the handler always returns "ok" (either "true" or a + // non-NULL closure). This is a hint that can allow code generators to + // generate more efficient code. + bool SetAlwaysOk(bool always_ok); + bool always_ok() const; + + private: + friend UPB_INLINE const void * ::upb_handlerattr_handlerdata( + const upb_handlerattr *attr); +, +UPB_DEFINE_STRUCT0(upb_handlerattr, + const void *handler_data_; + const void *closure_type_; + const void *return_closure_type_; + bool alwaysok_; +)); + +#define UPB_HANDLERATTR_INITIALIZER {NULL, NULL, NULL, false} + +typedef struct { + upb_func *func; + // It is wasteful to include the entire attributes here: + // + // * Some of the information is redundant (like storing the closure type + // separately for each handler that must match). + // * Some of the info is only needed prior to freeze() (like closure types). + // * alignment padding wastes a lot of space for alwaysok_. + // + // If/when the size and locality of handlers is an issue, we can optimize this + // not to store the entire attr like this. We do not expose the table's + // layout to allow this optimization in the future. + upb_handlerattr attr; +} upb_handlers_tabent; + +// Extra information about a buffer that is passed to a StringBuf handler. +// TODO(haberman): allow the handle to be pinned so that it will outlive +// the handler invocation. +UPB_DEFINE_CLASS0(upb::BufferHandle, + public: + BufferHandle(); + ~BufferHandle(); + + // The beginning of the buffer. This may be different than the pointer + // passed to a StringBuf handler because the handler may receive data + // that is from the middle or end of a larger buffer. + const char* buffer() const; + + // The offset within the attached object where this buffer begins. Only + // meaningful if there is an attached object. + size_t object_offset() const; + + // Note that object_offset is the offset of "buf" within the attached object. + void SetBuffer(const char* buf, size_t object_offset); + + // The BufferHandle can have an "attached object", which can be used to + // tunnel through a pointer to the buffer's underlying representation. + template + void SetAttachedObject(const T* obj); + + // Returns NULL if the attached object is not of this type. + template + const T* GetAttachedObject() const; + + private: + friend UPB_INLINE void ::upb_bufhandle_init(upb_bufhandle *h); + friend UPB_INLINE void ::upb_bufhandle_setobj(upb_bufhandle *h, + const void *obj, + const void *type); + friend UPB_INLINE void ::upb_bufhandle_setbuf(upb_bufhandle *h, + const char *buf, size_t ofs); + friend UPB_INLINE const void* ::upb_bufhandle_obj(const upb_bufhandle *h); + friend UPB_INLINE const void* ::upb_bufhandle_objtype( + const upb_bufhandle *h); + friend UPB_INLINE const char* ::upb_bufhandle_buf(const upb_bufhandle *h); +, +UPB_DEFINE_STRUCT0(upb_bufhandle, + const char *buf_; + const void *obj_; + const void *objtype_; + size_t objofs_; +)); + +// A upb::Handlers object represents the set of handlers associated with a +// message in the graph of messages. You can think of it as a big virtual +// table with functions corresponding to all the events that can fire while +// parsing or visiting a message of a specific type. +// +// Any handlers that are not set behave as if they had successfully consumed +// the value. Any unset Start* handlers will propagate their closure to the +// inner frame. +// +// The easiest way to create the *Handler objects needed by the Set* methods is +// with the UpbBind() and UpbMakeHandler() macros; see below. +UPB_DEFINE_CLASS1(upb::Handlers, upb::RefCounted, + public: + typedef upb_selector_t Selector; + typedef upb_handlertype_t Type; + + typedef Handler StartFieldHandler; + typedef Handler EndFieldHandler; + typedef Handler StartMessageHandler; + typedef Handler EndMessageHandler; + typedef Handler StartStringHandler; + typedef Handler StringHandler; + + template struct ValueHandler { + typedef Handler H; + }; + + typedef ValueHandler::H Int32Handler; + typedef ValueHandler::H Int64Handler; + typedef ValueHandler::H UInt32Handler; + typedef ValueHandler::H UInt64Handler; + typedef ValueHandler::H FloatHandler; + typedef ValueHandler::H DoubleHandler; + typedef ValueHandler::H BoolHandler; + + // Any function pointer can be converted to this and converted back to its + // correct type. + typedef void GenericFunction(); + + typedef void HandlersCallback(const void *closure, upb_handlers *h); + + // Returns a new handlers object for the given frozen msgdef. + // Returns NULL if memory allocation failed. + static reffed_ptr New(const MessageDef *m); + + // Convenience function for registering a graph of handlers that mirrors the + // graph of msgdefs for some message. For "m" and all its children a new set + // of handlers will be created and the given callback will be invoked, + // allowing the client to register handlers for this message. Note that any + // subhandlers set by the callback will be overwritten. + static reffed_ptr NewFrozen(const MessageDef *m, + HandlersCallback *callback, + const void *closure); + + // Functionality from upb::RefCounted. + bool IsFrozen() const; + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void *from, const void *to) const; + void CheckRef(const void *owner) const; + + // All handler registration functions return bool to indicate success or + // failure; details about failures are stored in this status object. If a + // failure does occur, it must be cleared before the Handlers are frozen, + // otherwise the freeze() operation will fail. The functions may *only* be + // used while the Handlers are mutable. + const Status* status(); + void ClearError(); + + // Call to freeze these Handlers. Requires that any SubHandlers are already + // frozen. For cycles, you must use the static version below and freeze the + // whole graph at once. + bool Freeze(Status* s); + + // Freezes the given set of handlers. You may not freeze a handler without + // also freezing any handlers they point to. + static bool Freeze(Handlers*const* handlers, int n, Status* s); + static bool Freeze(const std::vector& handlers, Status* s); + + // Returns the msgdef associated with this handlers object. + const MessageDef* message_def() const; + + // Adds the given pointer and function to the list of cleanup functions that + // will be run when these handlers are freed. If this pointer has previously + // been registered, the function returns false and does nothing. + bool AddCleanup(void *ptr, upb_handlerfree *cleanup); + + // Sets the startmsg handler for the message, which is defined as follows: + // + // bool startmsg(MyType* closure) { + // // Called when the message begins. Returns true if processing should + // // continue. + // return true; + // } + bool SetStartMessageHandler(const StartMessageHandler& handler); + + // Sets the endmsg handler for the message, which is defined as follows: + // + // bool endmsg(MyType* closure, upb_status *status) { + // // Called when processing of this message ends, whether in success or + // // failure. "status" indicates the final status of processing, and + // // can also be modified in-place to update the final status. + // } + bool SetEndMessageHandler(const EndMessageHandler& handler); + + // Sets the value handler for the given field, which is defined as follows + // (this is for an int32 field; other field types will pass their native + // C/C++ type for "val"): + // + // bool OnValue(MyClosure* c, const MyHandlerData* d, int32_t val) { + // // Called when the field's value is encountered. "d" contains + // // whatever data was bound to this field when it was registered. + // // Returns true if processing should continue. + // return true; + // } + // + // handers->SetInt32Handler(f, UpbBind(OnValue, new MyHandlerData(...))); + // + // The value type must exactly match f->type(). + // For example, a handler that takes an int32_t parameter may only be used for + // fields of type UPB_TYPE_INT32 and UPB_TYPE_ENUM. + // + // Returns false if the handler failed to register; in this case the cleanup + // handler (if any) will be called immediately. + bool SetInt32Handler (const FieldDef* f, const Int32Handler& h); + bool SetInt64Handler (const FieldDef* f, const Int64Handler& h); + bool SetUInt32Handler(const FieldDef* f, const UInt32Handler& h); + bool SetUInt64Handler(const FieldDef* f, const UInt64Handler& h); + bool SetFloatHandler (const FieldDef* f, const FloatHandler& h); + bool SetDoubleHandler(const FieldDef* f, const DoubleHandler& h); + bool SetBoolHandler (const FieldDef* f, const BoolHandler& h); + + // Like the previous, but templated on the type on the value (ie. int32). + // This is mostly useful to call from other templates. To call this you must + // specify the template parameter explicitly, ie: + // h->SetValueHandler(f, UpbBind(MyHandler, MyData)); + template + bool SetValueHandler( + const FieldDef *f, + const typename ValueHandler::Type>::H& handler); + + // Sets handlers for a string field, which are defined as follows: + // + // MySubClosure* startstr(MyClosure* c, const MyHandlerData* d, + // size_t size_hint) { + // // Called when a string value begins. The return value indicates the + // // closure for the string. "size_hint" indicates the size of the + // // string if it is known, however if the string is length-delimited + // // and the end-of-string is not available size_hint will be zero. + // // This case is indistinguishable from the case where the size is + // // known to be zero. + // // + // // TODO(haberman): is it important to distinguish these cases? + // // If we had ssize_t as a type we could make -1 "unknown", but + // // ssize_t is POSIX (not ANSI) and therefore less portable. + // // In practice I suspect it won't be important to distinguish. + // return closure; + // } + // + // size_t str(MyClosure* closure, const MyHandlerData* d, + // const char *str, size_t len) { + // // Called for each buffer of string data; the multiple physical buffers + // // are all part of the same logical string. The return value indicates + // // how many bytes were consumed. If this number is less than "len", + // // this will also indicate that processing should be halted for now, + // // like returning false or UPB_BREAK from any other callback. If + // // number is greater than "len", the excess bytes will be skipped over + // // and not passed to the callback. + // return len; + // } + // + // bool endstr(MyClosure* c, const MyHandlerData* d) { + // // Called when a string value ends. Return value indicates whether + // // processing should continue. + // return true; + // } + bool SetStartStringHandler(const FieldDef* f, const StartStringHandler& h); + bool SetStringHandler(const FieldDef* f, const StringHandler& h); + bool SetEndStringHandler(const FieldDef* f, const EndFieldHandler& h); + + // Sets the startseq handler, which is defined as follows: + // + // MySubClosure *startseq(MyClosure* c, const MyHandlerData* d) { + // // Called when a sequence (repeated field) begins. The returned + // // pointer indicates the closure for the sequence (or UPB_BREAK + // // to interrupt processing). + // return closure; + // } + // + // h->SetStartSequenceHandler(f, UpbBind(startseq, new MyHandlerData(...))); + // + // Returns "false" if "f" does not belong to this message or is not a + // repeated field. + bool SetStartSequenceHandler(const FieldDef* f, const StartFieldHandler& h); + + // Sets the startsubmsg handler for the given field, which is defined as + // follows: + // + // MySubClosure* startsubmsg(MyClosure* c, const MyHandlerData* d) { + // // Called when a submessage begins. The returned pointer indicates the + // // closure for the sequence (or UPB_BREAK to interrupt processing). + // return closure; + // } + // + // h->SetStartSubMessageHandler(f, UpbBind(startsubmsg, + // new MyHandlerData(...))); + // + // Returns "false" if "f" does not belong to this message or is not a + // submessage/group field. + bool SetStartSubMessageHandler(const FieldDef* f, const StartFieldHandler& h); + + // Sets the endsubmsg handler for the given field, which is defined as + // follows: + // + // bool endsubmsg(MyClosure* c, const MyHandlerData* d) { + // // Called when a submessage ends. Returns true to continue processing. + // return true; + // } + // + // Returns "false" if "f" does not belong to this message or is not a + // submessage/group field. + bool SetEndSubMessageHandler(const FieldDef *f, const EndFieldHandler &h); + + // Starts the endsubseq handler for the given field, which is defined as + // follows: + // + // bool endseq(MyClosure* c, const MyHandlerData* d) { + // // Called when a sequence ends. Returns true continue processing. + // return true; + // } + // + // Returns "false" if "f" does not belong to this message or is not a + // repeated field. + bool SetEndSequenceHandler(const FieldDef* f, const EndFieldHandler& h); + + // Sets or gets the object that specifies handlers for the given field, which + // must be a submessage or group. Returns NULL if no handlers are set. + bool SetSubHandlers(const FieldDef* f, const Handlers* sub); + const Handlers* GetSubHandlers(const FieldDef* f) const; + + // Equivalent to GetSubHandlers, but takes the STARTSUBMSG selector for the + // field. + const Handlers* GetSubHandlers(Selector startsubmsg) const; + + // A selector refers to a specific field handler in the Handlers object + // (for example: the STARTSUBMSG handler for field "field15"). + // On success, returns true and stores the selector in "s". + // If the FieldDef or Type are invalid, returns false. + // The returned selector is ONLY valid for Handlers whose MessageDef + // contains this FieldDef. + static bool GetSelector(const FieldDef* f, Type type, Selector* s); + + // Given a START selector of any kind, returns the corresponding END selector. + static Selector GetEndSelector(Selector start_selector); + + // Returns the function pointer for this handler. It is the client's + // responsibility to cast to the correct function type before calling it. + GenericFunction* GetHandler(Selector selector); + + // Sets the given attributes to the attributes for this selector. + bool GetAttributes(Selector selector, HandlerAttributes* attr); + + // Returns the handler data that was registered with this handler. + const void* GetHandlerData(Selector selector); + + // Could add any of the following functions as-needed, with some minor + // implementation changes: + // + // const FieldDef* GetFieldDef(Selector selector); + // static bool IsSequence(Selector selector); + + private: + UPB_DISALLOW_POD_OPS(Handlers, upb::Handlers); + + friend UPB_INLINE GenericFunction *::upb_handlers_gethandler( + const upb_handlers *h, upb_selector_t s); + friend UPB_INLINE const void *::upb_handlers_gethandlerdata( + const upb_handlers *h, upb_selector_t s); + +, +UPB_DEFINE_STRUCT(upb_handlers, upb_refcounted, + const upb_msgdef *msg; + const upb_handlers **sub; + const void *top_closure_type; + upb_inttable cleanup_; + upb_status status_; // Used only when mutable. + upb_handlers_tabent table[1]; // Dynamically-sized field handler array. +)); + + +#ifdef __cplusplus + +namespace upb { + +// Convenience macros for creating a Handler object that is wrapped with a +// type-safe wrapper function that converts the "void*" parameters/returns +// of the underlying C API into nice C++ function. +// +// Sample usage: +// void OnValue1(MyClosure* c, const MyHandlerData* d, int32_t val) { +// // do stuff ... +// } +// +// // Handler that doesn't need any data bound to it. +// void OnValue2(MyClosure* c, int32_t val) { +// // do stuff ... +// } +// +// // Handler that returns bool so it can return failure if necessary. +// bool OnValue3(MyClosure* c, int32_t val) { +// // do stuff ... +// return ok; +// } +// +// // Member function handler. +// class MyClosure { +// public: +// void OnValue(int32_t val) { +// // do stuff ... +// } +// }; +// +// // Takes ownership of the MyHandlerData. +// handlers->SetInt32Handler(f1, UpbBind(OnValue1, new MyHandlerData(...))); +// handlers->SetInt32Handler(f2, UpbMakeHandler(OnValue2)); +// handlers->SetInt32Handler(f1, UpbMakeHandler(OnValue3)); +// handlers->SetInt32Handler(f2, UpbMakeHandler(&MyClosure::OnValue)); + +#ifdef UPB_CXX11 + +// In C++11, the "template" disambiguator can appear even outside templates, +// so all calls can safely use this pair of macros. + +#define UpbMakeHandler(f) upb::MatchFunc(f).template GetFunc() + +// We have to be careful to only evaluate "d" once. +#define UpbBind(f, d) upb::MatchFunc(f).template GetFunc((d)) + +#else + +// Prior to C++11, the "template" disambiguator may only appear inside a +// template, so the regular macro must not use "template" + +#define UpbMakeHandler(f) upb::MatchFunc(f).GetFunc() + +#define UpbBind(f, d) upb::MatchFunc(f).GetFunc((d)) + +#endif // UPB_CXX11 + +// This macro must be used in C++98 for calls from inside a template. But we +// define this variant in all cases; code that wants to be compatible with both +// C++98 and C++11 should always use this macro when calling from a template. +#define UpbMakeHandlerT(f) upb::MatchFunc(f).template GetFunc() + +// We have to be careful to only evaluate "d" once. +#define UpbBindT(f, d) upb::MatchFunc(f).template GetFunc((d)) + +// Handler: a struct that contains the (handler, data, deleter) tuple that is +// used to register all handlers. Users can Make() these directly but it's +// more convenient to use the UpbMakeHandler/UpbBind macros above. +template class Handler { + public: + // The underlying, handler function signature that upb uses internally. + typedef T FuncPtr; + + // Intentionally implicit. + template Handler(F func); + ~Handler(); + + private: + void AddCleanup(Handlers* h) const { + if (cleanup_func_) { + bool ok = h->AddCleanup(cleanup_data_, cleanup_func_); + UPB_ASSERT_VAR(ok, ok); + } + } + + UPB_DISALLOW_COPY_AND_ASSIGN(Handler); + friend class Handlers; + FuncPtr handler_; + mutable HandlerAttributes attr_; + mutable bool registered_; + void *cleanup_data_; + upb_handlerfree *cleanup_func_; +}; + +} // namespace upb + +#endif // __cplusplus + +UPB_BEGIN_EXTERN_C + +// Native C API. + +// Handler function typedefs. +typedef bool upb_startmsg_handlerfunc(void *c, const void*); +typedef bool upb_endmsg_handlerfunc(void *c, const void *, upb_status *status); +typedef void* upb_startfield_handlerfunc(void *c, const void *hd); +typedef bool upb_endfield_handlerfunc(void *c, const void *hd); +typedef bool upb_int32_handlerfunc(void *c, const void *hd, int32_t val); +typedef bool upb_int64_handlerfunc(void *c, const void *hd, int64_t val); +typedef bool upb_uint32_handlerfunc(void *c, const void *hd, uint32_t val); +typedef bool upb_uint64_handlerfunc(void *c, const void *hd, uint64_t val); +typedef bool upb_float_handlerfunc(void *c, const void *hd, float val); +typedef bool upb_double_handlerfunc(void *c, const void *hd, double val); +typedef bool upb_bool_handlerfunc(void *c, const void *hd, bool val); +typedef void *upb_startstr_handlerfunc(void *c, const void *hd, + size_t size_hint); +typedef size_t upb_string_handlerfunc(void *c, const void *hd, const char *buf, + size_t n, const upb_bufhandle* handle); + +// upb_bufhandle +size_t upb_bufhandle_objofs(const upb_bufhandle *h); + +// upb_handlerattr +void upb_handlerattr_init(upb_handlerattr *attr); +void upb_handlerattr_uninit(upb_handlerattr *attr); + +bool upb_handlerattr_sethandlerdata(upb_handlerattr *attr, const void *hd); +bool upb_handlerattr_setclosuretype(upb_handlerattr *attr, const void *type); +const void *upb_handlerattr_closuretype(const upb_handlerattr *attr); +bool upb_handlerattr_setreturnclosuretype(upb_handlerattr *attr, + const void *type); +const void *upb_handlerattr_returnclosuretype(const upb_handlerattr *attr); +bool upb_handlerattr_setalwaysok(upb_handlerattr *attr, bool alwaysok); +bool upb_handlerattr_alwaysok(const upb_handlerattr *attr); + +UPB_INLINE const void *upb_handlerattr_handlerdata( + const upb_handlerattr *attr) { + return attr->handler_data_; +} + +// upb_handlers +typedef void upb_handlers_callback(const void *closure, upb_handlers *h); +upb_handlers *upb_handlers_new(const upb_msgdef *m, + const void *owner); +const upb_handlers *upb_handlers_newfrozen(const upb_msgdef *m, + const void *owner, + upb_handlers_callback *callback, + const void *closure); +bool upb_handlers_isfrozen(const upb_handlers *h); +void upb_handlers_ref(const upb_handlers *h, const void *owner); +void upb_handlers_unref(const upb_handlers *h, const void *owner); +void upb_handlers_donateref(const upb_handlers *h, const void *from, + const void *to); +void upb_handlers_checkref(const upb_handlers *h, const void *owner); + +const upb_status *upb_handlers_status(upb_handlers *h); +void upb_handlers_clearerr(upb_handlers *h); +const upb_msgdef *upb_handlers_msgdef(const upb_handlers *h); +bool upb_handlers_addcleanup(upb_handlers *h, void *p, upb_handlerfree *hfree); + +bool upb_handlers_setstartmsg(upb_handlers *h, upb_startmsg_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendmsg(upb_handlers *h, upb_endmsg_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setint32(upb_handlers *h, const upb_fielddef *f, + upb_int32_handlerfunc *func, upb_handlerattr *attr); +bool upb_handlers_setint64(upb_handlers *h, const upb_fielddef *f, + upb_int64_handlerfunc *func, upb_handlerattr *attr); +bool upb_handlers_setuint32(upb_handlers *h, const upb_fielddef *f, + upb_uint32_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setuint64(upb_handlers *h, const upb_fielddef *f, + upb_uint64_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setfloat(upb_handlers *h, const upb_fielddef *f, + upb_float_handlerfunc *func, upb_handlerattr *attr); +bool upb_handlers_setdouble(upb_handlers *h, const upb_fielddef *f, + upb_double_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setbool(upb_handlers *h, const upb_fielddef *f, + upb_bool_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstartstr(upb_handlers *h, const upb_fielddef *f, + upb_startstr_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstring(upb_handlers *h, const upb_fielddef *f, + upb_string_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendstr(upb_handlers *h, const upb_fielddef *f, + upb_endfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstartseq(upb_handlers *h, const upb_fielddef *f, + upb_startfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setstartsubmsg(upb_handlers *h, const upb_fielddef *f, + upb_startfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendsubmsg(upb_handlers *h, const upb_fielddef *f, + upb_endfield_handlerfunc *func, + upb_handlerattr *attr); +bool upb_handlers_setendseq(upb_handlers *h, const upb_fielddef *f, + upb_endfield_handlerfunc *func, + upb_handlerattr *attr); + +bool upb_handlers_setsubhandlers(upb_handlers *h, const upb_fielddef *f, + const upb_handlers *sub); +const upb_handlers *upb_handlers_getsubhandlers(const upb_handlers *h, + const upb_fielddef *f); +const upb_handlers *upb_handlers_getsubhandlers_sel(const upb_handlers *h, + upb_selector_t sel); + +UPB_INLINE upb_func *upb_handlers_gethandler(const upb_handlers *h, + upb_selector_t s) { + return (upb_func *)h->table[s].func; +} + +bool upb_handlers_getattr(const upb_handlers *h, upb_selector_t s, + upb_handlerattr *attr); + +UPB_INLINE const void *upb_handlers_gethandlerdata(const upb_handlers *h, + upb_selector_t s) { + return upb_handlerattr_handlerdata(&h->table[s].attr); +} + +// Handler types for single fields. +// Right now we only have one for TYPE_BYTES but ones for other types +// should follow. +// +// These follow the same handlers protocol for fields of a message. +UPB_DEFINE_CLASS0(upb::BytesHandler, + public: + BytesHandler(); + ~BytesHandler(); +, +UPB_DEFINE_STRUCT0(upb_byteshandler, + upb_handlers_tabent table[3]; +)); + +void upb_byteshandler_init(upb_byteshandler *h); +void upb_byteshandler_uninit(upb_byteshandler *h); + +// Caller must ensure that "d" outlives the handlers. +// TODO(haberman): support handlerfree function for the data. +// TODO(haberman): should this have a "freeze" operation? It's not necessary +// for memory management, but could be useful to force immutability and provide +// a convenient moment to verify that all registration succeeded. +bool upb_byteshandler_setstartstr(upb_byteshandler *h, + upb_startstr_handlerfunc *func, void *d); +bool upb_byteshandler_setstring(upb_byteshandler *h, + upb_string_handlerfunc *func, void *d); +bool upb_byteshandler_setendstr(upb_byteshandler *h, + upb_endfield_handlerfunc *func, void *d); + +// "Static" methods +bool upb_handlers_freeze(upb_handlers *const *handlers, int n, upb_status *s); +upb_handlertype_t upb_handlers_getprimitivehandlertype(const upb_fielddef *f); +bool upb_handlers_getselector(const upb_fielddef *f, upb_handlertype_t type, + upb_selector_t *s); +UPB_INLINE upb_selector_t upb_handlers_getendselector(upb_selector_t start) { + return start + 1; +} + +// Internal-only. +uint32_t upb_handlers_selectorbaseoffset(const upb_fielddef *f); +uint32_t upb_handlers_selectorcount(const upb_fielddef *f); + +UPB_END_EXTERN_C + +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2011-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Inline definitions for handlers.h, which are particularly long and a bit + * tricky. + */ + +#ifndef UPB_HANDLERS_INL_H_ +#define UPB_HANDLERS_INL_H_ + +#include + +// Type detection and typedefs for integer types. +// For platforms where there are multiple 32-bit or 64-bit types, we need to be +// able to enumerate them so we can properly create overloads for all variants. +// +// If any platform existed where there were three integer types with the same +// size, this would have to become more complicated. For example, short, int, +// and long could all be 32-bits. Even more diabolically, short, int, long, +// and long long could all be 64 bits and still be standard-compliant. +// However, few platforms are this strange, and it's unlikely that upb will be +// used on the strangest ones. + +// Can't count on stdint.h limits like INT32_MAX, because in C++ these are +// only defined when __STDC_LIMIT_MACROS are defined before the *first* include +// of stdint.h. We can't guarantee that someone else didn't include these first +// without defining __STDC_LIMIT_MACROS. +#define UPB_INT32_MAX 0x7fffffffLL +#define UPB_INT32_MIN (-UPB_INT32_MAX - 1) +#define UPB_INT64_MAX 0x7fffffffffffffffLL +#define UPB_INT64_MIN (-UPB_INT64_MAX - 1) + +#if INT_MAX == UPB_INT32_MAX && INT_MIN == UPB_INT32_MIN +#define UPB_INT_IS_32BITS 1 +#endif + +#if LONG_MAX == UPB_INT32_MAX && LONG_MIN == UPB_INT32_MIN +#define UPB_LONG_IS_32BITS 1 +#endif + +#if LONG_MAX == UPB_INT64_MAX && LONG_MIN == UPB_INT64_MIN +#define UPB_LONG_IS_64BITS 1 +#endif + +#if LLONG_MAX == UPB_INT64_MAX && LLONG_MIN == UPB_INT64_MIN +#define UPB_LLONG_IS_64BITS 1 +#endif + +// We use macros instead of typedefs so we can undefine them later and avoid +// leaking them outside this header file. +#if UPB_INT_IS_32BITS +#define UPB_INT32_T int +#define UPB_UINT32_T unsigned int + +#if UPB_LONG_IS_32BITS +#define UPB_TWO_32BIT_TYPES 1 +#define UPB_INT32ALT_T long +#define UPB_UINT32ALT_T unsigned long +#endif // UPB_LONG_IS_32BITS + +#elif UPB_LONG_IS_32BITS // && !UPB_INT_IS_32BITS +#define UPB_INT32_T long +#define UPB_UINT32_T unsigned long +#endif // UPB_INT_IS_32BITS + + +#if UPB_LONG_IS_64BITS +#define UPB_INT64_T long +#define UPB_UINT64_T unsigned long + +#if UPB_LLONG_IS_64BITS +#define UPB_TWO_64BIT_TYPES 1 +#define UPB_INT64ALT_T long long +#define UPB_UINT64ALT_T unsigned long long +#endif // UPB_LLONG_IS_64BITS + +#elif UPB_LLONG_IS_64BITS // && !UPB_LONG_IS_64BITS +#define UPB_INT64_T long long +#define UPB_UINT64_T unsigned long long +#endif // UPB_LONG_IS_64BITS + +#undef UPB_INT32_MAX +#undef UPB_INT32_MIN +#undef UPB_INT64_MAX +#undef UPB_INT64_MIN +#undef UPB_INT_IS_32BITS +#undef UPB_LONG_IS_32BITS +#undef UPB_LONG_IS_64BITS +#undef UPB_LLONG_IS_64BITS + +// C inline methods. + +// upb_bufhandle +UPB_INLINE void upb_bufhandle_init(upb_bufhandle *h) { + h->obj_ = NULL; + h->objtype_ = NULL; + h->buf_ = NULL; + h->objofs_ = 0; +} +UPB_INLINE void upb_bufhandle_uninit(upb_bufhandle *h) { + UPB_UNUSED(h); +} +UPB_INLINE void upb_bufhandle_setobj(upb_bufhandle *h, const void *obj, + const void *type) { + h->obj_ = obj; + h->objtype_ = type; +} +UPB_INLINE void upb_bufhandle_setbuf(upb_bufhandle *h, const char *buf, + size_t ofs) { + h->buf_ = buf; + h->objofs_ = ofs; +} +UPB_INLINE const void *upb_bufhandle_obj(const upb_bufhandle *h) { + return h->obj_; +} +UPB_INLINE const void *upb_bufhandle_objtype(const upb_bufhandle *h) { + return h->objtype_; +} +UPB_INLINE const char *upb_bufhandle_buf(const upb_bufhandle *h) { + return h->buf_; +} + + +#ifdef __cplusplus + +namespace upb { + +typedef void CleanupFunc(void *ptr); + +// Template to remove "const" from "const T*" and just return "T*". +// +// We define a nonsense default because otherwise it will fail to instantiate as +// a function parameter type even in cases where we don't expect any caller to +// actually match the overload. +class CouldntRemoveConst {}; +template struct remove_constptr { typedef CouldntRemoveConst type; }; +template struct remove_constptr { typedef T *type; }; + +// Template that we use below to remove a template specialization from +// consideration if it matches a specific type. +template struct disable_if_same { typedef void Type; }; +template struct disable_if_same {}; + +template void DeletePointer(void *p) { delete static_cast(p); } + +template +struct FirstUnlessVoid { + typedef T1 value; +}; + +template +struct FirstUnlessVoid { + typedef T2 value; +}; + +template +struct is_same { + static bool value; +}; + +template +struct is_same { + static bool value; +}; + +template +bool is_same::value = false; + +template +bool is_same::value = true; + +// FuncInfo //////////////////////////////////////////////////////////////////// + +// Info about the user's original, pre-wrapped function. +template +struct FuncInfo { + // The type of the closure that the function takes (its first param). + typedef C Closure; + + // The return type. + typedef R Return; +}; + +// Func //////////////////////////////////////////////////////////////////////// + +// Func1, Func2, Func3: Template classes representing a function and its +// signature. +// +// Since the function is a template parameter, calling the function can be +// inlined at compile-time and does not require a function pointer at runtime. +// These functions are not bound to a handler data so have no data or cleanup +// handler. +struct UnboundFunc { + CleanupFunc *GetCleanup() { return NULL; } + void *GetData() { return NULL; } +}; + +template +struct Func1 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1) { return F(p1); } +}; + +template +struct Func2 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2) { return F(p1, p2); } +}; + +template +struct Func3 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2, P3 p3) { return F(p1, p2, p3); } +}; + +template +struct Func4 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2, P3 p3, P4 p4) { return F(p1, p2, p3, p4); } +}; + +template +struct Func5 : public UnboundFunc { + typedef R Return; + typedef I FuncInfo; + static R Call(P1 p1, P2 p2, P3 p3, P4 p4, P5 p5) { + return F(p1, p2, p3, p4, p5); + } +}; + +// BoundFunc /////////////////////////////////////////////////////////////////// + +// BoundFunc2, BoundFunc3: Like Func2/Func3 except also contains a value that +// shall be bound to the function's second parameter. +// +// Note that the second parameter is a const pointer, but our stored bound value +// is non-const so we can free it when the handlers are destroyed. +template +struct BoundFunc { + typedef typename remove_constptr::type MutableP2; + explicit BoundFunc(MutableP2 data_) : data(data_) {} + CleanupFunc *GetCleanup() { return &DeletePointer; } + MutableP2 GetData() { return data; } + MutableP2 data; +}; + +template +struct BoundFunc2 : public BoundFunc { + typedef BoundFunc Base; + typedef I FuncInfo; + explicit BoundFunc2(typename Base::MutableP2 arg) : Base(arg) {} +}; + +template +struct BoundFunc3 : public BoundFunc { + typedef BoundFunc Base; + typedef I FuncInfo; + explicit BoundFunc3(typename Base::MutableP2 arg) : Base(arg) {} +}; + +template +struct BoundFunc4 : public BoundFunc { + typedef BoundFunc Base; + typedef I FuncInfo; + explicit BoundFunc4(typename Base::MutableP2 arg) : Base(arg) {} +}; + +template +struct BoundFunc5 : public BoundFunc { + typedef BoundFunc Base; + typedef I FuncInfo; + explicit BoundFunc5(typename Base::MutableP2 arg) : Base(arg) {} +}; + +// FuncSig ///////////////////////////////////////////////////////////////////// + +// FuncSig1, FuncSig2, FuncSig3: template classes reflecting a function +// *signature*, but without a specific function attached. +// +// These classes contain member functions that can be invoked with a +// specific function to return a Func/BoundFunc class. +template +struct FuncSig1 { + template + Func1 > GetFunc() { + return Func1 >(); + } +}; + +template +struct FuncSig2 { + template + Func2 > GetFunc() { + return Func2 >(); + } + + template + BoundFunc2 > GetFunc( + typename remove_constptr::type param2) { + return BoundFunc2 >(param2); + } +}; + +template +struct FuncSig3 { + template + Func3 > GetFunc() { + return Func3 >(); + } + + template + BoundFunc3 > GetFunc( + typename remove_constptr::type param2) { + return BoundFunc3 >(param2); + } +}; + +template +struct FuncSig4 { + template + Func4 > GetFunc() { + return Func4 >(); + } + + template + BoundFunc4 > GetFunc( + typename remove_constptr::type param2) { + return BoundFunc4 >(param2); + } +}; + +template +struct FuncSig5 { + template + Func5 > GetFunc() { + return Func5 >(); + } + + template + BoundFunc5 > GetFunc( + typename remove_constptr::type param2) { + return BoundFunc5 >(param2); + } +}; + +// Overloaded template function that can construct the appropriate FuncSig* +// class given a function pointer by deducing the template parameters. +template +inline FuncSig1 MatchFunc(R (*f)(P1)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return FuncSig1(); +} + +template +inline FuncSig2 MatchFunc(R (*f)(P1, P2)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return FuncSig2(); +} + +template +inline FuncSig3 MatchFunc(R (*f)(P1, P2, P3)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return FuncSig3(); +} + +template +inline FuncSig4 MatchFunc(R (*f)(P1, P2, P3, P4)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return FuncSig4(); +} + +template +inline FuncSig5 MatchFunc(R (*f)(P1, P2, P3, P4, P5)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return FuncSig5(); +} + +// MethodSig /////////////////////////////////////////////////////////////////// + +// CallMethod*: a function template that calls a given method. +template +R CallMethod0(C *obj) { + return ((*obj).*F)(); +} + +template +R CallMethod1(C *obj, P1 arg1) { + return ((*obj).*F)(arg1); +} + +template +R CallMethod2(C *obj, P1 arg1, P2 arg2) { + return ((*obj).*F)(arg1, arg2); +} + +template +R CallMethod3(C *obj, P1 arg1, P2 arg2, P3 arg3) { + return ((*obj).*F)(arg1, arg2, arg3); +} + +template +R CallMethod4(C *obj, P1 arg1, P2 arg2, P3 arg3, P4 arg4) { + return ((*obj).*F)(arg1, arg2, arg3, arg4); +} + +// MethodSig: like FuncSig, but for member functions. +// +// GetFunc() returns a normal FuncN object, so after calling GetFunc() no +// more logic is required to special-case methods. +template +struct MethodSig0 { + template + Func1, FuncInfo > GetFunc() { + return Func1, FuncInfo >(); + } +}; + +template +struct MethodSig1 { + template + Func2, FuncInfo > GetFunc() { + return Func2, FuncInfo >(); + } + + template + BoundFunc2, FuncInfo > GetFunc( + typename remove_constptr::type param1) { + return BoundFunc2, FuncInfo >( + param1); + } +}; + +template +struct MethodSig2 { + template + Func3, FuncInfo > + GetFunc() { + return Func3, + FuncInfo >(); + } + + template + BoundFunc3, FuncInfo > + GetFunc(typename remove_constptr::type param1) { + return BoundFunc3, + FuncInfo >(param1); + } +}; + +template +struct MethodSig3 { + template + Func4, FuncInfo > + GetFunc() { + return Func4, + FuncInfo >(); + } + + template + BoundFunc4, + FuncInfo > + GetFunc(typename remove_constptr::type param1) { + return BoundFunc4, + FuncInfo >(param1); + } +}; + +template +struct MethodSig4 { + template + Func5, + FuncInfo > + GetFunc() { + return Func5, + FuncInfo >(); + } + + template + BoundFunc5, + FuncInfo > + GetFunc(typename remove_constptr::type param1) { + return BoundFunc5, FuncInfo >( + param1); + } +}; + +template +inline MethodSig0 MatchFunc(R (C::*f)()) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return MethodSig0(); +} + +template +inline MethodSig1 MatchFunc(R (C::*f)(P1)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return MethodSig1(); +} + +template +inline MethodSig2 MatchFunc(R (C::*f)(P1, P2)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return MethodSig2(); +} + +template +inline MethodSig3 MatchFunc(R (C::*f)(P1, P2, P3)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return MethodSig3(); +} + +template +inline MethodSig4 MatchFunc(R (C::*f)(P1, P2, P3, P4)) { + UPB_UNUSED(f); // Only used for template parameter deduction. + return MethodSig4(); +} + +// MaybeWrapReturn ///////////////////////////////////////////////////////////// + +// Template class that attempts to wrap the return value of the function so it +// matches the expected type. There are two main adjustments it may make: +// +// 1. If the function returns void, make it return the expected type and with +// a value that always indicates success. +// 2. If the function is expected to return void* but doesn't, wrap it so it +// does (either by returning the closure param if the wrapped function +// returns void or by casting a different pointer type to void* for +// return). + +// Template parameters are FuncN type and desired return type. +template +struct MaybeWrapReturn; + +// If the return type matches, return the given function unwrapped. +template +struct MaybeWrapReturn { + typedef F Func; +}; + +// Function wrapper that munges the return value from void to (bool)true. +template +bool ReturnTrue2(P1 p1, P2 p2) { + F(p1, p2); + return true; +} + +template +bool ReturnTrue3(P1 p1, P2 p2, P3 p3) { + F(p1, p2, p3); + return true; +} + +// Function wrapper that munges the return value from void to (void*)arg1 +template +void *ReturnClosure2(P1 p1, P2 p2) { + F(p1, p2); + return p1; +} + +template +void *ReturnClosure3(P1 p1, P2 p2, P3 p3) { + F(p1, p2, p3); + return p1; +} + +// Function wrapper that munges the return value from R to void*. +template +void *CastReturnToVoidPtr2(P1 p1, P2 p2) { + return F(p1, p2); +} + +template +void *CastReturnToVoidPtr3(P1 p1, P2 p2, P3 p3) { + return F(p1, p2, p3); +} + +// Function wrapper that munges the return value from bool to void*. +template +void *ReturnClosureOrBreak2(P1 p1, P2 p2) { + return F(p1, p2) ? p1 : UPB_BREAK; +} + +template +void *ReturnClosureOrBreak3(P1 p1, P2 p2, P3 p3) { + return F(p1, p2, p3) ? p1 : UPB_BREAK; +} + +// For the string callback, which takes five params, returns the size param. +template +size_t ReturnStringLen(P1 p1, P2 p2, const char *p3, size_t p4, + const BufferHandle *p5) { + F(p1, p2, p3, p4, p5); + return p4; +} + +// For the string callback, which takes five params, returns the size param or +// zero. +template +size_t ReturnNOr0(P1 p1, P2 p2, const char *p3, size_t p4, + const BufferHandle *p5) { + return F(p1, p2, p3, p4, p5) ? p4 : 0; +} + +// If we have a function returning void but want a function returning bool, wrap +// it in a function that returns true. +template +struct MaybeWrapReturn, bool> { + typedef Func2, I> Func; +}; + +template +struct MaybeWrapReturn, bool> { + typedef Func3, I> Func; +}; + +// If our function returns void but we want one returning void*, wrap it in a +// function that returns the first argument. +template +struct MaybeWrapReturn, void *> { + typedef Func2, I> Func; +}; + +template +struct MaybeWrapReturn, void *> { + typedef Func3, I> Func; +}; + +// If our function returns R* but we want one returning void*, wrap it in a +// function that casts to void*. +template +struct MaybeWrapReturn, void *, + typename disable_if_same::Type> { + typedef Func2, I> Func; +}; + +template +struct MaybeWrapReturn, void *, + typename disable_if_same::Type> { + typedef Func3, I> + Func; +}; + +// If our function returns bool but we want one returning void*, wrap it in a +// function that returns either the first param or UPB_BREAK. +template +struct MaybeWrapReturn, void *> { + typedef Func2, I> Func; +}; + +template +struct MaybeWrapReturn, void *> { + typedef Func3, I> + Func; +}; + +// If our function returns void but we want one returning size_t, wrap it in a +// function that returns the size argument. +template +struct MaybeWrapReturn< + Func5, + size_t> { + typedef Func5, I> Func; +}; + +// If our function returns bool but we want one returning size_t, wrap it in a +// function that returns either 0 or the buf size. +template +struct MaybeWrapReturn< + Func5, + size_t> { + typedef Func5, I> Func; +}; + +// ConvertParams /////////////////////////////////////////////////////////////// + +// Template class that converts the function parameters if necessary, and +// ignores the HandlerData parameter if appropriate. +// +// Template parameter is the are FuncN function type. +template +struct ConvertParams; + +// Function that discards the handler data parameter. +template +R IgnoreHandlerData2(void *p1, const void *hd) { + UPB_UNUSED(hd); + return F(static_cast(p1)); +} + +template +R IgnoreHandlerData3(void *p1, const void *hd, P2Wrapper p2) { + UPB_UNUSED(hd); + return F(static_cast(p1), p2); +} + +template +R IgnoreHandlerData4(void *p1, const void *hd, P2 p2, P3 p3) { + UPB_UNUSED(hd); + return F(static_cast(p1), p2, p3); +} + +template +R IgnoreHandlerData5(void *p1, const void *hd, P2 p2, P3 p3, P4 p4) { + UPB_UNUSED(hd); + return F(static_cast(p1), p2, p3, p4); +} + +template +R IgnoreHandlerDataIgnoreHandle(void *p1, const void *hd, const char *p2, + size_t p3, const BufferHandle *handle) { + UPB_UNUSED(hd); + UPB_UNUSED(handle); + return F(static_cast(p1), p2, p3); +} + +// Function that casts the handler data parameter. +template +R CastHandlerData2(void *c, const void *hd) { + return F(static_cast(c), static_cast(hd)); +} + +template +R CastHandlerData3(void *c, const void *hd, P3Wrapper p3) { + return F(static_cast(c), static_cast(hd), p3); +} + +template +R CastHandlerData5(void *c, const void *hd, P3 p3, P4 p4, P5 p5) { + return F(static_cast(c), static_cast(hd), p3, p4, p5); +} + +template +R CastHandlerDataIgnoreHandle(void *c, const void *hd, const char *p3, + size_t p4, const BufferHandle *handle) { + UPB_UNUSED(handle); + return F(static_cast(c), static_cast(hd), p3, p4); +} + +// For unbound functions, ignore the handler data. +template +struct ConvertParams, T> { + typedef Func2, I> Func; +}; + +template +struct ConvertParams, + R2 (*)(P1_2, P2_2, P3_2)> { + typedef Func3, I> Func; +}; + +// For StringBuffer only; this ignores both the handler data and the +// BufferHandle. +template +struct ConvertParams, T> { + typedef Func5, + I> Func; +}; + +template +struct ConvertParams, T> { + typedef Func5, I> Func; +}; + +// For bound functions, cast the handler data. +template +struct ConvertParams, T> { + typedef Func2, I> + Func; +}; + +template +struct ConvertParams, + R2 (*)(P1_2, P2_2, P3_2)> { + typedef Func3, I> Func; +}; + +// For StringBuffer only; this ignores the BufferHandle. +template +struct ConvertParams, T> { + typedef Func5, + I> Func; +}; + +template +struct ConvertParams, T> { + typedef Func5, I> Func; +}; + +// utype/ltype are upper/lower-case, ctype is canonical C type, vtype is +// variant C type. +#define TYPE_METHODS(utype, ltype, ctype, vtype) \ + template <> struct CanonicalType { \ + typedef ctype Type; \ + }; \ + template <> \ + inline bool Handlers::SetValueHandler( \ + const FieldDef *f, \ + const Handlers::utype ## Handler& handler) { \ + assert(!handler.registered_); \ + handler.AddCleanup(this); \ + handler.registered_ = true; \ + return upb_handlers_set##ltype(this, f, handler.handler_, &handler.attr_); \ + } \ + +TYPE_METHODS(Double, double, double, double); +TYPE_METHODS(Float, float, float, float); +TYPE_METHODS(UInt64, uint64, uint64_t, UPB_UINT64_T); +TYPE_METHODS(UInt32, uint32, uint32_t, UPB_UINT32_T); +TYPE_METHODS(Int64, int64, int64_t, UPB_INT64_T); +TYPE_METHODS(Int32, int32, int32_t, UPB_INT32_T); +TYPE_METHODS(Bool, bool, bool, bool); + +#ifdef UPB_TWO_32BIT_TYPES +TYPE_METHODS(Int32, int32, int32_t, UPB_INT32ALT_T); +TYPE_METHODS(UInt32, uint32, uint32_t, UPB_UINT32ALT_T); +#endif + +#ifdef UPB_TWO_64BIT_TYPES +TYPE_METHODS(Int64, int64, int64_t, UPB_INT64ALT_T); +TYPE_METHODS(UInt64, uint64, uint64_t, UPB_UINT64ALT_T); +#endif +#undef TYPE_METHODS + +template <> struct CanonicalType { + typedef Status* Type; +}; + +// Type methods that are only one-per-canonical-type and not one-per-cvariant. + +#define TYPE_METHODS(utype, ctype) \ + inline bool Handlers::Set##utype##Handler(const FieldDef *f, \ + const utype##Handler &h) { \ + return SetValueHandler(f, h); \ + } \ + +TYPE_METHODS(Double, double); +TYPE_METHODS(Float, float); +TYPE_METHODS(UInt64, uint64_t); +TYPE_METHODS(UInt32, uint32_t); +TYPE_METHODS(Int64, int64_t); +TYPE_METHODS(Int32, int32_t); +TYPE_METHODS(Bool, bool); +#undef TYPE_METHODS + +template struct ReturnOf; + +template +struct ReturnOf { + typedef R Return; +}; + +template +struct ReturnOf { + typedef R Return; +}; + +template +struct ReturnOf { + typedef R Return; +}; + +template +struct ReturnOf { + typedef R Return; +}; + +template const void *UniquePtrForType() { + static const char ch = 0; + return &ch; +} + +template +template +inline Handler::Handler(F func) + : registered_(false), + cleanup_data_(func.GetData()), + cleanup_func_(func.GetCleanup()) { + upb_handlerattr_sethandlerdata(&attr_, func.GetData()); + typedef typename ReturnOf::Return Return; + typedef typename ConvertParams::Func ConvertedParamsFunc; + typedef typename MaybeWrapReturn::Func + ReturnWrappedFunc; + handler_ = ReturnWrappedFunc().Call; + + // Set attributes based on what templates can statically tell us about the + // user's function. + + // If the original function returns void, then we know that we wrapped it to + // always return ok. + bool always_ok = is_same::value; + attr_.SetAlwaysOk(always_ok); + + // Closure parameter and return type. + attr_.SetClosureType(UniquePtrForType()); + + // We use the closure type (from the first parameter) if the return type is + // void. This is all nonsense for non START* handlers, but it doesn't matter + // because in that case the value will be ignored. + typedef typename FirstUnlessVoid::value + EffectiveReturn; + attr_.SetReturnClosureType(UniquePtrForType()); +} + +template +inline Handler::~Handler() { + assert(registered_); +} + +inline HandlerAttributes::HandlerAttributes() { upb_handlerattr_init(this); } +inline HandlerAttributes::~HandlerAttributes() { upb_handlerattr_uninit(this); } +inline bool HandlerAttributes::SetHandlerData(const void *hd) { + return upb_handlerattr_sethandlerdata(this, hd); +} +inline const void* HandlerAttributes::handler_data() const { + return upb_handlerattr_handlerdata(this); +} +inline bool HandlerAttributes::SetClosureType(const void *type) { + return upb_handlerattr_setclosuretype(this, type); +} +inline const void* HandlerAttributes::closure_type() const { + return upb_handlerattr_closuretype(this); +} +inline bool HandlerAttributes::SetReturnClosureType(const void *type) { + return upb_handlerattr_setreturnclosuretype(this, type); +} +inline const void* HandlerAttributes::return_closure_type() const { + return upb_handlerattr_returnclosuretype(this); +} +inline bool HandlerAttributes::SetAlwaysOk(bool always_ok) { + return upb_handlerattr_setalwaysok(this, always_ok); +} +inline bool HandlerAttributes::always_ok() const { + return upb_handlerattr_alwaysok(this); +} + +inline BufferHandle::BufferHandle() { upb_bufhandle_init(this); } +inline BufferHandle::~BufferHandle() { upb_bufhandle_uninit(this); } +inline const char* BufferHandle::buffer() const { + return upb_bufhandle_buf(this); +} +inline size_t BufferHandle::object_offset() const { + return upb_bufhandle_objofs(this); +} +inline void BufferHandle::SetBuffer(const char* buf, size_t ofs) { + upb_bufhandle_setbuf(this, buf, ofs); +} +template +void BufferHandle::SetAttachedObject(const T* obj) { + upb_bufhandle_setobj(this, obj, UniquePtrForType()); +} +template +const T* BufferHandle::GetAttachedObject() const { + return upb_bufhandle_objtype(this) == UniquePtrForType() + ? static_cast(upb_bufhandle_obj(this)) + : NULL; +} + +inline reffed_ptr Handlers::New(const MessageDef *m) { + upb_handlers *h = upb_handlers_new(m, &h); + return reffed_ptr(h, &h); +} +inline reffed_ptr Handlers::NewFrozen( + const MessageDef *m, upb_handlers_callback *callback, + const void *closure) { + const upb_handlers *h = upb_handlers_newfrozen(m, &h, callback, closure); + return reffed_ptr(h, &h); +} +inline bool Handlers::IsFrozen() const { return upb_handlers_isfrozen(this); } +inline void Handlers::Ref(const void *owner) const { + upb_handlers_ref(this, owner); +} +inline void Handlers::Unref(const void *owner) const { + upb_handlers_unref(this, owner); +} +inline void Handlers::DonateRef(const void *from, const void *to) const { + upb_handlers_donateref(this, from, to); +} +inline void Handlers::CheckRef(const void *owner) const { + upb_handlers_checkref(this, owner); +} +inline const Status* Handlers::status() { + return upb_handlers_status(this); +} +inline void Handlers::ClearError() { + return upb_handlers_clearerr(this); +} +inline bool Handlers::Freeze(Status *s) { + upb::Handlers* h = this; + return upb_handlers_freeze(&h, 1, s); +} +inline bool Handlers::Freeze(Handlers *const *handlers, int n, Status *s) { + return upb_handlers_freeze(handlers, n, s); +} +inline bool Handlers::Freeze(const std::vector& h, Status* status) { + return upb_handlers_freeze((Handlers* const*)&h[0], h.size(), status); +} +inline const MessageDef *Handlers::message_def() const { + return upb_handlers_msgdef(this); +} +inline bool Handlers::AddCleanup(void *p, upb_handlerfree *func) { + return upb_handlers_addcleanup(this, p, func); +} +inline bool Handlers::SetStartMessageHandler( + const Handlers::StartMessageHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartmsg(this, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndMessageHandler( + const Handlers::EndMessageHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendmsg(this, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStartStringHandler(const FieldDef *f, + const StartStringHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartstr(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndStringHandler(const FieldDef *f, + const EndFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendstr(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStringHandler(const FieldDef *f, + const StringHandler& handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstring(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStartSequenceHandler( + const FieldDef *f, const StartFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartseq(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetStartSubMessageHandler( + const FieldDef *f, const StartFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setstartsubmsg(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndSubMessageHandler(const FieldDef *f, + const EndFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendsubmsg(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetEndSequenceHandler(const FieldDef *f, + const EndFieldHandler &handler) { + assert(!handler.registered_); + handler.registered_ = true; + handler.AddCleanup(this); + return upb_handlers_setendseq(this, f, handler.handler_, &handler.attr_); +} +inline bool Handlers::SetSubHandlers(const FieldDef *f, const Handlers *sub) { + return upb_handlers_setsubhandlers(this, f, sub); +} +inline const Handlers *Handlers::GetSubHandlers(const FieldDef *f) const { + return upb_handlers_getsubhandlers(this, f); +} +inline const Handlers *Handlers::GetSubHandlers(Handlers::Selector sel) const { + return upb_handlers_getsubhandlers_sel(this, sel); +} +inline bool Handlers::GetSelector(const FieldDef *f, Handlers::Type type, + Handlers::Selector *s) { + return upb_handlers_getselector(f, type, s); +} +inline Handlers::Selector Handlers::GetEndSelector(Handlers::Selector start) { + return upb_handlers_getendselector(start); +} +inline Handlers::GenericFunction *Handlers::GetHandler( + Handlers::Selector selector) { + return upb_handlers_gethandler(this, selector); +} +inline const void *Handlers::GetHandlerData(Handlers::Selector selector) { + return upb_handlers_gethandlerdata(this, selector); +} + +inline BytesHandler::BytesHandler() { + upb_byteshandler_init(this); +} + +inline BytesHandler::~BytesHandler() { + upb_byteshandler_uninit(this); +} + +} // namespace upb + +#endif // __cplusplus + + +#undef UPB_TWO_32BIT_TYPES +#undef UPB_TWO_64BIT_TYPES +#undef UPB_INT32_T +#undef UPB_UINT32_T +#undef UPB_INT32ALT_T +#undef UPB_UINT32ALT_T +#undef UPB_INT64_T +#undef UPB_UINT64_T +#undef UPB_INT64ALT_T +#undef UPB_UINT64ALT_T + +#endif // UPB_HANDLERS_INL_H_ + +#endif // UPB_HANDLERS_H +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2010-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * A upb_sink is an object that binds a upb_handlers object to some runtime + * state. It is the object that can actually receive data via the upb_handlers + * interface. + * + * Unlike upb_def and upb_handlers, upb_sink is never frozen, immutable, or + * thread-safe. You can create as many of them as you want, but each one may + * only be used in a single thread at a time. + * + * If we compare with class-based OOP, a you can think of a upb_def as an + * abstract base class, a upb_handlers as a concrete derived class, and a + * upb_sink as an object (class instance). + */ + +#ifndef UPB_SINK_H +#define UPB_SINK_H + + +#ifdef __cplusplus +namespace upb { +class BufferSource; +class BytesSink; +class Sink; +} +#endif + +UPB_DECLARE_TYPE(upb::BufferSource, upb_bufsrc); +UPB_DECLARE_TYPE(upb::BytesSink, upb_bytessink); +UPB_DECLARE_TYPE(upb::Sink, upb_sink); + +// Internal-only struct for the sink. +struct upb_sinkframe { + UPB_PRIVATE_FOR_CPP + const upb_handlers *h; + void *closure; + + // For any frames besides the top, this is the END* callback that will run + // when the subframe is popped (for example, for a "sequence" frame the frame + // above it will be a UPB_HANDLER_ENDSEQ handler). But this is only + // necessary for assertion checking inside upb_sink and can be omitted if the + // sink has only one caller. + // + // TODO(haberman): have a mechanism for ensuring that a sink only has one + // caller. + upb_selector_t selector; +}; + +// The maximum nesting depth that upb::Sink will allow. Matches proto2's limit. +// TODO: make this a runtime-settable property of Sink. +#define UPB_SINK_MAX_NESTING 64 + +// A upb::Sink is an object that binds a upb::Handlers object to some runtime +// state. It represents an endpoint to which data can be sent. +// +// TODO(haberman): right now all of these functions take selectors. Should they +// take selectorbase instead? +// +// ie. instead of calling: +// sink->StartString(FOO_FIELD_START_STRING, ...) +// a selector base would let you say: +// sink->StartString(FOO_FIELD, ...) +// +// This would make call sites a little nicer and require emitting fewer selector +// definitions in .h files. +// +// But the current scheme has the benefit that you can retrieve a function +// pointer for any handler with handlers->GetHandler(selector), without having +// to have a separate GetHandler() function for each handler type. The JIT +// compiler uses this. To accommodate we'd have to expose a separate +// GetHandler() for every handler type. +// +// Also to ponder: selectors right now are independent of a specific Handlers +// instance. In other words, they allocate a number to every possible handler +// that *could* be registered, without knowing anything about what handlers +// *are* registered. That means that using selectors as table offsets prohibits +// us from compacting the handler table at Freeze() time. If the table is very +// sparse, this could be wasteful. +// +// Having another selector-like thing that is specific to a Handlers instance +// would allow this compacting, but then it would be impossible to write code +// ahead-of-time that can be bound to any Handlers instance at runtime. For +// example, a .proto file parser written as straight C will not know what +// Handlers it will be bound to, so when it calls sink->StartString() what +// selector will it pass? It needs a selector like we have today, that is +// independent of any particular upb::Handlers. +// +// Is there a way then to allow Handlers table compaction? +UPB_DEFINE_CLASS0(upb::Sink, + public: + // Constructor with no initialization; must be Reset() before use. + Sink() {} + + // Constructs a new sink for the given frozen handlers and closure. + // + // TODO: once the Handlers know the expected closure type, verify that T + // matches it. + template Sink(const Handlers* handlers, T* closure); + + // Resets the value of the sink. + template void Reset(const Handlers* handlers, T* closure); + + // Returns the top-level object that is bound to this sink. + // + // TODO: once the Handlers know the expected closure type, verify that T + // matches it. + template T* GetObject() const; + + // Functions for pushing data into the sink. + // + // These return false if processing should stop (either due to error or just + // to suspend). + // + // These may not be called from within one of the same sink's handlers (in + // other words, handlers are not re-entrant). + + // Should be called at the start and end of every message; both the top-level + // message and submessages. This means that submessages should use the + // following sequence: + // sink->StartSubMessage(startsubmsg_selector); + // sink->StartMessage(); + // // ... + // sink->EndMessage(&status); + // sink->EndSubMessage(endsubmsg_selector); + bool StartMessage(); + bool EndMessage(Status* status); + + // Putting of individual values. These work for both repeated and + // non-repeated fields, but for repeated fields you must wrap them in + // calls to StartSequence()/EndSequence(). + bool PutInt32(Handlers::Selector s, int32_t val); + bool PutInt64(Handlers::Selector s, int64_t val); + bool PutUInt32(Handlers::Selector s, uint32_t val); + bool PutUInt64(Handlers::Selector s, uint64_t val); + bool PutFloat(Handlers::Selector s, float val); + bool PutDouble(Handlers::Selector s, double val); + bool PutBool(Handlers::Selector s, bool val); + + // Putting of string/bytes values. Each string can consist of zero or more + // non-contiguous buffers of data. + // + // For StartString(), the function will write a sink for the string to "sub." + // The sub-sink must be used for any/all PutStringBuffer() calls. + bool StartString(Handlers::Selector s, size_t size_hint, Sink* sub); + size_t PutStringBuffer(Handlers::Selector s, const char *buf, size_t len, + const BufferHandle *handle); + bool EndString(Handlers::Selector s); + + // For submessage fields. + // + // For StartSubMessage(), the function will write a sink for the string to + // "sub." The sub-sink must be used for any/all handlers called within the + // submessage. + bool StartSubMessage(Handlers::Selector s, Sink* sub); + bool EndSubMessage(Handlers::Selector s); + + // For repeated fields of any type, the sequence of values must be wrapped in + // these calls. + // + // For StartSequence(), the function will write a sink for the string to + // "sub." The sub-sink must be used for any/all handlers called within the + // sequence. + bool StartSequence(Handlers::Selector s, Sink* sub); + bool EndSequence(Handlers::Selector s); + + // Copy and assign specifically allowed. + // We don't even bother making these members private because so many + // functions need them and this is mainly just a dumb data container anyway. +, +UPB_DEFINE_STRUCT0(upb_sink, + const upb_handlers *handlers; + void *closure; +)); + +UPB_DEFINE_CLASS0(upb::BytesSink, + public: + BytesSink() {} + + // Constructs a new sink for the given frozen handlers and closure. + // + // TODO(haberman): once the Handlers know the expected closure type, verify + // that T matches it. + template BytesSink(const BytesHandler* handler, T* closure); + + // Resets the value of the sink. + template void Reset(const BytesHandler* handler, T* closure); + + bool Start(size_t size_hint, void **subc); + size_t PutBuffer(void *subc, const char *buf, size_t len, + const BufferHandle *handle); + bool End(); +, +UPB_DEFINE_STRUCT0(upb_bytessink, + const upb_byteshandler *handler; + void *closure; +)); + +// A class for pushing a flat buffer of data to a BytesSink. +// You can construct an instance of this to get a resumable source, +// or just call the static PutBuffer() to do a non-resumable push all in one go. +UPB_DEFINE_CLASS0(upb::BufferSource, + public: + BufferSource(); + BufferSource(const char* buf, size_t len, BytesSink* sink); + + // Returns true if the entire buffer was pushed successfully. Otherwise the + // next call to PutNext() will resume where the previous one left off. + // TODO(haberman): implement this. + bool PutNext(); + + // A static version; with this version is it not possible to resume in the + // case of failure or a partially-consumed buffer. + static bool PutBuffer(const char* buf, size_t len, BytesSink* sink); + + template static bool PutBuffer(const T& str, BytesSink* sink) { + return PutBuffer(str.c_str(), str.size(), sink); + } +, +UPB_DEFINE_STRUCT0(upb_bufsrc, +)); + +UPB_BEGIN_EXTERN_C // { + +// Inline definitions. + +UPB_INLINE void upb_bytessink_reset(upb_bytessink *s, const upb_byteshandler *h, + void *closure) { + s->handler = h; + s->closure = closure; +} + +UPB_INLINE bool upb_bytessink_start(upb_bytessink *s, size_t size_hint, + void **subc) { + *subc = s->closure; + if (!s->handler) return true; + upb_startstr_handlerfunc *start = + (upb_startstr_handlerfunc *)s->handler->table[UPB_STARTSTR_SELECTOR].func; + + if (!start) return true; + *subc = start(s->closure, upb_handlerattr_handlerdata( + &s->handler->table[UPB_STARTSTR_SELECTOR].attr), + size_hint); + return *subc != NULL; +} + +UPB_INLINE size_t upb_bytessink_putbuf(upb_bytessink *s, void *subc, + const char *buf, size_t size, + const upb_bufhandle* handle) { + if (!s->handler) return true; + upb_string_handlerfunc *putbuf = + (upb_string_handlerfunc *)s->handler->table[UPB_STRING_SELECTOR].func; + + if (!putbuf) return true; + return putbuf(subc, upb_handlerattr_handlerdata( + &s->handler->table[UPB_STRING_SELECTOR].attr), + buf, size, handle); +} + +UPB_INLINE bool upb_bytessink_end(upb_bytessink *s) { + if (!s->handler) return true; + upb_endfield_handlerfunc *end = + (upb_endfield_handlerfunc *)s->handler->table[UPB_ENDSTR_SELECTOR].func; + + if (!end) return true; + return end(s->closure, + upb_handlerattr_handlerdata( + &s->handler->table[UPB_ENDSTR_SELECTOR].attr)); +} + +UPB_INLINE bool upb_bufsrc_putbuf(const char *buf, size_t len, + upb_bytessink *sink) { + void *subc; + upb_bufhandle handle; + upb_bufhandle_init(&handle); + upb_bufhandle_setbuf(&handle, buf, 0); + bool ret = upb_bytessink_start(sink, len, &subc); + if (ret && len != 0) { + ret = (upb_bytessink_putbuf(sink, subc, buf, len, &handle) == len); + } + if (ret) { + ret = upb_bytessink_end(sink); + } + upb_bufhandle_uninit(&handle); + return ret; +} + +#define PUTVAL(type, ctype) \ + UPB_INLINE bool upb_sink_put##type(upb_sink *s, upb_selector_t sel, \ + ctype val) { \ + if (!s->handlers) return true; \ + upb_##type##_handlerfunc *func = \ + (upb_##type##_handlerfunc *)upb_handlers_gethandler(s->handlers, sel); \ + if (!func) return true; \ + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); \ + return func(s->closure, hd, val); \ + } + +PUTVAL(int32, int32_t); +PUTVAL(int64, int64_t); +PUTVAL(uint32, uint32_t); +PUTVAL(uint64, uint64_t); +PUTVAL(float, float); +PUTVAL(double, double); +PUTVAL(bool, bool); +#undef PUTVAL + +UPB_INLINE void upb_sink_reset(upb_sink *s, const upb_handlers *h, void *c) { + s->handlers = h; + s->closure = c; +} + +UPB_INLINE size_t upb_sink_putstring(upb_sink *s, upb_selector_t sel, + const char *buf, size_t n, + const upb_bufhandle *handle) { + if (!s->handlers) return n; + upb_string_handlerfunc *handler = + (upb_string_handlerfunc *)upb_handlers_gethandler(s->handlers, sel); + + if (!handler) return n; + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); + return handler(s->closure, hd, buf, n, handle); +} + +UPB_INLINE bool upb_sink_startmsg(upb_sink *s) { + if (!s->handlers) return true; + upb_startmsg_handlerfunc *startmsg = + (upb_startmsg_handlerfunc *)upb_handlers_gethandler(s->handlers, + UPB_STARTMSG_SELECTOR); + if (!startmsg) return true; + const void *hd = + upb_handlers_gethandlerdata(s->handlers, UPB_STARTMSG_SELECTOR); + return startmsg(s->closure, hd); +} + +UPB_INLINE bool upb_sink_endmsg(upb_sink *s, upb_status *status) { + if (!s->handlers) return true; + upb_endmsg_handlerfunc *endmsg = + (upb_endmsg_handlerfunc *)upb_handlers_gethandler(s->handlers, + UPB_ENDMSG_SELECTOR); + + if (!endmsg) return true; + const void *hd = + upb_handlers_gethandlerdata(s->handlers, UPB_ENDMSG_SELECTOR); + return endmsg(s->closure, hd, status); +} + +UPB_INLINE bool upb_sink_startseq(upb_sink *s, upb_selector_t sel, + upb_sink *sub) { + sub->closure = s->closure; + sub->handlers = s->handlers; + if (!s->handlers) return true; + upb_startfield_handlerfunc *startseq = + (upb_startfield_handlerfunc*)upb_handlers_gethandler(s->handlers, sel); + + if (!startseq) return true; + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); + sub->closure = startseq(s->closure, hd); + return sub->closure ? true : false; +} + +UPB_INLINE bool upb_sink_endseq(upb_sink *s, upb_selector_t sel) { + if (!s->handlers) return true; + upb_endfield_handlerfunc *endseq = + (upb_endfield_handlerfunc*)upb_handlers_gethandler(s->handlers, sel); + + if (!endseq) return true; + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); + return endseq(s->closure, hd); +} + +UPB_INLINE bool upb_sink_startstr(upb_sink *s, upb_selector_t sel, + size_t size_hint, upb_sink *sub) { + sub->closure = s->closure; + sub->handlers = s->handlers; + if (!s->handlers) return true; + upb_startstr_handlerfunc *startstr = + (upb_startstr_handlerfunc*)upb_handlers_gethandler(s->handlers, sel); + + if (!startstr) return true; + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); + sub->closure = startstr(s->closure, hd, size_hint); + return sub->closure ? true : false; +} + +UPB_INLINE bool upb_sink_endstr(upb_sink *s, upb_selector_t sel) { + if (!s->handlers) return true; + upb_endfield_handlerfunc *endstr = + (upb_endfield_handlerfunc*)upb_handlers_gethandler(s->handlers, sel); + + if (!endstr) return true; + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); + return endstr(s->closure, hd); +} + +UPB_INLINE bool upb_sink_startsubmsg(upb_sink *s, upb_selector_t sel, + upb_sink *sub) { + sub->closure = s->closure; + if (!s->handlers) { + sub->handlers = NULL; + return true; + } + sub->handlers = upb_handlers_getsubhandlers_sel(s->handlers, sel); + upb_startfield_handlerfunc *startsubmsg = + (upb_startfield_handlerfunc*)upb_handlers_gethandler(s->handlers, sel); + + if (!startsubmsg) return true; + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); + sub->closure = startsubmsg(s->closure, hd); + return sub->closure ? true : false; +} + +UPB_INLINE bool upb_sink_endsubmsg(upb_sink *s, upb_selector_t sel) { + if (!s->handlers) return true; + upb_endfield_handlerfunc *endsubmsg = + (upb_endfield_handlerfunc*)upb_handlers_gethandler(s->handlers, sel); + + if (!endsubmsg) return s->closure; + const void *hd = upb_handlers_gethandlerdata(s->handlers, sel); + return endsubmsg(s->closure, hd); +} + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus + +namespace upb { + +template Sink::Sink(const Handlers* handlers, T* closure) { + upb_sink_reset(this, handlers, closure); +} +template +inline void Sink::Reset(const Handlers* handlers, T* closure) { + upb_sink_reset(this, handlers, closure); +} +inline bool Sink::StartMessage() { + return upb_sink_startmsg(this); +} +inline bool Sink::EndMessage(Status* status) { + return upb_sink_endmsg(this, status); +} +inline bool Sink::PutInt32(Handlers::Selector sel, int32_t val) { + return upb_sink_putint32(this, sel, val); +} +inline bool Sink::PutInt64(Handlers::Selector sel, int64_t val) { + return upb_sink_putint64(this, sel, val); +} +inline bool Sink::PutUInt32(Handlers::Selector sel, uint32_t val) { + return upb_sink_putuint32(this, sel, val); +} +inline bool Sink::PutUInt64(Handlers::Selector sel, uint64_t val) { + return upb_sink_putuint64(this, sel, val); +} +inline bool Sink::PutFloat(Handlers::Selector sel, float val) { + return upb_sink_putfloat(this, sel, val); +} +inline bool Sink::PutDouble(Handlers::Selector sel, double val) { + return upb_sink_putdouble(this, sel, val); +} +inline bool Sink::PutBool(Handlers::Selector sel, bool val) { + return upb_sink_putbool(this, sel, val); +} +inline bool Sink::StartString(Handlers::Selector sel, size_t size_hint, + Sink *sub) { + return upb_sink_startstr(this, sel, size_hint, sub); +} +inline size_t Sink::PutStringBuffer(Handlers::Selector sel, const char *buf, + size_t len, const BufferHandle* handle) { + return upb_sink_putstring(this, sel, buf, len, handle); +} +inline bool Sink::EndString(Handlers::Selector sel) { + return upb_sink_endstr(this, sel); +} +inline bool Sink::StartSubMessage(Handlers::Selector sel, Sink* sub) { + return upb_sink_startsubmsg(this, sel, sub); +} +inline bool Sink::EndSubMessage(Handlers::Selector sel) { + return upb_sink_endsubmsg(this, sel); +} +inline bool Sink::StartSequence(Handlers::Selector sel, Sink* sub) { + return upb_sink_startseq(this, sel, sub); +} +inline bool Sink::EndSequence(Handlers::Selector sel) { + return upb_sink_endseq(this, sel); +} + +template +BytesSink::BytesSink(const BytesHandler* handler, T* closure) { + Reset(handler, closure); +} + +template +void BytesSink::Reset(const BytesHandler *handler, T *closure) { + upb_bytessink_reset(this, handler, closure); +} +inline bool BytesSink::Start(size_t size_hint, void **subc) { + return upb_bytessink_start(this, size_hint, subc); +} +inline size_t BytesSink::PutBuffer(void *subc, const char *buf, size_t len, + const BufferHandle *handle) { + return upb_bytessink_putbuf(this, subc, buf, len, handle); +} +inline bool BytesSink::End() { + return upb_bytessink_end(this); +} + +inline bool BufferSource::PutBuffer(const char *buf, size_t len, + BytesSink *sink) { + return upb_bufsrc_putbuf(buf, len, sink); +} + +} // namespace upb +#endif + +#endif +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2013 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * For handlers that do very tiny, very simple operations, the function call + * overhead of calling a handler can be significant. This file allows the + * user to define handlers that do something very simple like store the value + * to memory and/or set a hasbit. JIT compilers can then special-case these + * handlers and emit specialized code for them instead of actually calling the + * handler. + * + * The functionality is very simple/limited right now but may expand to be able + * to call another function. + */ + +#ifndef UPB_SHIM_H +#define UPB_SHIM_H + + +typedef struct { + size_t offset; + int32_t hasbit; +} upb_shim_data; + +#ifdef __cplusplus + +namespace upb { + +struct Shim { + typedef upb_shim_data Data; + + // Sets a handler for the given field that writes the value to the given + // offset and, if hasbit >= 0, sets a bit at the given bit offset. Returns + // true if the handler was set successfully. + static bool Set(Handlers *h, const FieldDef *f, size_t ofs, int32_t hasbit); + + // If this handler is a shim, returns the corresponding upb::Shim::Data and + // stores the type in "type". Otherwise returns NULL. + static const Data* GetData(const Handlers* h, Handlers::Selector s, + FieldDef::Type* type); +}; + +} // namespace upb + +#endif + +UPB_BEGIN_EXTERN_C // { + +// C API. +bool upb_shim_set(upb_handlers *h, const upb_fielddef *f, size_t offset, + int32_t hasbit); +const upb_shim_data *upb_shim_getdata(const upb_handlers *h, upb_selector_t s, + upb_fieldtype_t *type); + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus +// C++ Wrappers. +namespace upb { +inline bool Shim::Set(Handlers* h, const FieldDef* f, size_t ofs, + int32_t hasbit) { + return upb_shim_set(h, f, ofs, hasbit); +} +inline const Shim::Data* Shim::GetData(const Handlers* h, Handlers::Selector s, + FieldDef::Type* type) { + return upb_shim_getdata(h, s, type); +} +} // namespace upb +#endif + +#endif // UPB_SHIM_H +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2011 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * upb::descriptor::Reader provides a way of building upb::Defs from + * data in descriptor.proto format. + */ + +#ifndef UPB_DESCRIPTOR_H +#define UPB_DESCRIPTOR_H + + +#ifdef __cplusplus +namespace upb { +namespace descriptor { +class Reader; +} // namespace descriptor +} // namespace upb +#endif + +UPB_DECLARE_TYPE(upb::descriptor::Reader, upb_descreader); + +// Internal-only structs used by Reader. + +// upb_deflist is an internal-only dynamic array for storing a growing list of +// upb_defs. +typedef struct { + UPB_PRIVATE_FOR_CPP + upb_def **defs; + size_t len; + size_t size; + bool owned; +} upb_deflist; + +// We keep a stack of all the messages scopes we are currently in, as well as +// the top-level file scope. This is necessary to correctly qualify the +// definitions that are contained inside. "name" tracks the name of the +// message or package (a bare name -- not qualified by any enclosing scopes). +typedef struct { + UPB_PRIVATE_FOR_CPP + char *name; + // Index of the first def that is under this scope. For msgdefs, the + // msgdef itself is at start-1. + int start; +} upb_descreader_frame; + +// The maximum number of nested declarations that are allowed, ie. +// message Foo { +// message Bar { +// message Baz { +// } +// } +// } +// +// This is a resource limit that affects how big our runtime stack can grow. +// TODO: make this a runtime-settable property of the Reader instance. +#define UPB_MAX_MESSAGE_NESTING 64 + +// Class that receives descriptor data according to the descriptor.proto schema +// and use it to build upb::Defs corresponding to that schema. +UPB_DEFINE_CLASS0(upb::descriptor::Reader, + public: + // These handlers must have come from NewHandlers() and must outlive the + // Reader. + // + // TODO: generate the handlers statically (like we do with the + // descriptor.proto defs) so that there is no need to pass this parameter (or + // to build/memory-manage the handlers at runtime at all). Unfortunately this + // is a bit tricky to implement for Handlers, but necessary to simplify this + // interface. + Reader(const Handlers* handlers, Status* status); + ~Reader(); + + // Resets the reader's state and discards any defs it may have built. + void Reset(); + + // The reader's input; this is where descriptor.proto data should be sent. + Sink* input(); + + // Returns an array of all defs that have been parsed, and transfers ownership + // of them to "owner". The number of defs is stored in *n. Ownership of the + // returned array is retained and is invalidated by any other call into + // Reader. + // + // These defs are not frozen or resolved; they are ready to be added to a + // symtab. + upb::Def** GetDefs(void* owner, int* n); + + // Builds and returns handlers for the reader, owned by "owner." + static Handlers* NewHandlers(const void* owner); +, +UPB_DEFINE_STRUCT0(upb_descreader, + upb_sink sink; + upb_deflist defs; + upb_descreader_frame stack[UPB_MAX_MESSAGE_NESTING]; + int stack_len; + + uint32_t number; + char *name; + bool saw_number; + bool saw_name; + + char *default_string; + + upb_fielddef *f; +)); + +UPB_BEGIN_EXTERN_C // { + +// C API. +void upb_descreader_init(upb_descreader *r, const upb_handlers *handlers, + upb_status *status); +void upb_descreader_uninit(upb_descreader *r); +void upb_descreader_reset(upb_descreader *r); +upb_sink *upb_descreader_input(upb_descreader *r); +upb_def **upb_descreader_getdefs(upb_descreader *r, void *owner, int *n); +const upb_handlers *upb_descreader_newhandlers(const void *owner); + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus +// C++ implementation details. ///////////////////////////////////////////////// +namespace upb { +namespace descriptor { +inline Reader::Reader(const Handlers *h, Status *s) { + upb_descreader_init(this, h, s); +} +inline Reader::~Reader() { upb_descreader_uninit(this); } +inline void Reader::Reset() { upb_descreader_reset(this); } +inline Sink* Reader::input() { return upb_descreader_input(this); } +inline upb::Def** Reader::GetDefs(void* owner, int* n) { + return upb_descreader_getdefs(this, owner, n); +} +} // namespace descriptor +} // namespace upb +#endif + +#endif // UPB_DESCRIPTOR_H +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009-2014 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Internal-only definitions for the decoder. + */ + +#ifndef UPB_DECODER_INT_H_ +#define UPB_DECODER_INT_H_ + +#include +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009-2014 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * upb::pb::Decoder implements a high performance, streaming, resumable decoder + * for the binary protobuf format. + * + * This interface works the same regardless of what decoder backend is being + * used. A client of this class does not need to know whether decoding is using + * a JITted decoder (DynASM, LLVM, etc) or an interpreted decoder. By default, + * it will always use the fastest available decoder. However, you can call + * set_allow_jit(false) to disable any JIT decoder that might be available. + * This is primarily useful for testing purposes. + */ + +#ifndef UPB_DECODER_H_ +#define UPB_DECODER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace pb { +class CodeCache; +class Decoder; +class DecoderMethod; +class DecoderMethodOptions; +} // namespace pb +} // namespace upb +#endif + +UPB_DECLARE_TYPE(upb::pb::CodeCache, upb_pbcodecache); +UPB_DECLARE_TYPE(upb::pb::Decoder, upb_pbdecoder); +UPB_DECLARE_TYPE(upb::pb::DecoderMethod, upb_pbdecodermethod); +UPB_DECLARE_TYPE(upb::pb::DecoderMethodOptions, upb_pbdecodermethodopts); + +// The maximum that any submessages can be nested. Matches proto2's limit. +// This specifies the size of the decoder's statically-sized array and therefore +// setting it high will cause the upb::pb::Decoder object to be larger. +// +// If necessary we can add a runtime-settable property to Decoder that allow +// this to be larger than the compile-time setting, but this would add +// complexity, particularly since we would have to decide how/if to give users +// the ability to set a custom memory allocation function. +#define UPB_DECODER_MAX_NESTING 64 + +// Internal-only struct used by the decoder. +typedef struct { + UPB_PRIVATE_FOR_CPP + // Space optimization note: we store two pointers here that the JIT + // doesn't need at all; the upb_handlers* inside the sink and + // the dispatch table pointer. We can optimze so that the JIT uses + // smaller stack frames than the interpreter. The only thing we need + // to guarantee is that the fallback routines can find end_ofs. + upb_sink sink; + + // The absolute stream offset of the end-of-frame delimiter. + // Non-delimited frames (groups and non-packed repeated fields) reuse the + // delimiter of their parent, even though the frame may not end there. + // + // NOTE: the JIT stores a slightly different value here for non-top frames. + // It stores the value relative to the end of the enclosed message. But the + // top frame is still stored the same way, which is important for ensuring + // that calls from the JIT into C work correctly. + uint64_t end_ofs; + const uint32_t *base; + + // 0 indicates a length-delimited field. + // A positive number indicates a known group. + // A negative number indicates an unknown group. + int32_t groupnum; + upb_inttable *dispatch; // Not used by the JIT. +} upb_pbdecoder_frame; + +// The parameters one uses to construct a DecoderMethod. +// TODO(haberman): move allowjit here? Seems more convenient for users. +UPB_DEFINE_CLASS0(upb::pb::DecoderMethodOptions, + public: + // Parameter represents the destination handlers that this method will push + // to. + explicit DecoderMethodOptions(const Handlers* dest_handlers); + + // Should the decoder push submessages to lazy handlers for fields that have + // them? The caller should set this iff the lazy handlers expect data that is + // in protobuf binary format and the caller wishes to lazy parse it. + void set_lazy(bool lazy); +, +UPB_DEFINE_STRUCT0(upb_pbdecodermethodopts, + const upb_handlers *handlers; + bool lazy; +)); + +// Represents the code to parse a protobuf according to a destination Handlers. +UPB_DEFINE_CLASS1(upb::pb::DecoderMethod, upb::RefCounted, + public: + // From upb::ReferenceCounted. + void Ref(const void* owner) const; + void Unref(const void* owner) const; + void DonateRef(const void* from, const void* to) const; + void CheckRef(const void* owner) const; + + // The destination handlers that are statically bound to this method. + // This method is only capable of outputting to a sink that uses these + // handlers. + const Handlers* dest_handlers() const; + + // The input handlers for this decoder method. + const BytesHandler* input_handler() const; + + // Whether this method is native. + bool is_native() const; + + // Convenience method for generating a DecoderMethod without explicitly + // creating a CodeCache. + static reffed_ptr New(const DecoderMethodOptions& opts); + + private: + UPB_DISALLOW_POD_OPS(DecoderMethod, upb::pb::DecoderMethod); +, +UPB_DEFINE_STRUCT(upb_pbdecodermethod, upb_refcounted, + // While compiling, the base is relative in "ofs", after compiling it is + // absolute in "ptr". + union { + uint32_t ofs; // PC offset of method. + void *ptr; // Pointer to bytecode or machine code for this method. + } code_base; + + // The decoder method group to which this method belongs. We own a ref. + // Owning a ref on the entire group is more coarse-grained than is strictly + // necessary; all we truly require is that methods we directly reference + // outlive us, while the group could contain many other messages we don't + // require. But the group represents the messages that were + // allocated+compiled together, so it makes the most sense to free them + // together also. + const upb_refcounted *group; + + // Whether this method is native code or bytecode. + bool is_native_; + + // The handler one calls to invoke this method. + upb_byteshandler input_handler_; + + // The destination handlers this method is bound to. We own a ref. + const upb_handlers *dest_handlers_; + + // Dispatch table -- used by both bytecode decoder and JIT when encountering a + // field number that wasn't the one we were expecting to see. See + // decoder.int.h for the layout of this table. + upb_inttable dispatch; +)); + +// A Decoder receives binary protobuf data on its input sink and pushes the +// decoded data to its output sink. +UPB_DEFINE_CLASS0(upb::pb::Decoder, + public: + // Constructs a decoder instance for the given method, which must outlive this + // decoder. Any errors during parsing will be set on the given status, which + // must also outlive this decoder. + Decoder(const DecoderMethod* method, Status* status); + ~Decoder(); + + // Returns the DecoderMethod this decoder is parsing from. + // TODO(haberman): Do users need to be able to rebind this? + const DecoderMethod* method() const; + + // Resets the state of the decoder. + void Reset(); + + // Returns number of bytes successfully parsed. + // + // This can be useful for determining the stream position where an error + // occurred. + // + // This value may not be up-to-date when called from inside a parsing + // callback. + uint64_t BytesParsed() const; + + // Resets the output sink of the Decoder. + // The given sink must match method()->dest_handlers(). + // + // This must be called at least once before the decoder can be used. It may + // only be called with the decoder is in a state where it was just created or + // reset with pipeline.Reset(). The given sink must be from the same pipeline + // as this decoder. + bool ResetOutput(Sink* sink); + + // The sink on which this decoder receives input. + BytesSink* input(); + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(Decoder); +, +UPB_DEFINE_STRUCT0(upb_pbdecoder, UPB_QUOTE( + // Our input sink. + upb_bytessink input_; + + // The decoder method we are parsing with (owned). + const upb_pbdecodermethod *method_; + + size_t call_len; + const uint32_t *pc, *last; + + // Current input buffer and its stream offset. + const char *buf, *ptr, *end, *checkpoint; + + // End of the delimited region, relative to ptr, or NULL if not in this buf. + const char *delim_end; + + // End of the delimited region, relative to ptr, or end if not in this buf. + const char *data_end; + + // Overall stream offset of "buf." + uint64_t bufstart_ofs; + + // Buffer for residual bytes not parsed from the previous buffer. + // The maximum number of residual bytes we require is 12; a five-byte + // unknown tag plus an eight-byte value, less one because the value + // is only a partial value. + char residual[12]; + char *residual_end; + + // Stores the user buffer passed to our decode function. + const char *buf_param; + size_t size_param; + const upb_bufhandle *handle; + +#ifdef UPB_USE_JIT_X64 + // Used momentarily by the generated code to store a value while a user + // function is called. + uint32_t tmp_len; + + const void *saved_rsp; +#endif + + upb_status *status; + + // Our internal stack. + upb_pbdecoder_frame *top, *limit; + upb_pbdecoder_frame stack[UPB_DECODER_MAX_NESTING]; +#ifdef UPB_USE_JIT_X64 + // Each native stack frame needs two pointers, plus we need a few frames for + // the enter/exit trampolines. + const uint32_t *callstack[(UPB_DECODER_MAX_NESTING * 2) + 10]; +#else + const uint32_t *callstack[UPB_DECODER_MAX_NESTING]; +#endif +))); + +// A class for caching protobuf processing code, whether bytecode for the +// interpreted decoder or machine code for the JIT. +// +// This class is not thread-safe. +UPB_DEFINE_CLASS0(upb::pb::CodeCache, + public: + CodeCache(); + ~CodeCache(); + + // Whether the cache is allowed to generate machine code. Defaults to true. + // There is no real reason to turn it off except for testing or if you are + // having a specific problem with the JIT. + // + // Note that allow_jit = true does not *guarantee* that the code will be JIT + // compiled. If this platform is not supported or the JIT was not compiled + // in, the code may still be interpreted. + bool allow_jit() const; + + // This may only be called when the object is first constructed, and prior to + // any code generation, otherwise returns false and does nothing. + bool set_allow_jit(bool allow); + + // Returns a DecoderMethod that can push data to the given handlers. + // If a suitable method already exists, it will be returned from the cache. + // + // Specifying the destination handlers here allows the DecoderMethod to be + // statically bound to the destination handlers if possible, which can allow + // more efficient decoding. However the returned method may or may not + // actually be statically bound. But in all cases, the returned method can + // push data to the given handlers. + const DecoderMethod *GetDecoderMethod(const DecoderMethodOptions& opts); + + // If/when someone needs to explicitly create a dynamically-bound + // DecoderMethod*, we can add a method to get it here. + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(CodeCache); +, +UPB_DEFINE_STRUCT0(upb_pbcodecache, + bool allow_jit_; + + // Array of mgroups. + upb_inttable groups; +)); + +UPB_BEGIN_EXTERN_C // { + +void upb_pbdecoder_init(upb_pbdecoder *d, const upb_pbdecodermethod *method, + upb_status *status); +void upb_pbdecoder_uninit(upb_pbdecoder *d); +void upb_pbdecoder_reset(upb_pbdecoder *d); +const upb_pbdecodermethod *upb_pbdecoder_method(const upb_pbdecoder *d); +bool upb_pbdecoder_resetoutput(upb_pbdecoder *d, upb_sink *sink); +upb_bytessink *upb_pbdecoder_input(upb_pbdecoder *d); +uint64_t upb_pbdecoder_bytesparsed(const upb_pbdecoder *d); + +void upb_pbdecodermethodopts_init(upb_pbdecodermethodopts *opts, + const upb_handlers *h); +void upb_pbdecodermethodopts_setlazy(upb_pbdecodermethodopts *opts, bool lazy); + +void upb_pbdecodermethod_ref(const upb_pbdecodermethod *m, const void *owner); +void upb_pbdecodermethod_unref(const upb_pbdecodermethod *m, const void *owner); +void upb_pbdecodermethod_donateref(const upb_pbdecodermethod *m, + const void *from, const void *to); +void upb_pbdecodermethod_checkref(const upb_pbdecodermethod *m, + const void *owner); +const upb_handlers *upb_pbdecodermethod_desthandlers( + const upb_pbdecodermethod *m); +const upb_byteshandler *upb_pbdecodermethod_inputhandler( + const upb_pbdecodermethod *m); +bool upb_pbdecodermethod_isnative(const upb_pbdecodermethod *m); +const upb_pbdecodermethod *upb_pbdecodermethod_new( + const upb_pbdecodermethodopts *opts, const void *owner); + +void upb_pbcodecache_init(upb_pbcodecache *c); +void upb_pbcodecache_uninit(upb_pbcodecache *c); +bool upb_pbcodecache_allowjit(const upb_pbcodecache *c); +bool upb_pbcodecache_setallowjit(upb_pbcodecache *c, bool allow); +const upb_pbdecodermethod *upb_pbcodecache_getdecodermethod( + upb_pbcodecache *c, const upb_pbdecodermethodopts *opts); + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus + +namespace upb { + +namespace pb { + +inline Decoder::Decoder(const DecoderMethod* m, Status* s) { + upb_pbdecoder_init(this, m, s); +} +inline Decoder::~Decoder() { + upb_pbdecoder_uninit(this); +} +inline const DecoderMethod* Decoder::method() const { + return upb_pbdecoder_method(this); +} +inline void Decoder::Reset() { + upb_pbdecoder_reset(this); +} +inline uint64_t Decoder::BytesParsed() const { + return upb_pbdecoder_bytesparsed(this); +} +inline bool Decoder::ResetOutput(Sink* sink) { + return upb_pbdecoder_resetoutput(this, sink); +} +inline BytesSink* Decoder::input() { + return upb_pbdecoder_input(this); +} + +inline DecoderMethodOptions::DecoderMethodOptions(const Handlers* h) { + upb_pbdecodermethodopts_init(this, h); +} +inline void DecoderMethodOptions::set_lazy(bool lazy) { + upb_pbdecodermethodopts_setlazy(this, lazy); +} + +inline void DecoderMethod::Ref(const void *owner) const { + upb_pbdecodermethod_ref(this, owner); +} +inline void DecoderMethod::Unref(const void *owner) const { + upb_pbdecodermethod_unref(this, owner); +} +inline void DecoderMethod::DonateRef(const void *from, const void *to) const { + upb_pbdecodermethod_donateref(this, from, to); +} +inline void DecoderMethod::CheckRef(const void *owner) const { + upb_pbdecodermethod_checkref(this, owner); +} +inline const Handlers* DecoderMethod::dest_handlers() const { + return upb_pbdecodermethod_desthandlers(this); +} +inline const BytesHandler* DecoderMethod::input_handler() const { + return upb_pbdecodermethod_inputhandler(this); +} +inline bool DecoderMethod::is_native() const { + return upb_pbdecodermethod_isnative(this); +} +// static +inline reffed_ptr DecoderMethod::New( + const DecoderMethodOptions &opts) { + const upb_pbdecodermethod *m = upb_pbdecodermethod_new(&opts, &m); + return reffed_ptr(m, &m); +} + +inline CodeCache::CodeCache() { + upb_pbcodecache_init(this); +} +inline CodeCache::~CodeCache() { + upb_pbcodecache_uninit(this); +} +inline bool CodeCache::allow_jit() const { + return upb_pbcodecache_allowjit(this); +} +inline bool CodeCache::set_allow_jit(bool allow) { + return upb_pbcodecache_setallowjit(this, allow); +} +inline const DecoderMethod *CodeCache::GetDecoderMethod( + const DecoderMethodOptions& opts) { + return upb_pbcodecache_getdecodermethod(this, &opts); +} + +} // namespace pb +} // namespace upb + +#endif // __cplusplus + +#endif /* UPB_DECODER_H_ */ + +// Opcode definitions. The canonical meaning of each opcode is its +// implementation in the interpreter (the JIT is written to match this). +// +// All instructions have the opcode in the low byte. +// Instruction format for most instructions is: +// +// +-------------------+--------+ +// | arg (24) | op (8) | +// +-------------------+--------+ +// +// Exceptions are indicated below. A few opcodes are multi-word. +typedef enum { + // Opcodes 1-8, 13, 15-18 parse their respective descriptor types. + // Arg for all of these is the upb selector for this field. +#define T(type) OP_PARSE_ ## type = UPB_DESCRIPTOR_TYPE_ ## type + T(DOUBLE), T(FLOAT), T(INT64), T(UINT64), T(INT32), T(FIXED64), T(FIXED32), + T(BOOL), T(UINT32), T(SFIXED32), T(SFIXED64), T(SINT32), T(SINT64), +#undef T + OP_STARTMSG = 9, // No arg. + OP_ENDMSG = 10, // No arg. + OP_STARTSEQ = 11, + OP_ENDSEQ = 12, + OP_STARTSUBMSG = 14, + OP_ENDSUBMSG = 19, + OP_STARTSTR = 20, + OP_STRING = 21, + OP_ENDSTR = 22, + + OP_PUSHTAGDELIM = 23, // No arg. + OP_PUSHLENDELIM = 24, // No arg. + OP_POP = 25, // No arg. + OP_SETDELIM = 26, // No arg. + OP_SETBIGGROUPNUM = 27, // two words: | unused (24) | opc || groupnum (32) | + OP_CHECKDELIM = 28, + OP_CALL = 29, + OP_RET = 30, + OP_BRANCH = 31, + + // Different opcodes depending on how many bytes expected. + OP_TAG1 = 32, // | expected tag (16) | jump target (8) | opc (8) | + OP_TAG2 = 33, // | expected tag (16) | jump target (8) | opc (8) | + OP_TAGN = 34, // three words: + // | unused (16) | jump target(8) | opc (8) | + // | expected tag 1 (32) | + // | expected tag 2 (32) | + + OP_SETDISPATCH = 35, // N words: + // | unused (24) | opc | + // | upb_inttable* (32 or 64) | + + OP_DISPATCH = 36, // No arg. + + OP_HALT = 37, // No arg. +} opcode; + +#define OP_MAX OP_HALT + +UPB_INLINE opcode getop(uint32_t instr) { return instr & 0xff; } + +// Method group; represents a set of decoder methods that had their code +// emitted together, and must therefore be freed together. Immutable once +// created. It is possible we may want to expose this to users at some point. +// +// Overall ownership of Decoder objects looks like this: +// +// +----------+ +// | | <---> DecoderMethod +// | method | +// CodeCache ---> | group | <---> DecoderMethod +// | | +// | (mgroup) | <---> DecoderMethod +// +----------+ +typedef struct { + upb_refcounted base; + + // Maps upb_msgdef/upb_handlers -> upb_pbdecodermethod. We own refs on the + // methods. + upb_inttable methods; + + // When we add the ability to link to previously existing mgroups, we'll + // need an array of mgroups we reference here, and own refs on them. + + // The bytecode for our methods, if any exists. Owned by us. + uint32_t *bytecode; + uint32_t *bytecode_end; + +#ifdef UPB_USE_JIT_X64 + // JIT-generated machine code, if any. + upb_string_handlerfunc *jit_code; + // The size of the jit_code (required to munmap()). + size_t jit_size; + char *debug_info; + void *dl; +#endif +} mgroup; + +// Decoder entry points; used as handlers. +void *upb_pbdecoder_startbc(void *closure, const void *pc, size_t size_hint); +void *upb_pbdecoder_startjit(void *closure, const void *hd, size_t size_hint); +size_t upb_pbdecoder_decode(void *closure, const void *hd, const char *buf, + size_t size, const upb_bufhandle *handle); +bool upb_pbdecoder_end(void *closure, const void *handler_data); + +// Decoder-internal functions that the JIT calls to handle fallback paths. +int32_t upb_pbdecoder_resume(upb_pbdecoder *d, void *p, const char *buf, + size_t size, const upb_bufhandle *handle); +size_t upb_pbdecoder_suspend(upb_pbdecoder *d); +int32_t upb_pbdecoder_skipunknown(upb_pbdecoder *d, int32_t fieldnum, + uint8_t wire_type); +int32_t upb_pbdecoder_checktag_slow(upb_pbdecoder *d, uint64_t expected); +int32_t upb_pbdecoder_decode_varint_slow(upb_pbdecoder *d, uint64_t *u64); +int32_t upb_pbdecoder_decode_f32(upb_pbdecoder *d, uint32_t *u32); +int32_t upb_pbdecoder_decode_f64(upb_pbdecoder *d, uint64_t *u64); +void upb_pbdecoder_seterr(upb_pbdecoder *d, const char *msg); + +// Error messages that are shared between the bytecode and JIT decoders. +extern const char *kPbDecoderStackOverflow; + +// Access to decoderplan members needed by the decoder. +const char *upb_pbdecoder_getopname(unsigned int op); + +// JIT codegen entry point. +void upb_pbdecoder_jit(mgroup *group); +void upb_pbdecoder_freejit(mgroup *group); + +// A special label that means "do field dispatch for this message and branch to +// wherever that takes you." +#define LABEL_DISPATCH 0 + +// A special slot in the dispatch table that stores the epilogue (ENDMSG and/or +// RET) for branching to when we find an appropriate ENDGROUP tag. +#define DISPATCH_ENDMSG 0 + +// It's important to use this invalid wire type instead of 0 (which is a valid +// wire type). +#define NO_WIRE_TYPE 0xff + +// The dispatch table layout is: +// [field number] -> [ 48-bit offset ][ 8-bit wt2 ][ 8-bit wt1 ] +// +// If wt1 matches, jump to the 48-bit offset. If wt2 matches, lookup +// (UPB_MAX_FIELDNUMBER + fieldnum) and jump there. +// +// We need two wire types because of packed/non-packed compatibility. A +// primitive repeated field can use either wire type and be valid. While we +// could key the table on fieldnum+wiretype, the table would be 8x sparser. +// +// Storing two wire types in the primary value allows us to quickly rule out +// the second wire type without needing to do a separate lookup (this case is +// less common than an unknown field). +UPB_INLINE uint64_t upb_pbdecoder_packdispatch(uint64_t ofs, uint8_t wt1, + uint8_t wt2) { + return (ofs << 16) | (wt2 << 8) | wt1; +} + +UPB_INLINE void upb_pbdecoder_unpackdispatch(uint64_t dispatch, uint64_t *ofs, + uint8_t *wt1, uint8_t *wt2) { + *wt1 = (uint8_t)dispatch; + *wt2 = (uint8_t)(dispatch >> 8); + *ofs = dispatch >> 16; +} + +// All of the functions in decoder.c that return int32_t return values according +// to the following scheme: +// 1. negative values indicate a return code from the following list. +// 2. positive values indicate that error or end of buffer was hit, and +// that the decode function should immediately return the given value +// (the decoder state has already been suspended and is ready to be +// resumed). +#define DECODE_OK -1 +#define DECODE_MISMATCH -2 // Used only from checktag_slow(). +#define DECODE_ENDGROUP -3 // Used only from checkunknown(). + +#define CHECK_RETURN(x) { int32_t ret = x; if (ret >= 0) return ret; } + +#endif // UPB_DECODER_INT_H_ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2011 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * A number of routines for varint manipulation (we keep them all around to + * have multiple approaches available for benchmarking). + */ + +#ifndef UPB_VARINT_DECODER_H_ +#define UPB_VARINT_DECODER_H_ + +#include +#include +#include + +#ifdef __cplusplus +extern "C" { +#endif + +// A list of types as they are encoded on-the-wire. +typedef enum { + UPB_WIRE_TYPE_VARINT = 0, + UPB_WIRE_TYPE_64BIT = 1, + UPB_WIRE_TYPE_DELIMITED = 2, + UPB_WIRE_TYPE_START_GROUP = 3, + UPB_WIRE_TYPE_END_GROUP = 4, + UPB_WIRE_TYPE_32BIT = 5, +} upb_wiretype_t; + +#define UPB_MAX_WIRE_TYPE 5 + +// The maximum number of bytes that it takes to encode a 64-bit varint. +// Note that with a better encoding this could be 9 (TODO: write up a +// wiki document about this). +#define UPB_PB_VARINT_MAX_LEN 10 + +// Array of the "native" (ie. non-packed-repeated) wire type for the given a +// descriptor type (upb_descriptortype_t). +extern const uint8_t upb_pb_native_wire_types[]; + +/* Zig-zag encoding/decoding **************************************************/ + +UPB_INLINE int32_t upb_zzdec_32(uint32_t n) { + return (n >> 1) ^ -(int32_t)(n & 1); +} +UPB_INLINE int64_t upb_zzdec_64(uint64_t n) { + return (n >> 1) ^ -(int64_t)(n & 1); +} +UPB_INLINE uint32_t upb_zzenc_32(int32_t n) { return (n << 1) ^ (n >> 31); } +UPB_INLINE uint64_t upb_zzenc_64(int64_t n) { return (n << 1) ^ (n >> 63); } + +/* Decoding *******************************************************************/ + +// All decoding functions return this struct by value. +typedef struct { + const char *p; // NULL if the varint was unterminated. + uint64_t val; +} upb_decoderet; + +// Four functions for decoding a varint of at most eight bytes. They are all +// functionally identical, but are implemented in different ways and likely have +// different performance profiles. We keep them around for performance testing. +// +// Note that these functions may not read byte-by-byte, so they must not be used +// unless there are at least eight bytes left in the buffer! +upb_decoderet upb_vdecode_max8_branch32(upb_decoderet r); +upb_decoderet upb_vdecode_max8_branch64(upb_decoderet r); +upb_decoderet upb_vdecode_max8_wright(upb_decoderet r); +upb_decoderet upb_vdecode_max8_massimino(upb_decoderet r); + +// Template for a function that checks the first two bytes with branching +// and dispatches 2-10 bytes with a separate function. Note that this may read +// up to 10 bytes, so it must not be used unless there are at least ten bytes +// left in the buffer! +#define UPB_VARINT_DECODER_CHECK2(name, decode_max8_function) \ +UPB_INLINE upb_decoderet upb_vdecode_check2_ ## name(const char *_p) { \ + uint8_t *p = (uint8_t*)_p; \ + if ((*p & 0x80) == 0) { upb_decoderet r = {_p + 1, *p & 0x7fU}; return r; } \ + upb_decoderet r = {_p + 2, (*p & 0x7fU) | ((*(p + 1) & 0x7fU) << 7)}; \ + if ((*(p + 1) & 0x80) == 0) return r; \ + return decode_max8_function(r); \ +} + +UPB_VARINT_DECODER_CHECK2(branch32, upb_vdecode_max8_branch32); +UPB_VARINT_DECODER_CHECK2(branch64, upb_vdecode_max8_branch64); +UPB_VARINT_DECODER_CHECK2(wright, upb_vdecode_max8_wright); +UPB_VARINT_DECODER_CHECK2(massimino, upb_vdecode_max8_massimino); +#undef UPB_VARINT_DECODER_CHECK2 + +// Our canonical functions for decoding varints, based on the currently +// favored best-performing implementations. +UPB_INLINE upb_decoderet upb_vdecode_fast(const char *p) { + if (sizeof(long) == 8) + return upb_vdecode_check2_branch64(p); + else + return upb_vdecode_check2_branch32(p); +} + +UPB_INLINE upb_decoderet upb_vdecode_max8_fast(upb_decoderet r) { + return upb_vdecode_max8_massimino(r); +} + + +/* Encoding *******************************************************************/ + +UPB_INLINE int upb_value_size(uint64_t val) { +#ifdef __GNUC__ + int high_bit = 63 - __builtin_clzll(val); // 0-based, undef if val == 0. +#else + int high_bit = 0; + uint64_t tmp = val; + while(tmp >>= 1) high_bit++; +#endif + return val == 0 ? 1 : high_bit / 8 + 1; +} + +// Encodes a 64-bit varint into buf (which must be >=UPB_PB_VARINT_MAX_LEN +// bytes long), returning how many bytes were used. +// +// TODO: benchmark and optimize if necessary. +UPB_INLINE size_t upb_vencode64(uint64_t val, char *buf) { + if (val == 0) { buf[0] = 0; return 1; } + size_t i = 0; + while (val) { + uint8_t byte = val & 0x7fU; + val >>= 7; + if (val) byte |= 0x80U; + buf[i++] = byte; + } + return i; +} + +UPB_INLINE size_t upb_varint_size(uint64_t val) { + char buf[UPB_PB_VARINT_MAX_LEN]; + return upb_vencode64(val, buf); +} + +// Encodes a 32-bit varint, *not* sign-extended. +UPB_INLINE uint64_t upb_vencode32(uint32_t val) { + char buf[UPB_PB_VARINT_MAX_LEN]; + size_t bytes = upb_vencode64(val, buf); + uint64_t ret = 0; + assert(bytes <= 5); + memcpy(&ret, buf, bytes); + assert(ret <= 0xffffffffffU); + return ret; +} + +#ifdef __cplusplus +} /* extern "C" */ +#endif + +#endif /* UPB_VARINT_DECODER_H_ */ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009-2010 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * Implements a set of upb_handlers that write protobuf data to the binary wire + * format. + * + * This encoder implementation does not have any access to any out-of-band or + * precomputed lengths for submessages, so it must buffer submessages internally + * before it can emit the first byte. + */ + +#ifndef UPB_ENCODER_H_ +#define UPB_ENCODER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace pb { +class Encoder; +} // namespace pb +} // namespace upb +#endif + +UPB_DECLARE_TYPE(upb::pb::Encoder, upb_pb_encoder); + +#define UPB_PBENCODER_MAX_NESTING 100 + +/* upb::pb::Encoder ***********************************************************/ + +// The output buffer is divided into segments; a segment is a string of data +// that is "ready to go" -- it does not need any varint lengths inserted into +// the middle. The seams between segments are where varints will be inserted +// once they are known. +// +// We also use the concept of a "run", which is a range of encoded bytes that +// occur at a single submessage level. Every segment contains one or more runs. +// +// A segment can span messages. Consider: +// +// .--Submessage lengths---------. +// | | | +// | V V +// V | |--------------- | |----------------- +// Submessages: | |----------------------------------------------- +// Top-level msg: ------------------------------------------------------------ +// +// Segments: ----- ------------------- ----------------- +// Runs: *---- *--------------*--- *---------------- +// (* marks the start) +// +// Note that the top-level menssage is not in any segment because it does not +// have any length preceding it. +// +// A segment is only interrupted when another length needs to be inserted. So +// observe how the second segment spans both the inner submessage and part of +// the next enclosing message. +typedef struct { + UPB_PRIVATE_FOR_CPP + uint32_t msglen; // The length to varint-encode before this segment. + uint32_t seglen; // Length of the segment. +} upb_pb_encoder_segment; + +UPB_DEFINE_CLASS0(upb::pb::Encoder, + public: + Encoder(const upb::Handlers* handlers); + ~Encoder(); + + static reffed_ptr NewHandlers(const upb::MessageDef* msg); + + // Resets the state of the printer, so that it will expect to begin a new + // document. + void Reset(); + + // Resets the output pointer which will serve as our closure. + void ResetOutput(BytesSink* output); + + // The input to the encoder. + Sink* input(); + + private: + UPB_DISALLOW_COPY_AND_ASSIGN(Encoder); +, +UPB_DEFINE_STRUCT0(upb_pb_encoder, UPB_QUOTE( + // Our input and output. + upb_sink input_; + upb_bytessink *output_; + + // The "subclosure" -- used as the inner closure as part of the bytessink + // protocol. + void *subc; + + // The output buffer and limit, and our current write position. "buf" + // initially points to "initbuf", but is dynamically allocated if we need to + // grow beyond the initial size. + char *buf, *ptr, *limit; + + // The beginning of the current run, or undefined if we are at the top level. + char *runbegin; + + // The list of segments we are accumulating. + upb_pb_encoder_segment *segbuf, *segptr, *seglimit; + + // The stack of enclosing submessages. Each entry in the stack points to the + // segment where this submessage's length is being accumulated. + int stack[UPB_PBENCODER_MAX_NESTING], *top, *stacklimit; + + // Depth of startmsg/endmsg calls. + int depth; + + // Initial buffers for the output buffer and segment buffer. If we outgrow + // these we will dynamically allocate bigger ones. + char initbuf[256]; + upb_pb_encoder_segment seginitbuf[32]; +))); + +UPB_BEGIN_EXTERN_C + +const upb_handlers *upb_pb_encoder_newhandlers(const upb_msgdef *m, + const void *owner); +void upb_pb_encoder_reset(upb_pb_encoder *e); +upb_sink *upb_pb_encoder_input(upb_pb_encoder *p); +void upb_pb_encoder_init(upb_pb_encoder *e, const upb_handlers *h); +void upb_pb_encoder_resetoutput(upb_pb_encoder *e, upb_bytessink *output); +void upb_pb_encoder_uninit(upb_pb_encoder *e); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { +namespace pb { +inline Encoder::Encoder(const upb::Handlers* handlers) { + upb_pb_encoder_init(this, handlers); +} +inline Encoder::~Encoder() { + upb_pb_encoder_uninit(this); +} +inline void Encoder::Reset() { + upb_pb_encoder_reset(this); +} +inline void Encoder::ResetOutput(BytesSink* output) { + upb_pb_encoder_resetoutput(this, output); +} +inline Sink* Encoder::input() { + return upb_pb_encoder_input(this); +} +inline reffed_ptr Encoder::NewHandlers( + const upb::MessageDef *md) { + const Handlers* h = upb_pb_encoder_newhandlers(md, &h); + return reffed_ptr(h, &h); +} +} // namespace pb +} // namespace upb + +#endif + +#endif /* UPB_ENCODER_H_ */ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2011-2012 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * upb's core components like upb_decoder and upb_msg are carefully designed to + * avoid depending on each other for maximum orthogonality. In other words, + * you can use a upb_decoder to decode into *any* kind of structure; upb_msg is + * just one such structure. A upb_msg can be serialized/deserialized into any + * format, protobuf binary format is just one such format. + * + * However, for convenience we provide functions here for doing common + * operations like deserializing protobuf binary format into a upb_msg. The + * compromise is that this file drags in almost all of upb as a dependency, + * which could be undesirable if you're trying to use a trimmed-down build of + * upb. + * + * While these routines are convenient, they do not reuse any encoding/decoding + * state. For example, if a decoder is JIT-based, it will be re-JITted every + * time these functions are called. For this reason, if you are parsing lots + * of data and efficiency is an issue, these may not be the best functions to + * use (though they are useful for prototyping, before optimizing). + */ + +#ifndef UPB_GLUE_H +#define UPB_GLUE_H + +#include + +#ifdef __cplusplus +extern "C" { +#endif + +// Loads all defs from the given protobuf binary descriptor, setting default +// accessors and a default layout on all messages. The caller owns the +// returned array of defs, which will be of length *n. On error NULL is +// returned and status is set (if non-NULL). +upb_def **upb_load_defs_from_descriptor(const char *str, size_t len, int *n, + void *owner, upb_status *status); + +// Like the previous but also adds the loaded defs to the given symtab. +bool upb_load_descriptor_into_symtab(upb_symtab *symtab, const char *str, + size_t len, upb_status *status); + +// Like the previous but also reads the descriptor from the given filename. +bool upb_load_descriptor_file_into_symtab(upb_symtab *symtab, const char *fname, + upb_status *status); + +// Reads the given filename into a character string, returning NULL if there +// was an error. +char *upb_readfile(const char *filename, size_t *len); + +#ifdef __cplusplus +} /* extern "C" */ + +namespace upb { + +// All routines that load descriptors expect the descriptor to be a +// FileDescriptorSet. +inline bool LoadDescriptorFileIntoSymtab(SymbolTable* s, const char *fname, + Status* status) { + return upb_load_descriptor_file_into_symtab(s, fname, status); +} + +inline bool LoadDescriptorIntoSymtab(SymbolTable* s, const char* str, + size_t len, Status* status) { + return upb_load_descriptor_into_symtab(s, str, len, status); +} + +// Templated so it can accept both string and std::string. +template +bool LoadDescriptorIntoSymtab(SymbolTable* s, const T& desc, Status* status) { + return upb_load_descriptor_into_symtab(s, desc.c_str(), desc.size(), status); +} + +} // namespace upb + +#endif + +#endif +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2009 Google Inc. See LICENSE for details. + * Author: Josh Haberman + */ + +#ifndef UPB_TEXT_H_ +#define UPB_TEXT_H_ + + +#ifdef __cplusplus +namespace upb { +namespace pb { +class TextPrinter; +} // namespace pb +} // namespace upb +#endif + +UPB_DECLARE_TYPE(upb::pb::TextPrinter, upb_textprinter); + +UPB_DEFINE_CLASS0(upb::pb::TextPrinter, + public: + // The given handlers must have come from NewHandlers(). It must outlive the + // TextPrinter. + explicit TextPrinter(const upb::Handlers* handlers); + + void SetSingleLineMode(bool single_line); + + bool ResetOutput(BytesSink* output); + Sink* input(); + + // If handler caching becomes a requirement we can add a code cache as in + // decoder.h + static reffed_ptr NewHandlers(const MessageDef* md); + + private: +, +UPB_DEFINE_STRUCT0(upb_textprinter, + upb_sink input_; + upb_bytessink *output_; + int indent_depth_; + bool single_line_; + void *subc; +)); + +UPB_BEGIN_EXTERN_C // { + +// C API. +void upb_textprinter_init(upb_textprinter *p, const upb_handlers *h); +void upb_textprinter_uninit(upb_textprinter *p); +bool upb_textprinter_resetoutput(upb_textprinter *p, upb_bytessink *output); +void upb_textprinter_setsingleline(upb_textprinter *p, bool single_line); +upb_sink *upb_textprinter_input(upb_textprinter *p); + +const upb_handlers *upb_textprinter_newhandlers(const upb_msgdef *m, + const void *owner); + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus + +namespace upb { +namespace pb { +inline TextPrinter::TextPrinter(const upb::Handlers* handlers) { + upb_textprinter_init(this, handlers); +} +inline void TextPrinter::SetSingleLineMode(bool single_line) { + upb_textprinter_setsingleline(this, single_line); +} +inline bool TextPrinter::ResetOutput(BytesSink* output) { + return upb_textprinter_resetoutput(this, output); +} +inline Sink* TextPrinter::input() { + return upb_textprinter_input(this); +} +inline reffed_ptr TextPrinter::NewHandlers( + const MessageDef *md) { + const Handlers* h = upb_textprinter_newhandlers(md, &h); + return reffed_ptr(h, &h); +} +} // namespace pb +} // namespace upb + +#endif + +#endif /* UPB_TEXT_H_ */ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2014 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * upb::json::Parser can parse JSON according to a specific schema. + * Support for parsing arbitrary JSON (schema-less) will be added later. + */ + +#ifndef UPB_JSON_PARSER_H_ +#define UPB_JSON_PARSER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace json { +class Parser; +} // namespace json +} // namespace upb +#endif + +UPB_DECLARE_TYPE(upb::json::Parser, upb_json_parser); + +// Internal-only struct used by the parser. A parser frame corresponds +// one-to-one with a handler (sink) frame. +typedef struct { + UPB_PRIVATE_FOR_CPP + upb_sink sink; + // The current message in which we're parsing, and the field whose value we're + // expecting next. + const upb_msgdef *m; + const upb_fielddef *f; + + // We are in a repeated-field context, ready to emit mapentries as + // submessages. This flag alters the start-of-object (open-brace) behavior to + // begin a sequence of mapentry messages rather than a single submessage. + bool is_map; + // We are in a map-entry message context. This flag is set when parsing the + // value field of a single map entry and indicates to all value-field parsers + // (subobjects, strings, numbers, and bools) that the map-entry submessage + // should end as soon as the value is parsed. + bool is_mapentry; + // If |is_map| or |is_mapentry| is true, |mapfield| refers to the parent + // message's map field that we're currently parsing. This differs from |f| + // because |f| is the field in the *current* message (i.e., the map-entry + // message itself), not the parent's field that leads to this map. + const upb_fielddef *mapfield; +} upb_jsonparser_frame; + + +/* upb::json::Parser **********************************************************/ + +#define UPB_JSON_MAX_DEPTH 64 + +// Parses an incoming BytesStream, pushing the results to the destination sink. +UPB_DEFINE_CLASS0(upb::json::Parser, + public: + Parser(Status* status); + ~Parser(); + + // Resets the state of the printer, so that it will expect to begin a new + // document. + void Reset(); + + // Resets the output pointer which will serve as our closure. Implies + // Reset(). + void ResetOutput(Sink* output); + + // The input to the printer. + BytesSink* input(); +, +UPB_DEFINE_STRUCT0(upb_json_parser, + upb_byteshandler input_handler_; + upb_bytessink input_; + + // Stack to track the JSON scopes we are in. + upb_jsonparser_frame stack[UPB_JSON_MAX_DEPTH]; + upb_jsonparser_frame *top; + upb_jsonparser_frame *limit; + + upb_status *status; + + // Ragel's internal parsing stack for the parsing state machine. + int current_state; + int parser_stack[UPB_JSON_MAX_DEPTH]; + int parser_top; + + // The handle for the current buffer. + const upb_bufhandle *handle; + + // Accumulate buffer. See details in parser.rl. + const char *accumulated; + size_t accumulated_len; + char *accumulate_buf; + size_t accumulate_buf_size; + + // Multi-part text data. See details in parser.rl. + int multipart_state; + upb_selector_t string_selector; + + // Input capture. See details in parser.rl. + const char *capture; + + // Intermediate result of parsing a unicode escape sequence. + uint32_t digit; +)); + +UPB_BEGIN_EXTERN_C + +void upb_json_parser_init(upb_json_parser *p, upb_status *status); +void upb_json_parser_uninit(upb_json_parser *p); +void upb_json_parser_reset(upb_json_parser *p); +void upb_json_parser_resetoutput(upb_json_parser *p, upb_sink *output); +upb_bytessink *upb_json_parser_input(upb_json_parser *p); + +UPB_END_EXTERN_C + +#ifdef __cplusplus + +namespace upb { +namespace json { +inline Parser::Parser(Status* status) { upb_json_parser_init(this, status); } +inline Parser::~Parser() { upb_json_parser_uninit(this); } +inline void Parser::Reset() { upb_json_parser_reset(this); } +inline void Parser::ResetOutput(Sink* output) { + upb_json_parser_resetoutput(this, output); +} +inline BytesSink* Parser::input() { + return upb_json_parser_input(this); +} +} // namespace json +} // namespace upb + +#endif + + +#endif // UPB_JSON_PARSER_H_ +/* + * upb - a minimalist implementation of protocol buffers. + * + * Copyright (c) 2014 Google Inc. See LICENSE for details. + * Author: Josh Haberman + * + * upb::json::Printer allows you to create handlers that emit JSON + * according to a specific protobuf schema. + */ + +#ifndef UPB_JSON_TYPED_PRINTER_H_ +#define UPB_JSON_TYPED_PRINTER_H_ + + +#ifdef __cplusplus +namespace upb { +namespace json { +class Printer; +} // namespace json +} // namespace upb +#endif + +UPB_DECLARE_TYPE(upb::json::Printer, upb_json_printer); + + +/* upb::json::Printer *********************************************************/ + +// Prints an incoming stream of data to a BytesSink in JSON format. +UPB_DEFINE_CLASS0(upb::json::Printer, + public: + Printer(const upb::Handlers* handlers); + ~Printer(); + + // Resets the state of the printer, so that it will expect to begin a new + // document. + void Reset(); + + // Resets the output pointer which will serve as our closure. Implies + // Reset(). + void ResetOutput(BytesSink* output); + + // The input to the printer. + Sink* input(); + + // Returns handlers for printing according to the specified schema. + static reffed_ptr NewHandlers(const upb::MessageDef* md); +, +UPB_DEFINE_STRUCT0(upb_json_printer, + upb_sink input_; + // BytesSink closure. + void *subc_; + upb_bytessink *output_; + + // We track the depth so that we know when to emit startstr/endstr on the + // output. + int depth_; + // Have we emitted the first element? This state is necessary to emit commas + // without leaving a trailing comma in arrays/maps. We keep this state per + // frame depth. + // + // Why max_depth * 2? UPB_MAX_HANDLER_DEPTH counts depth as nested messages. + // We count frames (contexts in which we separate elements by commas) as both + // repeated fields and messages (maps), and the worst case is a + // message->repeated field->submessage->repeated field->... nesting. + bool first_elem_[UPB_MAX_HANDLER_DEPTH * 2]; +)); + +UPB_BEGIN_EXTERN_C // { + +// Native C API. + +void upb_json_printer_init(upb_json_printer *p, const upb_handlers *h); +void upb_json_printer_uninit(upb_json_printer *p); +void upb_json_printer_reset(upb_json_printer *p); +void upb_json_printer_resetoutput(upb_json_printer *p, upb_bytessink *output); +upb_sink *upb_json_printer_input(upb_json_printer *p); +const upb_handlers *upb_json_printer_newhandlers(const upb_msgdef *md, + const void *owner); + +UPB_END_EXTERN_C // } + +#ifdef __cplusplus + +namespace upb { +namespace json { +inline Printer::Printer(const upb::Handlers* handlers) { + upb_json_printer_init(this, handlers); +} +inline Printer::~Printer() { upb_json_printer_uninit(this); } +inline void Printer::Reset() { upb_json_printer_reset(this); } +inline void Printer::ResetOutput(BytesSink* output) { + upb_json_printer_resetoutput(this, output); +} +inline Sink* Printer::input() { return upb_json_printer_input(this); } +inline reffed_ptr Printer::NewHandlers( + const upb::MessageDef *md) { + const Handlers* h = upb_json_printer_newhandlers(md, &h); + return reffed_ptr(h, &h); +} +} // namespace json +} // namespace upb + +#endif + +#endif // UPB_JSON_TYPED_PRINTER_H_ diff --git a/ruby/ext/protobuf_c/extconf.rb b/ruby/ext/protobuf_c/extconf.rb deleted file mode 100644 index 7f23b1a829..0000000000 --- a/ruby/ext/protobuf_c/extconf.rb +++ /dev/null @@ -1,23 +0,0 @@ -#!/usr/bin/ruby - -require 'mkmf' - -upb_path = File.absolute_path(File.dirname($0)) + "/../../../upb" -libs = ["upb_pic", "upb.pb_pic", "upb.json_pic"] -system("cd #{upb_path}; make " + libs.map{|l| "lib/lib#{l}.a"}.join(" ")) - -$CFLAGS += " -O3 -std=c99 -Wno-unused-function -DNDEBUG" - -find_header("upb/upb.h", upb_path) or - raise "Can't find upb headers" -find_library("upb_pic", "upb_msgdef_new", upb_path + "/lib") or - raise "Can't find upb lib" -find_library("upb.pb_pic", "upb_pbdecoder_init", upb_path + "/lib") or - raise "Can't find upb.pb lib" -find_library("upb.json_pic", "upb_json_printer_init", upb_path + "/lib") or - raise "Can't find upb.pb lib" - -$objs = ["protobuf.o", "defs.o", "storage.o", "message.o", - "repeated_field.o", "encode_decode.o"] - -create_makefile("protobuf_c") diff --git a/ruby/google-protobuf.gemspec b/ruby/google-protobuf.gemspec new file mode 100644 index 0000000000..7bfa533c6e --- /dev/null +++ b/ruby/google-protobuf.gemspec @@ -0,0 +1,24 @@ +class << Gem::Specification + def find_c_source(dir) + `cd #{dir}; git ls-files "*.c" "*.h" extconf.rb Makefile`.split + .map{|f| "#{dir}/#{f}"} + end +end + +Gem::Specification.new do |s| + s.name = "google-protobuf" + s.version = "3.0.0.alpha.2" + s.licenses = ["BSD"] + s.summary = "Protocol Buffers" + s.description = "Protocol Buffers are Google's data interchange format." + s.authors = ["Protobuf Authors"] + s.email = "protobuf@googlegroups.com" + s.require_paths = ["lib"] + s.extensions = ["ext/google/protobuf_c/extconf.rb"] + s.files = ["lib/google/protobuf.rb"] + + # extension C source + find_c_source("ext/google/protobuf_c") + s.test_files = ["tests/basic.rb", + "tests/stress.rb", + "tests/generated_code_test.rb"] +end diff --git a/ruby/lib/protobuf.rb b/ruby/lib/google/protobuf.rb similarity index 98% rename from ruby/lib/protobuf.rb rename to ruby/lib/google/protobuf.rb index 2ace48f955..74053332f5 100644 --- a/ruby/lib/protobuf.rb +++ b/ruby/lib/google/protobuf.rb @@ -28,4 +28,4 @@ # (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -require 'protobuf_c' +require 'google/protobuf_c' diff --git a/ruby/tests/basic.rb b/ruby/tests/basic.rb index 237fbb06f8..a78cc39449 100644 --- a/ruby/tests/basic.rb +++ b/ruby/tests/basic.rb @@ -1,6 +1,6 @@ #!/usr/bin/ruby -require 'protobuf' +require 'google/protobuf' require 'test/unit' # ------------- generated code -------------- @@ -36,23 +36,52 @@ module BasicTest add_message "TestMessage2" do optional :foo, :int32, 1 end + add_message "Recursive1" do optional :foo, :message, 1, "Recursive2" end add_message "Recursive2" do optional :foo, :message, 1, "Recursive1" end + add_enum "TestEnum" do value :Default, 0 value :A, 1 value :B, 2 value :C, 3 end + add_message "BadFieldNames" do optional :dup, :int32, 1 optional :class, :int32, 2 optional :"a.b", :int32, 3 end + + add_message "MapMessage" do + map :map_string_int32, :string, :int32, 1 + map :map_string_msg, :string, :message, 2, "TestMessage2" + end + add_message "MapMessageWireEquiv" do + repeated :map_string_int32, :message, 1, "MapMessageWireEquiv_entry1" + repeated :map_string_msg, :message, 2, "MapMessageWireEquiv_entry2" + end + add_message "MapMessageWireEquiv_entry1" do + optional :key, :string, 1 + optional :value, :int32, 2 + end + add_message "MapMessageWireEquiv_entry2" do + optional :key, :string, 1 + optional :value, :message, 2, "TestMessage2" + end + + add_message "OneofMessage" do + oneof :my_oneof do + optional :a, :string, 1 + optional :b, :int32, 2 + optional :c, :message, 3, "TestMessage2" + optional :d, :enum, 4, "TestEnum" + end + end end TestMessage = pool.lookup("TestMessage").msgclass @@ -61,6 +90,13 @@ module BasicTest Recursive2 = pool.lookup("Recursive2").msgclass TestEnum = pool.lookup("TestEnum").enummodule BadFieldNames = pool.lookup("BadFieldNames").msgclass + MapMessage = pool.lookup("MapMessage").msgclass + MapMessageWireEquiv = pool.lookup("MapMessageWireEquiv").msgclass + MapMessageWireEquiv_entry1 = + pool.lookup("MapMessageWireEquiv_entry1").msgclass + MapMessageWireEquiv_entry2 = + pool.lookup("MapMessageWireEquiv_entry2").msgclass + OneofMessage = pool.lookup("OneofMessage").msgclass # ------------ test cases --------------- @@ -200,7 +236,8 @@ module BasicTest assert l.count == 0 l = Google::Protobuf::RepeatedField.new(:int32, [1, 2, 3]) assert l.count == 3 - assert l == [1, 2, 3] + assert_equal [1, 2, 3], l + assert_equal l, [1, 2, 3] l.push 4 assert l == [1, 2, 3, 4] dst_list = [] @@ -300,7 +337,7 @@ module BasicTest l.push :B l.push :C assert l.count == 3 - assert_raise NameError do + assert_raise RangeError do l.push :D end assert l[0] == :A @@ -324,12 +361,329 @@ module BasicTest end end + def test_map_basic + # allowed key types: + # :int32, :int64, :uint32, :uint64, :bool, :string, :bytes. + + m = Google::Protobuf::Map.new(:string, :int32) + m["asdf"] = 1 + assert m["asdf"] == 1 + m["jkl;"] = 42 + assert m == { "jkl;" => 42, "asdf" => 1 } + assert m.has_key?("asdf") + assert !m.has_key?("qwerty") + assert m.length == 2 + + m2 = m.dup + assert m == m2 + assert m.hash != 0 + assert m.hash == m2.hash + + collected = {} + m.each { |k,v| collected[v] = k } + assert collected == { 42 => "jkl;", 1 => "asdf" } + + assert m.delete("asdf") == 1 + assert !m.has_key?("asdf") + assert m["asdf"] == nil + assert !m.has_key?("asdf") + + # We only assert on inspect value when there is one map entry because the + # order in which elements appear is unspecified (depends on the internal + # hash function). We don't want a brittle test. + assert m.inspect == "{\"jkl;\"=>42}" + + assert m.keys == ["jkl;"] + assert m.values == [42] + + m.clear + assert m.length == 0 + assert m == {} + + assert_raise TypeError do + m[1] = 1 + end + assert_raise RangeError do + m["asdf"] = 0x1_0000_0000 + end + end + + def test_map_ctor + m = Google::Protobuf::Map.new(:string, :int32, + {"a" => 1, "b" => 2, "c" => 3}) + assert m == {"a" => 1, "c" => 3, "b" => 2} + end + + def test_map_keytypes + m = Google::Protobuf::Map.new(:int32, :int32) + m[1] = 42 + m[-1] = 42 + assert_raise RangeError do + m[0x8000_0000] = 1 + end + assert_raise TypeError do + m["asdf"] = 1 + end + + m = Google::Protobuf::Map.new(:int64, :int32) + m[0x1000_0000_0000_0000] = 1 + assert_raise RangeError do + m[0x1_0000_0000_0000_0000] = 1 + end + assert_raise TypeError do + m["asdf"] = 1 + end + + m = Google::Protobuf::Map.new(:uint32, :int32) + m[0x8000_0000] = 1 + assert_raise RangeError do + m[0x1_0000_0000] = 1 + end + assert_raise RangeError do + m[-1] = 1 + end + + m = Google::Protobuf::Map.new(:uint64, :int32) + m[0x8000_0000_0000_0000] = 1 + assert_raise RangeError do + m[0x1_0000_0000_0000_0000] = 1 + end + assert_raise RangeError do + m[-1] = 1 + end + + m = Google::Protobuf::Map.new(:bool, :int32) + m[true] = 1 + m[false] = 2 + assert_raise TypeError do + m[1] = 1 + end + assert_raise TypeError do + m["asdf"] = 1 + end + + m = Google::Protobuf::Map.new(:string, :int32) + m["asdf"] = 1 + assert_raise TypeError do + m[1] = 1 + end + assert_raise TypeError do + bytestring = ["FFFF"].pack("H*") + m[bytestring] = 1 + end + + m = Google::Protobuf::Map.new(:bytes, :int32) + bytestring = ["FFFF"].pack("H*") + m[bytestring] = 1 + assert_raise TypeError do + m["asdf"] = 1 + end + assert_raise TypeError do + m[1] = 1 + end + end + + def test_map_msg_enum_valuetypes + m = Google::Protobuf::Map.new(:string, :message, TestMessage) + m["asdf"] = TestMessage.new + assert_raise TypeError do + m["jkl;"] = TestMessage2.new + end + + m = Google::Protobuf::Map.new( + :string, :message, TestMessage, + { "a" => TestMessage.new(:optional_int32 => 42), + "b" => TestMessage.new(:optional_int32 => 84) }) + assert m.length == 2 + assert m.values.map{|msg| msg.optional_int32}.sort == [42, 84] + + m = Google::Protobuf::Map.new(:string, :enum, TestEnum, + { "x" => :A, "y" => :B, "z" => :C }) + assert m.length == 3 + assert m["z"] == :C + m["z"] = 2 + assert m["z"] == :B + m["z"] = 4 + assert m["z"] == 4 + assert_raise RangeError do + m["z"] = :Z + end + assert_raise TypeError do + m["z"] = "z" + end + end + + def test_map_dup_deep_copy + m = Google::Protobuf::Map.new( + :string, :message, TestMessage, + { "a" => TestMessage.new(:optional_int32 => 42), + "b" => TestMessage.new(:optional_int32 => 84) }) + + m2 = m.dup + assert m == m2 + assert m.object_id != m2.object_id + assert m["a"].object_id == m2["a"].object_id + assert m["b"].object_id == m2["b"].object_id + + m2 = Google::Protobuf.deep_copy(m) + assert m == m2 + assert m.object_id != m2.object_id + assert m["a"].object_id != m2["a"].object_id + assert m["b"].object_id != m2["b"].object_id + end + + def test_map_field + m = MapMessage.new + assert m.map_string_int32 == {} + assert m.map_string_msg == {} + + m = MapMessage.new( + :map_string_int32 => {"a" => 1, "b" => 2}, + :map_string_msg => {"a" => TestMessage2.new(:foo => 1), + "b" => TestMessage2.new(:foo => 2)}) + assert m.map_string_int32.keys.sort == ["a", "b"] + assert m.map_string_int32["a"] == 1 + assert m.map_string_msg["b"].foo == 2 + + m.map_string_int32["c"] = 3 + assert m.map_string_int32["c"] == 3 + m.map_string_msg["c"] = TestMessage2.new(:foo => 3) + assert m.map_string_msg["c"] == TestMessage2.new(:foo => 3) + m.map_string_msg.delete("b") + m.map_string_msg.delete("c") + assert m.map_string_msg == { "a" => TestMessage2.new(:foo => 1) } + + assert_raise TypeError do + m.map_string_msg["e"] = TestMessage.new # wrong value type + end + # ensure nothing was added by the above + assert m.map_string_msg == { "a" => TestMessage2.new(:foo => 1) } + + m.map_string_int32 = Google::Protobuf::Map.new(:string, :int32) + assert_raise TypeError do + m.map_string_int32 = Google::Protobuf::Map.new(:string, :int64) + end + assert_raise TypeError do + m.map_string_int32 = {} + end + + assert_raise TypeError do + m = MapMessage.new(:map_string_int32 => { 1 => "I am not a number" }) + end + end + + def test_map_encode_decode + m = MapMessage.new( + :map_string_int32 => {"a" => 1, "b" => 2}, + :map_string_msg => {"a" => TestMessage2.new(:foo => 1), + "b" => TestMessage2.new(:foo => 2)}) + m2 = MapMessage.decode(MapMessage.encode(m)) + assert m == m2 + + m3 = MapMessageWireEquiv.decode(MapMessage.encode(m)) + assert m3.map_string_int32.length == 2 + + kv = {} + m3.map_string_int32.map { |msg| kv[msg.key] = msg.value } + assert kv == {"a" => 1, "b" => 2} + + kv = {} + m3.map_string_msg.map { |msg| kv[msg.key] = msg.value } + assert kv == {"a" => TestMessage2.new(:foo => 1), + "b" => TestMessage2.new(:foo => 2)} + end + + def test_oneof_descriptors + d = OneofMessage.descriptor + o = d.lookup_oneof("my_oneof") + assert o != nil + assert o.class == Google::Protobuf::OneofDescriptor + assert o.name == "my_oneof" + oneof_count = 0 + d.each_oneof{ |oneof| + oneof_count += 1 + assert oneof == o + } + assert oneof_count == 1 + assert o.count == 4 + field_names = o.map{|f| f.name}.sort + assert field_names == ["a", "b", "c", "d"] + end + + def test_oneof + d = OneofMessage.new + assert d.a == nil + assert d.b == nil + assert d.c == nil + assert d.d == nil + assert d.my_oneof == nil + + d.a = "hi" + assert d.a == "hi" + assert d.b == nil + assert d.c == nil + assert d.d == nil + assert d.my_oneof == :a + + d.b = 42 + assert d.a == nil + assert d.b == 42 + assert d.c == nil + assert d.d == nil + assert d.my_oneof == :b + + d.c = TestMessage2.new(:foo => 100) + assert d.a == nil + assert d.b == nil + assert d.c.foo == 100 + assert d.d == nil + assert d.my_oneof == :c + + d.d = :C + assert d.a == nil + assert d.b == nil + assert d.c == nil + assert d.d == :C + assert d.my_oneof == :d + + d2 = OneofMessage.decode(OneofMessage.encode(d)) + assert d2 == d + + encoded_field_a = OneofMessage.encode(OneofMessage.new(:a => "string")) + encoded_field_b = OneofMessage.encode(OneofMessage.new(:b => 1000)) + encoded_field_c = OneofMessage.encode( + OneofMessage.new(:c => TestMessage2.new(:foo => 1))) + encoded_field_d = OneofMessage.encode(OneofMessage.new(:d => :B)) + + d3 = OneofMessage.decode( + encoded_field_c + encoded_field_a + encoded_field_d) + assert d3.a == nil + assert d3.b == nil + assert d3.c == nil + assert d3.d == :B + + d4 = OneofMessage.decode( + encoded_field_c + encoded_field_a + encoded_field_d + + encoded_field_c) + assert d4.a == nil + assert d4.b == nil + assert d4.c.foo == 1 + assert d4.d == nil + + d5 = OneofMessage.new(:a => "hello") + assert d5.a != nil + d5.a = nil + assert d5.a == nil + assert OneofMessage.encode(d5) == '' + assert d5.my_oneof == nil + end + def test_enum_field m = TestMessage.new assert m.optional_enum == :Default m.optional_enum = :A assert m.optional_enum == :A - assert_raise NameError do + assert_raise RangeError do m.optional_enum = :ASDF end m.optional_enum = 1 @@ -363,6 +717,14 @@ module BasicTest assert m.repeated_msg[0].object_id != m2.repeated_msg[0].object_id end + def test_eq + m = TestMessage.new(:optional_int32 => 42, + :repeated_int32 => [1, 2, 3]) + m2 = TestMessage.new(:optional_int32 => 43, + :repeated_int32 => [1, 2, 3]) + assert m != m2 + end + def test_enum_lookup assert TestEnum::A == 1 assert TestEnum::B == 2 @@ -384,7 +746,8 @@ module BasicTest :repeated_string => ["a", "b", "c"], :repeated_int32 => [42, 43, 44], :repeated_enum => [:A, :B, :C, 100], - :repeated_msg => [TestMessage2.new(:foo => 1), TestMessage2.new(:foo => 2)]) + :repeated_msg => [TestMessage2.new(:foo => 1), + TestMessage2.new(:foo => 2)]) data = TestMessage.encode m m2 = TestMessage.decode data assert m == m2 @@ -629,5 +992,13 @@ module BasicTest m2 = TestMessage.decode_json(json_text) assert m == m2 end + + def test_json_maps + m = MapMessage.new(:map_string_int32 => {"a" => 1}) + expected = '{"map_string_int32":{"a":1},"map_string_msg":{}}' + assert MapMessage.encode_json(m) == expected + m2 = MapMessage.decode_json(MapMessage.encode_json(m)) + assert m == m2 + end end end diff --git a/ruby/tests/generated_code.proto b/ruby/tests/generated_code.proto new file mode 100644 index 0000000000..b1d6323243 --- /dev/null +++ b/ruby/tests/generated_code.proto @@ -0,0 +1,67 @@ +syntax = "proto3"; + +package A.B.C; + +message TestMessage { + optional int32 optional_int32 = 1; + optional int64 optional_int64 = 2; + optional uint32 optional_uint32 = 3; + optional uint64 optional_uint64 = 4; + optional bool optional_bool = 5; + optional double optional_double = 6; + optional float optional_float = 7; + optional string optional_string = 8; + optional bytes optional_bytes = 9; + optional TestEnum optional_enum = 10; + optional TestMessage optional_msg = 11; + + repeated int32 repeated_int32 = 21; + repeated int64 repeated_int64 = 22; + repeated uint32 repeated_uint32 = 23; + repeated uint64 repeated_uint64 = 24; + repeated bool repeated_bool = 25; + repeated double repeated_double = 26; + repeated float repeated_float = 27; + repeated string repeated_string = 28; + repeated bytes repeated_bytes = 29; + repeated TestEnum repeated_enum = 30; + repeated TestMessage repeated_msg = 31; + + oneof my_oneof { + int32 oneof_int32 = 41; + int64 oneof_int64 = 42; + uint32 oneof_uint32 = 43; + uint64 oneof_uint64 = 44; + bool oneof_bool = 45; + double oneof_double = 46; + float oneof_float = 47; + string oneof_string = 48; + bytes oneof_bytes = 49; + TestEnum oneof_enum = 50; + TestMessage oneof_msg = 51; + } + + map map_int32_string = 61; + map map_int64_string = 62; + map map_uint32_string = 63; + map map_uint64_string = 64; + map map_bool_string = 65; + map map_string_string = 66; + map map_string_msg = 67; + map map_string_enum = 68; + map map_string_int32 = 69; + map map_string_bool = 70; + + message NestedMessage { + optional int32 foo = 1; + } + + optional NestedMessage nested_message = 80; +} + +enum TestEnum { + Default = 0; + A = 1; + B = 2; + C = 3; +} diff --git a/ruby/tests/generated_code.rb b/ruby/tests/generated_code.rb new file mode 100644 index 0000000000..5a6854331f --- /dev/null +++ b/ruby/tests/generated_code.rb @@ -0,0 +1,74 @@ +# Generated by the protocol buffer compiler. DO NOT EDIT! +# source: generated_code.proto + +require 'google/protobuf' + +Google::Protobuf::DescriptorPool.generated_pool.build do + add_message "A.B.C.TestMessage" do + optional :optional_int32, :int32, 1 + optional :optional_int64, :int64, 2 + optional :optional_uint32, :uint32, 3 + optional :optional_uint64, :uint64, 4 + optional :optional_bool, :bool, 5 + optional :optional_double, :double, 6 + optional :optional_float, :float, 7 + optional :optional_string, :string, 8 + optional :optional_bytes, :string, 9 + optional :optional_enum, :enum, 10, "A.B.C.TestEnum" + optional :optional_msg, :message, 11, "A.B.C.TestMessage" + repeated :repeated_int32, :int32, 21 + repeated :repeated_int64, :int64, 22 + repeated :repeated_uint32, :uint32, 23 + repeated :repeated_uint64, :uint64, 24 + repeated :repeated_bool, :bool, 25 + repeated :repeated_double, :double, 26 + repeated :repeated_float, :float, 27 + repeated :repeated_string, :string, 28 + repeated :repeated_bytes, :string, 29 + repeated :repeated_enum, :enum, 30, "A.B.C.TestEnum" + repeated :repeated_msg, :message, 31, "A.B.C.TestMessage" + map :map_int32_string, :int32, :string, 61 + map :map_int64_string, :int64, :string, 62 + map :map_uint32_string, :uint32, :string, 63 + map :map_uint64_string, :uint64, :string, 64 + map :map_bool_string, :bool, :string, 65 + map :map_string_string, :string, :string, 66 + map :map_string_msg, :string, :message, 67, "A.B.C.TestMessage" + map :map_string_enum, :string, :enum, 68, "A.B.C.TestEnum" + map :map_string_int32, :string, :int32, 69 + map :map_string_bool, :string, :bool, 70 + optional :nested_message, :message, 80, "A.B.C.TestMessage.NestedMessage" + oneof :my_oneof do + optional :oneof_int32, :int32, 41 + optional :oneof_int64, :int64, 42 + optional :oneof_uint32, :uint32, 43 + optional :oneof_uint64, :uint64, 44 + optional :oneof_bool, :bool, 45 + optional :oneof_double, :double, 46 + optional :oneof_float, :float, 47 + optional :oneof_string, :string, 48 + optional :oneof_bytes, :string, 49 + optional :oneof_enum, :enum, 50, "A.B.C.TestEnum" + optional :oneof_msg, :message, 51, "A.B.C.TestMessage" + end + end + add_message "A.B.C.TestMessage.NestedMessage" do + optional :foo, :int32, 1 + end + add_enum "A.B.C.TestEnum" do + value :Default, 0 + value :A, 1 + value :B, 2 + value :C, 3 + end +end + +module A + module B + module C + TestMessage = Google::Protobuf::DescriptorPool.generated_pool.lookup("A.B.C.TestMessage").msgclass + TestMessage::NestedMessage = Google::Protobuf::DescriptorPool.generated_pool.lookup("A.B.C.TestMessage.NestedMessage").msgclass + TestEnum = Google::Protobuf::DescriptorPool.generated_pool.lookup("A.B.C.TestEnum").enummodule + end + end +end diff --git a/ruby/tests/generated_code_test.rb b/ruby/tests/generated_code_test.rb new file mode 100644 index 0000000000..daef357a10 --- /dev/null +++ b/ruby/tests/generated_code_test.rb @@ -0,0 +1,17 @@ +#!/usr/bin/ruby + +# generated_code.rb is in the same directory as this test. +$LOAD_PATH.unshift(File.expand_path(File.dirname(__FILE__))) + +require 'generated_code' +require 'test/unit' + +class GeneratedCodeTest < Test::Unit::TestCase + def test_generated_msg + # just test that we can instantiate the message. The purpose of this test + # is to ensure that the output of the code generator is valid Ruby and + # successfully creates message definitions and classes, not to test every + # aspect of the extension (basic.rb is for that). + m = A::B::C::TestMessage.new() + end +end diff --git a/ruby/tests/stress.rb b/ruby/tests/stress.rb index cc4bbf5cf4..1bd768ed7b 100644 --- a/ruby/tests/stress.rb +++ b/ruby/tests/stress.rb @@ -1,6 +1,6 @@ #!/usr/bin/ruby -require 'protobuf' +require 'google/protobuf' require 'test/unit' module StressTest diff --git a/src/Makefile.am b/src/Makefile.am index e1a2e16942..0f0bd089d0 100644 --- a/src/Makefile.am +++ b/src/Makefile.am @@ -107,7 +107,9 @@ nobase_include_HEADERS = \ google/protobuf/compiler/cpp/cpp_generator.h \ google/protobuf/compiler/java/java_generator.h \ google/protobuf/compiler/java/java_names.h \ - google/protobuf/compiler/python/python_generator.h + google/protobuf/compiler/javanano/javanano_generator.h \ + google/protobuf/compiler/python/python_generator.h \ + google/protobuf/compiler/ruby/ruby_generator.h lib_LTLIBRARIES = libprotobuf-lite.la libprotobuf.la libprotoc.la @@ -241,26 +243,28 @@ libprotoc_la_SOURCES = \ google/protobuf/compiler/java/java_doc_comment.cc \ google/protobuf/compiler/java/java_doc_comment.h \ google/protobuf/compiler/javanano/javanano_enum.cc \ + google/protobuf/compiler/javanano/javanano_enum.h \ + google/protobuf/compiler/javanano/javanano_enum_field.cc \ google/protobuf/compiler/javanano/javanano_enum_field.h \ google/protobuf/compiler/javanano/javanano_extension.cc \ + google/protobuf/compiler/javanano/javanano_extension.h \ google/protobuf/compiler/javanano/javanano_field.cc \ + google/protobuf/compiler/javanano/javanano_field.h \ google/protobuf/compiler/javanano/javanano_file.cc \ + google/protobuf/compiler/javanano/javanano_file.h \ google/protobuf/compiler/javanano/javanano_generator.cc \ + google/protobuf/compiler/javanano/javanano_generator.h \ google/protobuf/compiler/javanano/javanano_helpers.cc \ + google/protobuf/compiler/javanano/javanano_helpers.h \ + google/protobuf/compiler/javanano/javanano_map_field.cc \ + google/protobuf/compiler/javanano/javanano_map_field.h \ google/protobuf/compiler/javanano/javanano_message.cc \ + google/protobuf/compiler/javanano/javanano_message.h \ + google/protobuf/compiler/javanano/javanano_message_field.cc \ google/protobuf/compiler/javanano/javanano_message_field.h \ google/protobuf/compiler/javanano/javanano_params.h \ - google/protobuf/compiler/javanano/javanano_primitive_field.h \ - google/protobuf/compiler/javanano/javanano_enum_field.cc \ - google/protobuf/compiler/javanano/javanano_enum.h \ - google/protobuf/compiler/javanano/javanano_extension.h \ - google/protobuf/compiler/javanano/javanano_field.h \ - google/protobuf/compiler/javanano/javanano_file.h \ - google/protobuf/compiler/javanano/javanano_generator.h \ - google/protobuf/compiler/javanano/javanano_helpers.h \ - google/protobuf/compiler/javanano/javanano_message_field.cc \ - google/protobuf/compiler/javanano/javanano_message.h \ google/protobuf/compiler/javanano/javanano_primitive_field.cc \ + google/protobuf/compiler/javanano/javanano_primitive_field.h \ google/protobuf/compiler/python/python_generator.cc \ google/protobuf/compiler/ruby/ruby_generator.cc @@ -456,6 +460,7 @@ protobuf_test_SOURCES = \ google/protobuf/compiler/java/java_plugin_unittest.cc \ google/protobuf/compiler/java/java_doc_comment_unittest.cc \ google/protobuf/compiler/python/python_plugin_unittest.cc \ + google/protobuf/compiler/ruby/ruby_generator_unittest.cc \ $(COMMON_TEST_SOURCES) nodist_protobuf_test_SOURCES = $(protoc_outputs) diff --git a/src/google/protobuf/compiler/cpp/cpp_service.cc b/src/google/protobuf/compiler/cpp/cpp_service.cc index a8f303daea..226c2aa03b 100644 --- a/src/google/protobuf/compiler/cpp/cpp_service.cc +++ b/src/google/protobuf/compiler/cpp/cpp_service.cc @@ -301,7 +301,7 @@ void ServiceGenerator::GenerateGetPrototype(RequestOrResponse which, printer->Print(vars_, " default:\n" " GOOGLE_LOG(FATAL) << \"Bad method index; this should never happen.\";\n" - " return *reinterpret_cast< ::google::protobuf::Message*>(NULL);\n" + " return *static_cast< ::google::protobuf::Message*>(NULL);\n" " }\n" "}\n" "\n"); diff --git a/src/google/protobuf/compiler/java/java_names.h b/src/google/protobuf/compiler/java/java_names.h new file mode 100644 index 0000000000..0d6143353a --- /dev/null +++ b/src/google/protobuf/compiler/java/java_names.h @@ -0,0 +1,87 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2008 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +// Author: kenton@google.com (Kenton Varda) +// Based on original Protocol Buffers design by +// Sanjay Ghemawat, Jeff Dean, and others. +// +// Provides a mechanism for mapping a descriptor to the +// fully-qualified name of the corresponding Java class. + +#ifndef GOOGLE_PROTOBUF_COMPILER_JAVA_NAMES_H__ +#define GOOGLE_PROTOBUF_COMPILER_JAVA_NAMES_H__ + +#include + +namespace google { +namespace protobuf { + +class Descriptor; +class EnumDescriptor; +class FileDescriptor; +class ServiceDescriptor; + +namespace compiler { +namespace java { + +// Requires: +// descriptor != NULL +// +// Returns: +// The fully-qualified Java class name. +string ClassName(const Descriptor* descriptor); + +// Requires: +// descriptor != NULL +// +// Returns: +// The fully-qualified Java class name. +string ClassName(const EnumDescriptor* descriptor); + +// Requires: +// descriptor != NULL +// +// Returns: +// The fully-qualified Java class name. +string ClassName(const FileDescriptor* descriptor); + +// Requires: +// descriptor != NULL +// +// Returns: +// The fully-qualified Java class name. +string ClassName(const ServiceDescriptor* descriptor); + +} // namespace java +} // namespace compiler +} // namespace protobuf +} // namespace google + +#endif // GOOGLE_PROTOBUF_COMPILER_JAVA_NAMES_H__ diff --git a/src/google/protobuf/compiler/javanano/javanano_field.cc b/src/google/protobuf/compiler/javanano/javanano_field.cc index e3e4cefee0..b9be90b15f 100644 --- a/src/google/protobuf/compiler/javanano/javanano_field.cc +++ b/src/google/protobuf/compiler/javanano/javanano_field.cc @@ -36,6 +36,7 @@ #include #include #include +#include #include #include @@ -97,7 +98,11 @@ FieldGenerator* FieldGeneratorMap::MakeGenerator(const FieldDescriptor* field, if (field->is_repeated()) { switch (java_type) { case JAVATYPE_MESSAGE: - return new RepeatedMessageFieldGenerator(field, params); + if (IsMapEntry(field->message_type())) { + return new MapFieldGenerator(field, params); + } else { + return new RepeatedMessageFieldGenerator(field, params); + } case JAVATYPE_ENUM: return new RepeatedEnumFieldGenerator(field, params); default: diff --git a/src/google/protobuf/compiler/javanano/javanano_helpers.cc b/src/google/protobuf/compiler/javanano/javanano_helpers.cc index 2149418af6..7b23d9f43d 100644 --- a/src/google/protobuf/compiler/javanano/javanano_helpers.cc +++ b/src/google/protobuf/compiler/javanano/javanano_helpers.cc @@ -560,6 +560,17 @@ void SetBitOperationVariables(const string name, (*variables)["different_" + name] = GenerateDifferentBit(bitIndex); } +bool HasMapField(const Descriptor* descriptor) { + for (int i = 0; i < descriptor->field_count(); ++i) { + const FieldDescriptor* field = descriptor->field(i); + if (field->type() == FieldDescriptor::TYPE_MESSAGE && + IsMapEntry(field->message_type())) { + return true; + } + } + return false; +} + } // namespace javanano } // namespace compiler } // namespace protobuf diff --git a/src/google/protobuf/compiler/javanano/javanano_helpers.h b/src/google/protobuf/compiler/javanano/javanano_helpers.h index 2931074365..b4fee649a3 100644 --- a/src/google/protobuf/compiler/javanano/javanano_helpers.h +++ b/src/google/protobuf/compiler/javanano/javanano_helpers.h @@ -181,6 +181,14 @@ string GenerateDifferentBit(int bit_index); void SetBitOperationVariables(const string name, int bitIndex, map* variables); +inline bool IsMapEntry(const Descriptor* descriptor) { + // TODO(liujisi): Add an option to turn on maps for proto2 syntax as well. + return descriptor->options().map_entry() && + descriptor->file()->syntax() == FileDescriptor::SYNTAX_PROTO3; +} + +bool HasMapField(const Descriptor* descriptor); + } // namespace javanano } // namespace compiler } // namespace protobuf diff --git a/src/google/protobuf/compiler/javanano/javanano_map_field.cc b/src/google/protobuf/compiler/javanano/javanano_map_field.cc new file mode 100644 index 0000000000..83b2b0ce8f --- /dev/null +++ b/src/google/protobuf/compiler/javanano/javanano_map_field.cc @@ -0,0 +1,186 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2008 Google Inc. All rights reserved. +// http://code.google.com/p/protobuf/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include +#include +#include +#include +#include +#include + +namespace google { +namespace protobuf { +namespace compiler { +namespace javanano { + +namespace { + +string TypeName(const Params& params, const FieldDescriptor* field, + bool boxed) { + JavaType java_type = GetJavaType(field); + switch (java_type) { + case JAVATYPE_MESSAGE: + return ClassName(params, field->message_type()); + case JAVATYPE_INT: + case JAVATYPE_LONG: + case JAVATYPE_FLOAT: + case JAVATYPE_DOUBLE: + case JAVATYPE_BOOLEAN: + case JAVATYPE_STRING: + case JAVATYPE_BYTES: + case JAVATYPE_ENUM: + if (boxed) { + return BoxedPrimitiveTypeName(java_type); + } else { + return PrimitiveTypeName(java_type); + } + // No default because we want the compiler to complain if any new JavaTypes + // are added.. + } + + GOOGLE_LOG(FATAL) << "should not reach here."; + return ""; +} + +const FieldDescriptor* KeyField(const FieldDescriptor* descriptor) { + GOOGLE_CHECK_EQ(FieldDescriptor::TYPE_MESSAGE, descriptor->type()); + const Descriptor* message = descriptor->message_type(); + GOOGLE_CHECK(message->options().map_entry()); + return message->FindFieldByName("key"); +} + +const FieldDescriptor* ValueField(const FieldDescriptor* descriptor) { + GOOGLE_CHECK_EQ(FieldDescriptor::TYPE_MESSAGE, descriptor->type()); + const Descriptor* message = descriptor->message_type(); + GOOGLE_CHECK(message->options().map_entry()); + return message->FindFieldByName("value"); +} + +void SetMapVariables(const Params& params, + const FieldDescriptor* descriptor, map* variables) { + const FieldDescriptor* key = KeyField(descriptor); + const FieldDescriptor* value = ValueField(descriptor); + (*variables)["name"] = + RenameJavaKeywords(UnderscoresToCamelCase(descriptor)); + (*variables)["number"] = SimpleItoa(descriptor->number()); + (*variables)["key_type"] = TypeName(params, key, false); + (*variables)["boxed_key_type"] = TypeName(params,key, true); + (*variables)["key_desc_type"] = + "TYPE_" + ToUpper(FieldDescriptor::TypeName(key->type())); + (*variables)["key_tag"] = SimpleItoa(internal::WireFormat::MakeTag(key)); + (*variables)["value_type"] = TypeName(params, value, false); + (*variables)["boxed_value_type"] = TypeName(params, value, true); + (*variables)["value_desc_type"] = + "TYPE_" + ToUpper(FieldDescriptor::TypeName(value->type())); + (*variables)["value_tag"] = SimpleItoa(internal::WireFormat::MakeTag(value)); + (*variables)["type_parameters"] = + (*variables)["boxed_key_type"] + ", " + (*variables)["boxed_value_type"]; + (*variables)["value_default"] = + value->type() == FieldDescriptor::TYPE_MESSAGE + ? "new " + (*variables)["value_type"] + "()" + : "null"; +} +} // namespace + +// =================================================================== +MapFieldGenerator::MapFieldGenerator(const FieldDescriptor* descriptor, + const Params& params) + : FieldGenerator(params), descriptor_(descriptor) { + SetMapVariables(params, descriptor, &variables_); +} + +MapFieldGenerator::~MapFieldGenerator() {} + +void MapFieldGenerator:: +GenerateMembers(io::Printer* printer, bool /* unused lazy_init */) const { + printer->Print(variables_, + "public java.util.Map<$type_parameters$> $name$;\n"); +} + +void MapFieldGenerator:: +GenerateClearCode(io::Printer* printer) const { + printer->Print(variables_, + "$name$ = null;\n"); +} + +void MapFieldGenerator:: +GenerateMergingCode(io::Printer* printer) const { + printer->Print(variables_, + "this.$name$ = com.google.protobuf.nano.InternalNano.mergeMapEntry(\n" + " input, this.$name$, mapFactory,\n" + " com.google.protobuf.nano.InternalNano.$key_desc_type$,\n" + " com.google.protobuf.nano.InternalNano.$value_desc_type$,\n" + " $value_default$,\n" + " $key_tag$, $value_tag$);\n" + "\n"); +} + +void MapFieldGenerator:: +GenerateSerializationCode(io::Printer* printer) const { + printer->Print(variables_, + "if (this.$name$ != null) {\n" + " com.google.protobuf.nano.InternalNano.serializeMapField(\n" + " output, this.$name$, $number$,\n" + " com.google.protobuf.nano.InternalNano.$key_desc_type$,\n" + " com.google.protobuf.nano.InternalNano.$value_desc_type$);\n" + "}\n"); +} + +void MapFieldGenerator:: +GenerateSerializedSizeCode(io::Printer* printer) const { + printer->Print(variables_, + "if (this.$name$ != null) {\n" + " size += com.google.protobuf.nano.InternalNano.computeMapFieldSize(\n" + " this.$name$, $number$,\n" + " com.google.protobuf.nano.InternalNano.$key_desc_type$,\n" + " com.google.protobuf.nano.InternalNano.$value_desc_type$);\n" + "}\n"); +} + +void MapFieldGenerator:: +GenerateEqualsCode(io::Printer* printer) const { + printer->Print(variables_, + "if (!com.google.protobuf.nano.InternalNano.equals(\n" + " this.$name$, other.$name$)) {\n" + " return false;\n" + "}\n"); +} + +void MapFieldGenerator:: +GenerateHashCodeCode(io::Printer* printer) const { + printer->Print(variables_, + "result = 31 * result +\n" + " com.google.protobuf.nano.InternalNano.hashCode(this.$name$);\n"); +} + +} // namespace javanano +} // namespace compiler +} // namespace protobuf +} // namespace google diff --git a/src/google/protobuf/compiler/javanano/javanano_map_field.h b/src/google/protobuf/compiler/javanano/javanano_map_field.h new file mode 100644 index 0000000000..c01bde3815 --- /dev/null +++ b/src/google/protobuf/compiler/javanano/javanano_map_field.h @@ -0,0 +1,70 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2008 Google Inc. All rights reserved. +// http://code.google.com/p/protobuf/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef GOOGLE_PROTOBUF_COMPILER_JAVANANO_MAP_FIELD_H__ +#define GOOGLE_PROTOBUF_COMPILER_JAVANANO_MAP_FIELD_H__ + +#include +#include +#include +#include + +namespace google { +namespace protobuf { +namespace compiler { +namespace javanano { + +class MapFieldGenerator : public FieldGenerator { + public: + explicit MapFieldGenerator( + const FieldDescriptor* descriptor, const Params& params); + ~MapFieldGenerator(); + + // implements FieldGenerator --------------------------------------- + void GenerateMembers(io::Printer* printer, bool lazy_init) const; + void GenerateClearCode(io::Printer* printer) const; + void GenerateMergingCode(io::Printer* printer) const; + void GenerateSerializationCode(io::Printer* printer) const; + void GenerateSerializedSizeCode(io::Printer* printer) const; + void GenerateEqualsCode(io::Printer* printer) const; + void GenerateHashCodeCode(io::Printer* printer) const; + + private: + const FieldDescriptor* descriptor_; + map variables_; + + GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(MapFieldGenerator); +}; + +} // namespace javanano +} // namespace compiler +} // namespace protobuf +} // namespace google +#endif // GOOGLE_PROTOBUF_COMPILER_JAVANANO_MAP_FIELD_H__ diff --git a/src/google/protobuf/compiler/javanano/javanano_message.cc b/src/google/protobuf/compiler/javanano/javanano_message.cc index 7c52ca31bc..da7c656f38 100644 --- a/src/google/protobuf/compiler/javanano/javanano_message.cc +++ b/src/google/protobuf/compiler/javanano/javanano_message.cc @@ -90,6 +90,7 @@ void MessageGenerator::GenerateStaticVariables(io::Printer* printer) { // Generate static members for all nested types. for (int i = 0; i < descriptor_->nested_type_count(); i++) { // TODO(kenton): Reuse MessageGenerator objects? + if (IsMapEntry(descriptor_->nested_type(i))) continue; MessageGenerator(descriptor_->nested_type(i), params_) .GenerateStaticVariables(printer); } @@ -100,6 +101,7 @@ void MessageGenerator::GenerateStaticVariableInitializers( // Generate static member initializers for all nested types. for (int i = 0; i < descriptor_->nested_type_count(); i++) { // TODO(kenton): Reuse MessageGenerator objects? + if (IsMapEntry(descriptor_->nested_type(i))) continue; MessageGenerator(descriptor_->nested_type(i), params_) .GenerateStaticVariableInitializers(printer); } @@ -159,6 +161,7 @@ void MessageGenerator::Generate(io::Printer* printer) { } for (int i = 0; i < descriptor_->nested_type_count(); i++) { + if (IsMapEntry(descriptor_->nested_type(i))) continue; MessageGenerator(descriptor_->nested_type(i), params_).Generate(printer); } @@ -342,6 +345,11 @@ void MessageGenerator::GenerateMergeFromMethods(io::Printer* printer) { "classname", descriptor_->name()); printer->Indent(); + if (HasMapField(descriptor_)) { + printer->Print( + "com.google.protobuf.nano.MapFactories.MapFactory mapFactory =\n" + " com.google.protobuf.nano.MapFactories.getMapFactory();\n"); + } printer->Print( "while (true) {\n"); diff --git a/src/google/protobuf/compiler/ruby/ruby_generator.cc b/src/google/protobuf/compiler/ruby/ruby_generator.cc index 6fed1d462f..a9b6837e9b 100644 --- a/src/google/protobuf/compiler/ruby/ruby_generator.cc +++ b/src/google/protobuf/compiler/ruby/ruby_generator.cc @@ -100,15 +100,35 @@ std::string TypeName(const google::protobuf::FieldDescriptor* field) { } } -void GenerateMessage(const google::protobuf::Descriptor* message, - google::protobuf::io::Printer* printer) { - printer->Print( - "add_message \"$name$\" do\n", - "name", message->full_name()); - printer->Indent(); +void GenerateField(const google::protobuf::FieldDescriptor* field, + google::protobuf::io::Printer* printer) { + + if (field->is_map()) { + const FieldDescriptor* key_field = + field->message_type()->FindFieldByNumber(1); + const FieldDescriptor* value_field = + field->message_type()->FindFieldByNumber(2); + + printer->Print( + "map :$name$, :$key_type$, :$value_type$, $number$", + "name", field->name(), + "key_type", TypeName(key_field), + "value_type", TypeName(value_field), + "number", IntToString(field->number())); + + if (value_field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) { + printer->Print( + ", \"$subtype$\"\n", + "subtype", value_field->message_type()->full_name()); + } else if (value_field->cpp_type() == FieldDescriptor::CPPTYPE_ENUM) { + printer->Print( + ", \"$subtype$\"\n", + "subtype", value_field->enum_type()->full_name()); + } else { + printer->Print("\n"); + } + } else { - for (int i = 0; i < message->field_count(); i++) { - const FieldDescriptor* field = message->field(i); printer->Print( "$label$ :$name$, ", "label", LabelForField(field), @@ -117,6 +137,7 @@ void GenerateMessage(const google::protobuf::Descriptor* message, ":$type$, $number$", "type", TypeName(field), "number", IntToString(field->number())); + if (field->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE) { printer->Print( ", \"$subtype$\"\n", @@ -129,6 +150,49 @@ void GenerateMessage(const google::protobuf::Descriptor* message, printer->Print("\n"); } } +} + +void GenerateOneof(const google::protobuf::OneofDescriptor* oneof, + google::protobuf::io::Printer* printer) { + printer->Print( + "oneof :$name$ do\n", + "name", oneof->name()); + printer->Indent(); + + for (int i = 0; i < oneof->field_count(); i++) { + const FieldDescriptor* field = oneof->field(i); + GenerateField(field, printer); + } + + printer->Outdent(); + printer->Print("end\n"); +} + +void GenerateMessage(const google::protobuf::Descriptor* message, + google::protobuf::io::Printer* printer) { + + // Don't generate MapEntry messages -- we use the Ruby extension's native + // support for map fields instead. + if (message->options().map_entry()) { + return; + } + + printer->Print( + "add_message \"$name$\" do\n", + "name", message->full_name()); + printer->Indent(); + + for (int i = 0; i < message->field_count(); i++) { + const FieldDescriptor* field = message->field(i); + if (!field->containing_oneof()) { + GenerateField(field, printer); + } + } + + for (int i = 0; i < message->oneof_decl_count(); i++) { + const OneofDescriptor* oneof = message->oneof_decl(i); + GenerateOneof(oneof, printer); + } printer->Outdent(); printer->Print("end\n"); @@ -185,6 +249,13 @@ void GenerateMessageAssignment( const std::string& prefix, const google::protobuf::Descriptor* message, google::protobuf::io::Printer* printer) { + + // Don't generate MapEntry messages -- we use the Ruby extension's native + // support for map fields instead. + if (message->options().map_entry()) { + return; + } + printer->Print( "$prefix$$name$ = ", "prefix", prefix, @@ -261,7 +332,7 @@ void GenerateFile(const google::protobuf::FileDescriptor* file, "filename", file->name()); printer->Print( - "require 'protobuf'\n\n"); + "require 'google/protobuf'\n\n"); for (int i = 0; i < file->dependency_count(); i++) { const std::string& name = file->dependency(i)->name(); @@ -297,6 +368,14 @@ bool Generator::Generate( const string& parameter, GeneratorContext* generator_context, string* error) const { + + if (file->syntax() != FileDescriptor::SYNTAX_PROTO3) { + *error = + "Can only generate Ruby code for proto3 .proto files.\n" + "Please add 'syntax = \"proto3\";' to the top of your .proto file.\n"; + return false; + } + std::string filename = StripDotProto(file->name()) + ".rb"; scoped_ptr output( diff --git a/src/google/protobuf/compiler/ruby/ruby_generator.h b/src/google/protobuf/compiler/ruby/ruby_generator.h index 48dbefd125..75555c31b5 100644 --- a/src/google/protobuf/compiler/ruby/ruby_generator.h +++ b/src/google/protobuf/compiler/ruby/ruby_generator.h @@ -40,7 +40,8 @@ namespace protobuf { namespace compiler { namespace ruby { -class Generator : public google::protobuf::compiler::CodeGenerator { +class LIBPROTOC_EXPORT Generator + : public google::protobuf::compiler::CodeGenerator { virtual bool Generate( const FileDescriptor* file, const string& parameter, diff --git a/src/google/protobuf/compiler/ruby/ruby_generator_unittest.cc b/src/google/protobuf/compiler/ruby/ruby_generator_unittest.cc new file mode 100644 index 0000000000..e35ca695b5 --- /dev/null +++ b/src/google/protobuf/compiler/ruby/ruby_generator_unittest.cc @@ -0,0 +1,119 @@ +// Protocol Buffers - Google's data interchange format +// Copyright 2014 Google Inc. All rights reserved. +// https://developers.google.com/protocol-buffers/ +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following disclaimer +// in the documentation and/or other materials provided with the +// distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived from +// this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#include + +#include +#include +#include +#include + +#include +#include +#include + +namespace google { +namespace protobuf { +namespace compiler { +namespace ruby { +namespace { + +string FindRubyTestDir() { + // Inspired by TestSourceDir() in src/google/protobuf/testing/googletest.cc. + string prefix = "."; + while (!File::Exists(prefix + "/ruby/tests")) { + if (!File::Exists(prefix)) { + GOOGLE_LOG(FATAL) + << "Could not find Ruby test directory. Please run tests from " + "somewhere within the protobuf source package."; + } + prefix += "/.."; + } + return prefix + "/ruby/tests"; +} + +// This test is a simple golden-file test over the output of the Ruby code +// generator. When we make changes to the Ruby extension and alter the Ruby code +// generator to use those changes, we should (i) manually test the output of the +// code generator with the extension, and (ii) update the golden output above. +// Some day, we may integrate build systems between protoc and the language +// extensions to the point where we can do this test in a more automated way. + +TEST(RubyGeneratorTest, GeneratorTest) { + string ruby_tests = FindRubyTestDir(); + + google::protobuf::compiler::CommandLineInterface cli; + cli.SetInputsAreProtoPathRelative(true); + + ruby::Generator ruby_generator; + cli.RegisterGenerator("--ruby_out", &ruby_generator, ""); + + // Copy generated_code.proto to the temporary test directory. + string test_input; + GOOGLE_CHECK_OK(File::GetContents( + ruby_tests + "/generated_code.proto", + &test_input, + true)); + GOOGLE_CHECK_OK(File::SetContents( + TestTempDir() + "/generated_code.proto", + test_input, + true)); + + // Invoke the proto compiler (we will be inside TestTempDir() at this point). + string ruby_out = "--ruby_out=" + TestTempDir(); + string proto_path = "--proto_path=" + TestTempDir(); + const char* argv[] = { + "protoc", + ruby_out.c_str(), + proto_path.c_str(), + "generated_code.proto", + }; + + EXPECT_EQ(0, cli.Run(4, argv)); + + // Load the generated output and compare to the expected result. + string output; + GOOGLE_CHECK_OK(File::GetContents( + TestTempDir() + "/generated_code.rb", + &output, + true)); + string expected_output; + GOOGLE_CHECK_OK(File::GetContents( + ruby_tests + "/generated_code.rb", + &expected_output, + true)); + EXPECT_EQ(expected_output, output); +} + +} // namespace +} // namespace ruby +} // namespace compiler +} // namespace protobuf +} // namespace google diff --git a/src/google/protobuf/io/coded_stream.h b/src/google/protobuf/io/coded_stream.h index b9c30fa3cb..978cc19de1 100644 --- a/src/google/protobuf/io/coded_stream.h +++ b/src/google/protobuf/io/coded_stream.h @@ -123,7 +123,8 @@ #endif #else #include // __BYTE_ORDER - #if defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN && \ + #if ((defined(__LITTLE_ENDIAN__) && !defined(__BIG_ENDIAN__)) || \ + (defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN)) && \ !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST) #define PROTOBUF_LITTLE_ENDIAN 1 #endif diff --git a/src/google/protobuf/stubs/atomicops_internals_mips_gcc.h b/src/google/protobuf/stubs/atomicops_internals_mips_gcc.h index e3cd14cf80..f5837c9ec2 100644 --- a/src/google/protobuf/stubs/atomicops_internals_mips_gcc.h +++ b/src/google/protobuf/stubs/atomicops_internals_mips_gcc.h @@ -65,7 +65,7 @@ inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, "2:\n" ".set pop\n" : "=&r" (prev), "=m" (*ptr), "=&r" (tmp) - : "Ir" (old_value), "r" (new_value), "m" (*ptr) + : "r" (old_value), "r" (new_value), "m" (*ptr) : "memory"); return prev; } @@ -197,7 +197,7 @@ inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, "2:\n" ".set pop\n" : "=&r" (prev), "=m" (*ptr), "=&r" (tmp) - : "Ir" (old_value), "r" (new_value), "m" (*ptr) + : "r" (old_value), "r" (new_value), "m" (*ptr) : "memory"); return prev; } diff --git a/upb b/upb deleted file mode 160000 index 56913be6bb..0000000000 --- a/upb +++ /dev/null @@ -1 +0,0 @@ -Subproject commit 56913be6bb57f81dbbf7baf9cc9a0a2cd1a36493 diff --git a/vsprojects/libprotoc.vcproj b/vsprojects/libprotoc.vcproj index 45a5936f83..6b1e886445 100644 --- a/vsprojects/libprotoc.vcproj +++ b/vsprojects/libprotoc.vcproj @@ -223,6 +223,10 @@ RelativePath="..\src\google\protobuf\compiler\cpp\cpp_string_field.h" > + + @@ -375,6 +379,10 @@ RelativePath="..\src\google\protobuf\compiler\cpp\cpp_string_field.cc" > + +