Protocol Buffers - Google's data interchange format (grpc依赖) https://developers.google.com/protocol-buffers/
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

671 lines
22 KiB

// Protocol Buffers - Google's data interchange format
// Copyright 2023 Google LLC. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file or at
// https://developers.google.com/open-source/licenses/bsd
// We encode backwards, to avoid pre-computing lengths (one-pass encode).
#include "upb/wire/encode.h"
#include <setjmp.h>
#include <stdbool.h>
#include <stdint.h>
6 years ago
#include <string.h>
#include "upb/base/descriptor_constants.h"
#include "upb/base/internal/endian.h"
#include "upb/base/string_view.h"
#include "upb/hash/common.h"
#include "upb/hash/str_table.h"
#include "upb/mem/arena.h"
#include "upb/message/array.h"
#include "upb/message/internal/accessors.h"
#include "upb/message/internal/array.h"
#include "upb/message/internal/extension.h"
#include "upb/message/internal/map.h"
#include "upb/message/internal/map_entry.h"
#include "upb/message/internal/map_sorter.h"
#include "upb/message/internal/tagged_ptr.h"
#include "upb/message/map.h"
#include "upb/message/message.h"
#include "upb/message/tagged_ptr.h"
#include "upb/mini_table/extension.h"
#include "upb/mini_table/field.h"
#include "upb/mini_table/internal/field.h"
#include "upb/mini_table/internal/message.h"
#include "upb/mini_table/message.h"
#include "upb/mini_table/sub.h"
#include "upb/wire/internal/constants.h"
#include "upb/wire/types.h"
// Must be last.
#include "upb/port/def.inc"
#define UPB_PB_VARINT_MAX_LEN 10
UPB_NOINLINE
static size_t encode_varint64(uint64_t val, char* buf) {
size_t i = 0;
do {
uint8_t byte = val & 0x7fU;
val >>= 7;
if (val) byte |= 0x80U;
buf[i++] = byte;
} while (val);
return i;
}
static uint32_t encode_zz32(int32_t n) {
return ((uint32_t)n << 1) ^ (n >> 31);
}
static uint64_t encode_zz64(int64_t n) {
return ((uint64_t)n << 1) ^ (n >> 63);
}
typedef struct {
upb_EncodeStatus status;
jmp_buf err;
upb_Arena* arena;
char *buf, *ptr, *limit;
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
int options;
int depth;
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
_upb_mapsorter sorter;
} upb_encstate;
static size_t upb_roundup_pow2(size_t bytes) {
size_t ret = 128;
while (ret < bytes) {
ret *= 2;
}
return ret;
}
UPB_NORETURN static void encode_err(upb_encstate* e, upb_EncodeStatus s) {
UPB_ASSERT(s != kUpb_EncodeStatus_Ok);
e->status = s;
UPB_LONGJMP(e->err, 1);
}
UPB_NOINLINE
static void encode_growbuffer(upb_encstate* e, size_t bytes) {
size_t old_size = e->limit - e->buf;
size_t new_size = upb_roundup_pow2(bytes + (e->limit - e->ptr));
char* new_buf = upb_Arena_Realloc(e->arena, e->buf, old_size, new_size);
if (!new_buf) encode_err(e, kUpb_EncodeStatus_OutOfMemory);
// We want previous data at the end, realloc() put it at the beginning.
// TODO: This is somewhat inefficient since we are copying twice.
// Maybe create a realloc() that copies to the end of the new buffer?
if (old_size > 0) {
memmove(new_buf + new_size - old_size, e->buf, old_size);
}
e->ptr = new_buf + new_size - (e->limit - e->ptr);
e->limit = new_buf + new_size;
e->buf = new_buf;
e->ptr -= bytes;
}
/* Call to ensure that at least "bytes" bytes are available for writing at
* e->ptr. Returns false if the bytes could not be allocated. */
UPB_FORCEINLINE
void encode_reserve(upb_encstate* e, size_t bytes) {
if ((size_t)(e->ptr - e->buf) < bytes) {
encode_growbuffer(e, bytes);
return;
}
e->ptr -= bytes;
}
/* Writes the given bytes to the buffer, handling reserve/advance. */
static void encode_bytes(upb_encstate* e, const void* data, size_t len) {
if (len == 0) return; /* memcpy() with zero size is UB */
encode_reserve(e, len);
memcpy(e->ptr, data, len);
}
static void encode_fixed64(upb_encstate* e, uint64_t val) {
val = upb_BigEndian64(val);
encode_bytes(e, &val, sizeof(uint64_t));
}
static void encode_fixed32(upb_encstate* e, uint32_t val) {
val = upb_BigEndian32(val);
encode_bytes(e, &val, sizeof(uint32_t));
}
UPB_NOINLINE
static void encode_longvarint(upb_encstate* e, uint64_t val) {
size_t len;
char* start;
encode_reserve(e, UPB_PB_VARINT_MAX_LEN);
len = encode_varint64(val, e->ptr);
start = e->ptr + UPB_PB_VARINT_MAX_LEN - len;
memmove(start, e->ptr, len);
e->ptr = start;
}
UPB_FORCEINLINE
void encode_varint(upb_encstate* e, uint64_t val) {
if (val < 128 && e->ptr != e->buf) {
--e->ptr;
*e->ptr = val;
} else {
encode_longvarint(e, val);
}
}
static void encode_double(upb_encstate* e, double d) {
uint64_t u64;
UPB_ASSERT(sizeof(double) == sizeof(uint64_t));
memcpy(&u64, &d, sizeof(uint64_t));
encode_fixed64(e, u64);
}
static void encode_float(upb_encstate* e, float d) {
uint32_t u32;
UPB_ASSERT(sizeof(float) == sizeof(uint32_t));
memcpy(&u32, &d, sizeof(uint32_t));
encode_fixed32(e, u32);
}
static void encode_tag(upb_encstate* e, uint32_t field_number,
uint8_t wire_type) {
encode_varint(e, (field_number << 3) | wire_type);
}
static void encode_fixedarray(upb_encstate* e, const upb_Array* arr,
size_t elem_size, uint32_t tag) {
size_t bytes = upb_Array_Size(arr) * elem_size;
const char* data = upb_Array_DataPtr(arr);
const char* ptr = data + bytes - elem_size;
if (tag || !upb_IsLittleEndian()) {
while (true) {
if (elem_size == 4) {
uint32_t val;
memcpy(&val, ptr, sizeof(val));
val = upb_BigEndian32(val);
encode_bytes(e, &val, elem_size);
} else {
UPB_ASSERT(elem_size == 8);
uint64_t val;
memcpy(&val, ptr, sizeof(val));
val = upb_BigEndian64(val);
encode_bytes(e, &val, elem_size);
}
if (tag) encode_varint(e, tag);
if (ptr == data) break;
ptr -= elem_size;
}
} else {
encode_bytes(e, data, bytes);
}
}
static void encode_message(upb_encstate* e, const upb_Message* msg,
const upb_MiniTable* m, size_t* size);
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
static void encode_TaggedMessagePtr(upb_encstate* e,
upb_TaggedMessagePtr tagged,
const upb_MiniTable* m, size_t* size) {
if (upb_TaggedMessagePtr_IsEmpty(tagged)) {
m = UPB_PRIVATE(_upb_MiniTable_Empty)();
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
}
encode_message(e, UPB_PRIVATE(_upb_TaggedMessagePtr_GetMessage)(tagged), m,
size);
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
}
static void encode_scalar(upb_encstate* e, const void* _field_mem,
const upb_MiniTableSub* subs,
const upb_MiniTableField* f) {
const char* field_mem = _field_mem;
int wire_type;
#define CASE(ctype, type, wtype, encodeval) \
{ \
ctype val = *(ctype*)field_mem; \
encode_##type(e, encodeval); \
wire_type = wtype; \
break; \
}
switch (f->UPB_PRIVATE(descriptortype)) {
case kUpb_FieldType_Double:
CASE(double, double, kUpb_WireType_64Bit, val);
case kUpb_FieldType_Float:
CASE(float, float, kUpb_WireType_32Bit, val);
case kUpb_FieldType_Int64:
case kUpb_FieldType_UInt64:
CASE(uint64_t, varint, kUpb_WireType_Varint, val);
case kUpb_FieldType_UInt32:
CASE(uint32_t, varint, kUpb_WireType_Varint, val);
case kUpb_FieldType_Int32:
case kUpb_FieldType_Enum:
CASE(int32_t, varint, kUpb_WireType_Varint, (int64_t)val);
case kUpb_FieldType_SFixed64:
case kUpb_FieldType_Fixed64:
CASE(uint64_t, fixed64, kUpb_WireType_64Bit, val);
case kUpb_FieldType_Fixed32:
case kUpb_FieldType_SFixed32:
CASE(uint32_t, fixed32, kUpb_WireType_32Bit, val);
case kUpb_FieldType_Bool:
CASE(bool, varint, kUpb_WireType_Varint, val);
case kUpb_FieldType_SInt32:
CASE(int32_t, varint, kUpb_WireType_Varint, encode_zz32(val));
case kUpb_FieldType_SInt64:
CASE(int64_t, varint, kUpb_WireType_Varint, encode_zz64(val));
case kUpb_FieldType_String:
case kUpb_FieldType_Bytes: {
upb_StringView view = *(upb_StringView*)field_mem;
encode_bytes(e, view.data, view.size);
encode_varint(e, view.size);
wire_type = kUpb_WireType_Delimited;
break;
}
case kUpb_FieldType_Group: {
size_t size;
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
upb_TaggedMessagePtr submsg = *(upb_TaggedMessagePtr*)field_mem;
const upb_MiniTable* subm =
upb_MiniTableSub_Message(subs[f->UPB_PRIVATE(submsg_index)]);
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
if (submsg == 0) {
return;
}
if (--e->depth == 0) encode_err(e, kUpb_EncodeStatus_MaxDepthExceeded);
encode_tag(e, upb_MiniTableField_Number(f), kUpb_WireType_EndGroup);
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
encode_TaggedMessagePtr(e, submsg, subm, &size);
wire_type = kUpb_WireType_StartGroup;
e->depth++;
break;
}
case kUpb_FieldType_Message: {
size_t size;
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
upb_TaggedMessagePtr submsg = *(upb_TaggedMessagePtr*)field_mem;
const upb_MiniTable* subm =
upb_MiniTableSub_Message(subs[f->UPB_PRIVATE(submsg_index)]);
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
if (submsg == 0) {
return;
}
if (--e->depth == 0) encode_err(e, kUpb_EncodeStatus_MaxDepthExceeded);
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
encode_TaggedMessagePtr(e, submsg, subm, &size);
encode_varint(e, size);
wire_type = kUpb_WireType_Delimited;
e->depth++;
break;
}
default:
UPB_UNREACHABLE();
}
#undef CASE
encode_tag(e, upb_MiniTableField_Number(f), wire_type);
}
static void encode_array(upb_encstate* e, const upb_Message* msg,
const upb_MiniTableSub* subs,
const upb_MiniTableField* f) {
const upb_Array* arr = *UPB_PTR_AT(msg, f->UPB_PRIVATE(offset), upb_Array*);
bool packed = upb_MiniTableField_IsPacked(f);
size_t pre_len = e->limit - e->ptr;
if (arr == NULL || upb_Array_Size(arr) == 0) {
return;
}
#define VARINT_CASE(ctype, encode) \
{ \
const ctype* start = upb_Array_DataPtr(arr); \
const ctype* ptr = start + upb_Array_Size(arr); \
uint32_t tag = \
packed ? 0 : (f->UPB_PRIVATE(number) << 3) | kUpb_WireType_Varint; \
do { \
ptr--; \
encode_varint(e, encode); \
if (tag) encode_varint(e, tag); \
} while (ptr != start); \
} \
break;
#define TAG(wire_type) (packed ? 0 : (f->UPB_PRIVATE(number) << 3 | wire_type))
switch (f->UPB_PRIVATE(descriptortype)) {
case kUpb_FieldType_Double:
encode_fixedarray(e, arr, sizeof(double), TAG(kUpb_WireType_64Bit));
break;
case kUpb_FieldType_Float:
encode_fixedarray(e, arr, sizeof(float), TAG(kUpb_WireType_32Bit));
break;
case kUpb_FieldType_SFixed64:
case kUpb_FieldType_Fixed64:
encode_fixedarray(e, arr, sizeof(uint64_t), TAG(kUpb_WireType_64Bit));
break;
case kUpb_FieldType_Fixed32:
case kUpb_FieldType_SFixed32:
encode_fixedarray(e, arr, sizeof(uint32_t), TAG(kUpb_WireType_32Bit));
break;
case kUpb_FieldType_Int64:
case kUpb_FieldType_UInt64:
VARINT_CASE(uint64_t, *ptr);
case kUpb_FieldType_UInt32:
VARINT_CASE(uint32_t, *ptr);
case kUpb_FieldType_Int32:
case kUpb_FieldType_Enum:
VARINT_CASE(int32_t, (int64_t)*ptr);
case kUpb_FieldType_Bool:
VARINT_CASE(bool, *ptr);
case kUpb_FieldType_SInt32:
VARINT_CASE(int32_t, encode_zz32(*ptr));
case kUpb_FieldType_SInt64:
VARINT_CASE(int64_t, encode_zz64(*ptr));
case kUpb_FieldType_String:
case kUpb_FieldType_Bytes: {
const upb_StringView* start = upb_Array_DataPtr(arr);
const upb_StringView* ptr = start + upb_Array_Size(arr);
do {
ptr--;
encode_bytes(e, ptr->data, ptr->size);
encode_varint(e, ptr->size);
encode_tag(e, upb_MiniTableField_Number(f), kUpb_WireType_Delimited);
} while (ptr != start);
return;
}
case kUpb_FieldType_Group: {
const upb_TaggedMessagePtr* start = upb_Array_DataPtr(arr);
const upb_TaggedMessagePtr* ptr = start + upb_Array_Size(arr);
const upb_MiniTable* subm =
upb_MiniTableSub_Message(subs[f->UPB_PRIVATE(submsg_index)]);
if (--e->depth == 0) encode_err(e, kUpb_EncodeStatus_MaxDepthExceeded);
do {
size_t size;
ptr--;
encode_tag(e, upb_MiniTableField_Number(f), kUpb_WireType_EndGroup);
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
encode_TaggedMessagePtr(e, *ptr, subm, &size);
encode_tag(e, upb_MiniTableField_Number(f), kUpb_WireType_StartGroup);
} while (ptr != start);
e->depth++;
return;
}
case kUpb_FieldType_Message: {
const upb_TaggedMessagePtr* start = upb_Array_DataPtr(arr);
const upb_TaggedMessagePtr* ptr = start + upb_Array_Size(arr);
const upb_MiniTable* subm =
upb_MiniTableSub_Message(subs[f->UPB_PRIVATE(submsg_index)]);
if (--e->depth == 0) encode_err(e, kUpb_EncodeStatus_MaxDepthExceeded);
do {
size_t size;
ptr--;
Added a new dynamic tree shaking model to upb, with the intention of removing the old model once YouTube has migrated. The `kUpb_DecodeOption_ExperimentalAllowUnlinked` flag to the decoder will enable the new behavior. When that flag is not passed, tree shaking with the old model will still be possible. "Dynamic tree shaking" in upb is a feature that allows messages to be parsed even if the MiniTables have not been fully linked. Unlinked sub-message fields can be parsed by preserving their data in the unknown fields. If the application later discovers that the message field is actually needed, the MiniTable can be patched to properly link that field, and existing message instances can "promote" the data from the unknown fields to an actual message of the correct type. Before this change, dynamic tree shaking stored unparsed message data in the unknown fields of the *parent*. In effect, we were treating the field as if it did not exist at all. This meant that parsing an unlinked field did not affect the hasbits or oneof cases of the parent, nor did it create a `upb_Array` or `upb_Map` for array/map fields. Only when a message was linked and promoted did any of these things occur. While this model had some amount of conceptual simplicity, it caused significant problems with oneofs. When multiple fields inside a single oneof are parsed from the wire, order matters, because later oneof fields must overwrite earlier ones. Dynamic tree shaking can mean that some fields in a oneof are linked while others are not. It is essential that we preserve this ordering semantic even when dynamic tree shaking is being used, but it is difficult to do if the oneof's data can be split between linked fields (which have been reified into parsed field data) and unlinked fields (whose data lives in the unknown fields of the parent). To solve this problem, this CL changes the representation for unlinked fields. Instead of being placed in the parent's unknown fields, we create an actual message instance for each unlinked message we parse, but we use a placeholder "empty message" MiniTable as the message's type. All of the message's data will therefore be placed into the "empty message's" unknown fields. But unlike before, this "empty message" is actually present according to the hasbits, oneof case, and `upb_Array`/`upb_Map` of the parent. This means that all of the oneof presence logic works as normal. Since the MiniTable can be patched at any time, we need a bit in the message instance itself to signal whether a pointer to a sub-message is an "empty message" or not. When dynamic tree shaking is in use, all users must be capable of recognizing an empty message and acting accordingly (promoting, etc) even if the MiniTable itself says that the field is linked. Because dynamic tree shaking imposes this extra requirement on users, we require that users pass an extra option to the decoder to allow parsing of unlinked sub-messages. Many existing users of upb (Ruby, PHP, Python, etc) will always have fully-linked MiniTables, so there is no reason for them to add extra logic to handle empty messages. By omitting the `kUpb_DecodeOption_ExperimentalAllowUnlinked` option, they will be relieved of the duty to check the tagged pointer that would indicate an empty, unlinked message. For existing users of dynamic tree shaking, there are three main changes: 1. The APIs in message/promote.h have changed, and users will need to update to the new interfaces. 2. The model for maps has changed slightly. Before, we required that map entries always had their values linked; for dynamic tree shaking to apply to maps, we required that the *entry* was left unlinked, not the entry's value. In the new model, that is reversed: map entries must always be linked, but a map entry's value can be unlinked. 3. The presence model for unlinked fields has changed. Unlinked fields will now register as "present" from the perspective of hasbits, oneof cases, and array/map entries. Users must test the tagged pointer to know if a message is of the correct, linked type or whether it is a placeholder "empty" message. There is a new function `upb_Message_GetTaggedMessagePtr()`, as well as a new accessor `upb_MessageValue.tagged_msg_val` that can be used to read and test the tagged pointer directly. PiperOrigin-RevId: 535288031
2 years ago
encode_TaggedMessagePtr(e, *ptr, subm, &size);
encode_varint(e, size);
encode_tag(e, upb_MiniTableField_Number(f), kUpb_WireType_Delimited);
} while (ptr != start);
e->depth++;
return;
}
}
#undef VARINT_CASE
if (packed) {
encode_varint(e, e->limit - e->ptr - pre_len);
encode_tag(e, upb_MiniTableField_Number(f), kUpb_WireType_Delimited);
}
}
static void encode_mapentry(upb_encstate* e, uint32_t number,
const upb_MiniTable* layout,
const upb_MapEntry* ent) {
const upb_MiniTableField* key_field = upb_MiniTable_MapKey(layout);
const upb_MiniTableField* val_field = upb_MiniTable_MapValue(layout);
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
size_t pre_len = e->limit - e->ptr;
size_t size;
encode_scalar(e, &ent->v, layout->UPB_PRIVATE(subs), val_field);
encode_scalar(e, &ent->k, layout->UPB_PRIVATE(subs), key_field);
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
size = (e->limit - e->ptr) - pre_len;
encode_varint(e, size);
encode_tag(e, number, kUpb_WireType_Delimited);
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
}
static void encode_map(upb_encstate* e, const upb_Message* msg,
const upb_MiniTableSub* subs,
const upb_MiniTableField* f) {
const upb_Map* map = *UPB_PTR_AT(msg, f->UPB_PRIVATE(offset), const upb_Map*);
const upb_MiniTable* layout =
upb_MiniTableSub_Message(subs[f->UPB_PRIVATE(submsg_index)]);
UPB_ASSERT(upb_MiniTable_FieldCount(layout) == 2);
if (!map || !upb_Map_Size(map)) return;
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
if (e->options & kUpb_EncodeOption_Deterministic) {
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
_upb_sortedmap sorted;
_upb_mapsorter_pushmap(
&e->sorter, layout->UPB_PRIVATE(fields)[0].UPB_PRIVATE(descriptortype),
map, &sorted);
upb_MapEntry ent;
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
while (_upb_sortedmap_next(&e->sorter, map, &sorted, &ent)) {
encode_mapentry(e, upb_MiniTableField_Number(f), layout, &ent);
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
}
_upb_mapsorter_popmap(&e->sorter, &sorted);
} else {
intptr_t iter = UPB_STRTABLE_BEGIN;
upb_StringView key;
upb_value val;
while (upb_strtable_next2(&map->table, &key, &val, &iter)) {
upb_MapEntry ent;
_upb_map_fromkey(key, &ent.k, map->key_size);
_upb_map_fromvalue(val, &ent.v, map->val_size);
encode_mapentry(e, upb_MiniTableField_Number(f), layout, &ent);
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
}
}
}
static bool encode_shouldencode(upb_encstate* e, const upb_Message* msg,
const upb_MiniTableSub* subs,
const upb_MiniTableField* f) {
if (f->presence == 0) {
// Proto3 presence or map/array.
const void* mem = UPB_PTR_AT(msg, f->UPB_PRIVATE(offset), void);
switch (UPB_PRIVATE(_upb_MiniTableField_GetRep)(f)) {
case kUpb_FieldRep_1Byte: {
char ch;
memcpy(&ch, mem, 1);
return ch != 0;
}
case kUpb_FieldRep_4Byte: {
uint32_t u32;
memcpy(&u32, mem, 4);
return u32 != 0;
}
case kUpb_FieldRep_8Byte: {
uint64_t u64;
memcpy(&u64, mem, 8);
return u64 != 0;
}
case kUpb_FieldRep_StringView: {
const upb_StringView* str = (const upb_StringView*)mem;
return str->size != 0;
}
default:
UPB_UNREACHABLE();
}
} else if (UPB_PRIVATE(_upb_MiniTableField_HasHasbit)(f)) {
// Proto2 presence: hasbit.
return UPB_PRIVATE(_upb_Message_GetHasbit)(msg, f);
} else {
// Field is in a oneof.
return UPB_PRIVATE(_upb_Message_GetOneofCase)(msg, f) ==
upb_MiniTableField_Number(f);
}
}
static void encode_field(upb_encstate* e, const upb_Message* msg,
const upb_MiniTableSub* subs,
const upb_MiniTableField* field) {
switch (UPB_PRIVATE(_upb_MiniTableField_Mode)(field)) {
case kUpb_FieldMode_Array:
encode_array(e, msg, subs, field);
break;
case kUpb_FieldMode_Map:
encode_map(e, msg, subs, field);
break;
case kUpb_FieldMode_Scalar:
encode_scalar(e, UPB_PTR_AT(msg, field->UPB_PRIVATE(offset), void), subs,
field);
break;
default:
UPB_UNREACHABLE();
}
}
static void encode_msgset_item(upb_encstate* e, const upb_Extension* ext) {
size_t size;
encode_tag(e, kUpb_MsgSet_Item, kUpb_WireType_EndGroup);
encode_message(e, ext->data.msg_val,
upb_MiniTableExtension_GetSubMessage(ext->ext), &size);
encode_varint(e, size);
encode_tag(e, kUpb_MsgSet_Message, kUpb_WireType_Delimited);
encode_varint(e, upb_MiniTableExtension_Number(ext->ext));
encode_tag(e, kUpb_MsgSet_TypeId, kUpb_WireType_Varint);
encode_tag(e, kUpb_MsgSet_Item, kUpb_WireType_StartGroup);
}
static void encode_ext(upb_encstate* e, const upb_Extension* ext,
bool is_message_set) {
if (UPB_UNLIKELY(is_message_set)) {
encode_msgset_item(e, ext);
} else {
encode_field(e, (upb_Message*)&ext->data, &ext->ext->UPB_PRIVATE(sub),
&ext->ext->UPB_PRIVATE(field));
}
}
static void encode_message(upb_encstate* e, const upb_Message* msg,
const upb_MiniTable* m, size_t* size) {
size_t pre_len = e->limit - e->ptr;
if (e->options & kUpb_EncodeOption_CheckRequired) {
if (m->UPB_PRIVATE(required_count)) {
if (!UPB_PRIVATE(_upb_Message_IsInitializedShallow)(msg, m)) {
encode_err(e, kUpb_EncodeStatus_MissingRequired);
}
}
}
if ((e->options & kUpb_EncodeOption_SkipUnknown) == 0) {
size_t unknown_size;
const char* unknown = upb_Message_GetUnknown(msg, &unknown_size);
if (unknown) {
encode_bytes(e, unknown, unknown_size);
}
}
if (m->UPB_PRIVATE(ext) != kUpb_ExtMode_NonExtendable) {
/* Encode all extensions together. Unlike C++, we do not attempt to keep
* these in field number order relative to normal fields or even to each
* other. */
size_t ext_count;
const upb_Extension* ext =
UPB_PRIVATE(_upb_Message_Getexts)(msg, &ext_count);
if (ext_count) {
if (e->options & kUpb_EncodeOption_Deterministic) {
_upb_sortedmap sorted;
_upb_mapsorter_pushexts(&e->sorter, ext, ext_count, &sorted);
while (_upb_sortedmap_nextext(&e->sorter, &sorted, &ext)) {
encode_ext(e, ext, m->UPB_PRIVATE(ext) == kUpb_ExtMode_IsMessageSet);
}
_upb_mapsorter_popmap(&e->sorter, &sorted);
} else {
const upb_Extension* end = ext + ext_count;
for (; ext != end; ext++) {
encode_ext(e, ext, m->UPB_PRIVATE(ext) == kUpb_ExtMode_IsMessageSet);
}
}
}
}
if (upb_MiniTable_FieldCount(m)) {
const upb_MiniTableField* f =
&m->UPB_PRIVATE(fields)[m->UPB_PRIVATE(field_count)];
const upb_MiniTableField* first = &m->UPB_PRIVATE(fields)[0];
while (f != first) {
f--;
if (encode_shouldencode(e, msg, m->UPB_PRIVATE(subs), f)) {
encode_field(e, msg, m->UPB_PRIVATE(subs), f);
}
}
}
*size = (e->limit - e->ptr) - pre_len;
}
static upb_EncodeStatus upb_Encoder_Encode(upb_encstate* const encoder,
const upb_Message* const msg,
const upb_MiniTable* const l,
char** const buf, size_t* const size,
bool prepend_len) {
// Unfortunately we must continue to perform hackery here because there are
// code paths which blindly copy the returned pointer without bothering to
// check for errors until much later (b/235839510). So we still set *buf to
// NULL on error and we still set it to non-NULL on a successful empty result.
if (UPB_SETJMP(encoder->err) == 0) {
size_t encoded_msg_size;
encode_message(encoder, msg, l, &encoded_msg_size);
if (prepend_len) {
encode_varint(encoder, encoded_msg_size);
}
*size = encoder->limit - encoder->ptr;
if (*size == 0) {
static char ch;
*buf = &ch;
} else {
UPB_ASSERT(encoder->ptr);
*buf = encoder->ptr;
}
} else {
UPB_ASSERT(encoder->status != kUpb_EncodeStatus_Ok);
*buf = NULL;
*size = 0;
}
_upb_mapsorter_destroy(&encoder->sorter);
return encoder->status;
}
static upb_EncodeStatus _upb_Encode(const upb_Message* msg,
const upb_MiniTable* l, int options,
upb_Arena* arena, char** buf, size_t* size,
bool prepend_len) {
upb_encstate e;
unsigned depth = (unsigned)options >> 16;
e.status = kUpb_EncodeStatus_Ok;
e.arena = arena;
e.buf = NULL;
e.limit = NULL;
e.ptr = NULL;
e.depth = depth ? depth : kUpb_WireFormat_DefaultDepthLimit;
Added map sorting to binary and text encoders. For the binary encoder, sorting is off by default. For the text encoder, sorting is on by default. Both defaults can be explicitly overridden. This grows code size a bit. I think we could potentially shave this (and other map-related code size) by having the generated code inject a function pointer to the map-related parsing/serialization code if maps are present. FILE SIZE VM SIZE -------------- -------------- +86% +1.07Ki +71% +768 upb/msg.c [NEW] +391 [NEW] +344 _upb_mapsorter_pushmap [NEW] +158 [NEW] +112 _upb_mapsorter_cmpstr [NEW] +111 [NEW] +64 _upb_mapsorter_cmpbool [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpi64 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu32 [NEW] +110 [NEW] +64 _upb_mapsorter_cmpu64 -3.6% -8 -4.3% -8 _upb_map_new +9.5% +464 +9.2% +424 upb/text_encode.c [NEW] +656 [NEW] +616 txtenc_mapentry +15% +32 +20% +32 upb_text_encode -20.1% -224 -20.7% -224 txtenc_msg +5.7% +342 +5.3% +296 upb/encode.c [NEW] +344 [NEW] +304 encode_mapentry [NEW] +246 [NEW] +208 upb_encode_ex [NEW] +41 [NEW] +16 upb_encode_ex.ch +0.7% +8 +0.7% +8 encode_scalar -1.0% -32 -1.0% -32 encode_message [DEL] -38 [DEL] -16 upb_encode.ch [DEL] -227 [DEL] -192 upb_encode +2.0% +152 +2.2% +152 upb/decode.c +44% +128 +44% +128 [section .rodata] +3.4% +24 +3.4% +24 _GLOBAL_OFFSET_TABLE_ +0.6% +107 +0.3% +48 upb/def.c [NEW] +100 [NEW] +48 upb_fielddef_descriptortype +7.1% +7 [ = ] 0 upb_fielddef_defaultint32 +2.9% +24 +2.9% +24 [section .dynsym] +1.2% +24 [ = ] 0 [section .symtab] +3.2% +16 +3.2% +16 [section .plt] [NEW] +16 [NEW] +16 memcmp@plt +0.5% +16 +0.6% +16 tests/conformance_upb.c +1.5% +16 +1.6% +16 DoTestIo +0.1% +16 +0.1% +16 upb/json_decode.c +0.4% +16 +0.4% +16 jsondec_wellknown +3.0% +8 +3.0% +8 [section .got.plt] +3.0% +8 +3.0% +8 _GLOBAL_OFFSET_TABLE_ +1.6% +7 +1.6% +7 [section .dynstr] +1.8% +4 +1.8% +4 [section .hash] +0.5% +3 +0.5% +3 [LOAD #2 [RX]] +2.8% +2 +2.8% +2 [section .gnu.version] -60.0% -1.74Ki [ = ] 0 [Unmapped] +0.3% +496 +1.4% +1.74Ki TOTAL
4 years ago
e.options = options;
_upb_mapsorter_init(&e.sorter);
return upb_Encoder_Encode(&e, msg, l, buf, size, prepend_len);
}
upb_EncodeStatus upb_Encode(const upb_Message* msg, const upb_MiniTable* l,
int options, upb_Arena* arena, char** buf,
size_t* size) {
return _upb_Encode(msg, l, options, arena, buf, size, false);
}
upb_EncodeStatus upb_EncodeLengthPrefixed(const upb_Message* msg,
const upb_MiniTable* l, int options,
upb_Arena* arena, char** buf,
size_t* size) {
return _upb_Encode(msg, l, options, arena, buf, size, true);
}