|
|
|
// 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
|
|
|
|
|
|
|
|
#include "google/protobuf/compiler/hpb/gen_accessors.h"
|
|
|
|
|
|
|
|
#include <string>
|
|
|
|
|
|
|
|
#include "absl/container/flat_hash_set.h"
|
|
|
|
#include "absl/strings/match.h"
|
|
|
|
#include "absl/strings/str_cat.h"
|
|
|
|
#include "absl/strings/string_view.h"
|
|
|
|
#include "google/protobuf/descriptor.h"
|
|
|
|
#include "google/protobuf/compiler/hpb/gen_repeated_fields.h"
|
|
|
|
#include "google/protobuf/compiler/hpb/gen_utils.h"
|
|
|
|
#include "google/protobuf/compiler/hpb/names.h"
|
|
|
|
#include "google/protobuf/compiler/hpb/output.h"
|
|
|
|
#include "upb_generator/common.h"
|
|
|
|
#include "upb_generator/keywords.h"
|
|
|
|
#include "upb_generator/names.h"
|
|
|
|
|
|
|
|
namespace google::protobuf::hpb_generator {
|
|
|
|
|
|
|
|
namespace protobuf = ::proto2;
|
|
|
|
|
|
|
|
using NameToFieldDescriptorMap =
|
|
|
|
absl::flat_hash_map<absl::string_view, const protobuf::FieldDescriptor*>;
|
|
|
|
|
|
|
|
void WriteFieldAccessorHazzer(const protobuf::Descriptor* desc,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
absl::string_view resolved_field_name,
|
|
|
|
absl::string_view resolved_upbc_name,
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
Output& output);
|
|
|
|
void WriteFieldAccessorClear(const protobuf::Descriptor* desc,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
absl::string_view resolved_field_name,
|
|
|
|
absl::string_view resolved_upbc_name,
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
Output& output);
|
|
|
|
void WriteMapFieldAccessors(const protobuf::Descriptor* desc,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
absl::string_view resolved_field_name,
|
|
|
|
absl::string_view resolved_upbc_name,
|
|
|
|
Output& output);
|
|
|
|
|
|
|
|
void WriteMapAccessorDefinitions(const protobuf::Descriptor* message,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
absl::string_view resolved_field_name,
|
|
|
|
absl::string_view class_name, Output& output);
|
|
|
|
|
|
|
|
// Returns C++ class member name by resolving naming conflicts across
|
|
|
|
// proto field names (such as clear_ prefixes) and keyword collisions.
|
|
|
|
//
|
|
|
|
// The Upb C generator prefixes all accessors with package and class names
|
|
|
|
// avoiding collisions. Therefore we need to use raw field names when calling
|
|
|
|
// into C accessors but need to fully resolve conflicts for C++ class members.
|
|
|
|
std::string ResolveFieldName(const protobuf::FieldDescriptor* field,
|
|
|
|
const NameToFieldDescriptorMap& field_names);
|
|
|
|
|
|
|
|
NameToFieldDescriptorMap CreateFieldNameMap(
|
|
|
|
const protobuf::Descriptor* message) {
|
|
|
|
NameToFieldDescriptorMap field_names;
|
|
|
|
for (int i = 0; i < message->field_count(); i++) {
|
|
|
|
const protobuf::FieldDescriptor* field = message->field(i);
|
|
|
|
field_names.emplace(field->name(), field);
|
|
|
|
}
|
|
|
|
return field_names;
|
|
|
|
}
|
|
|
|
|
|
|
|
void WriteFieldAccessorsInHeader(const protobuf::Descriptor* desc,
|
|
|
|
Output& output) {
|
|
|
|
// Generate const methods.
|
|
|
|
OutputIndenter i(output);
|
|
|
|
|
|
|
|
auto field_names = CreateFieldNameMap(desc);
|
|
|
|
auto upbc_field_names = upb::generator::CreateFieldNameMap(desc);
|
|
|
|
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
for (const auto* field : FieldNumberOrder(desc)) {
|
|
|
|
std::string resolved_field_name = ResolveFieldName(field, field_names);
|
|
|
|
std::string resolved_upbc_name =
|
|
|
|
upb::generator::ResolveFieldName(field, upbc_field_names);
|
|
|
|
WriteFieldAccessorHazzer(desc, field, resolved_field_name,
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
resolved_upbc_name, output);
|
|
|
|
WriteFieldAccessorClear(desc, field, resolved_field_name,
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
resolved_upbc_name, output);
|
|
|
|
|
|
|
|
if (field->is_map()) {
|
|
|
|
WriteMapFieldAccessors(desc, field, resolved_field_name,
|
|
|
|
resolved_upbc_name, output);
|
|
|
|
} else if (desc->options().map_entry()) {
|
|
|
|
// TODO Implement map entry
|
|
|
|
} else if (field->is_repeated()) {
|
|
|
|
WriteRepeatedFieldsInMessageHeader(desc, field, resolved_field_name,
|
|
|
|
resolved_upbc_name, output);
|
|
|
|
} else {
|
|
|
|
// non-repeated.
|
|
|
|
if (field->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_STRING) {
|
|
|
|
output(R"cc(
|
|
|
|
$0 $1() const;
|
|
|
|
void set_$1($0 value);
|
|
|
|
)cc",
|
|
|
|
CppConstType(field), resolved_field_name);
|
|
|
|
} else if (field->cpp_type() ==
|
|
|
|
protobuf::FieldDescriptor::CPPTYPE_MESSAGE) {
|
|
|
|
output(R"cc(
|
|
|
|
$1 $2() const;
|
|
|
|
$0 mutable_$2();
|
|
|
|
)cc",
|
|
|
|
MessagePtrConstType(field, /* const */ false),
|
|
|
|
MessagePtrConstType(field, /* const */ true),
|
|
|
|
resolved_field_name, resolved_upbc_name);
|
|
|
|
} else {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
inline $0 $1() const { return $2_$3(msg_); }
|
|
|
|
inline void set_$1($0 value) { return $2_set_$3(msg_, value); }
|
|
|
|
)cc",
|
|
|
|
CppConstType(field), resolved_field_name, MessageName(desc),
|
|
|
|
resolved_upbc_name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void WriteFieldAccessorHazzer(const protobuf::Descriptor* desc,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
const absl::string_view resolved_field_name,
|
|
|
|
const absl::string_view resolved_upbc_name,
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
Output& output) {
|
|
|
|
// Generate hazzer (if any).
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
if (field->has_presence()) {
|
|
|
|
// Has presence.
|
|
|
|
output("inline bool has_$0() const { return $1_has_$2(msg_); }\n",
|
|
|
|
resolved_field_name, MessageName(desc), resolved_upbc_name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void WriteFieldAccessorClear(const protobuf::Descriptor* desc,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
const absl::string_view resolved_field_name,
|
|
|
|
const absl::string_view resolved_upbc_name,
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
Output& output) {
|
|
|
|
if (field->has_presence()) {
|
|
|
|
output("void clear_$0() { $2_clear_$1(msg_); }\n", resolved_field_name,
|
|
|
|
resolved_upbc_name, MessageName(desc));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void WriteMapFieldAccessors(const protobuf::Descriptor* desc,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
const absl::string_view resolved_field_name,
|
|
|
|
const absl::string_view resolved_upbc_name,
|
|
|
|
Output& output) {
|
|
|
|
const protobuf::Descriptor* entry = field->message_type();
|
|
|
|
const protobuf::FieldDescriptor* key = entry->FindFieldByNumber(1);
|
|
|
|
const protobuf::FieldDescriptor* val = entry->FindFieldByNumber(2);
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
inline size_t $0_size() const { return $1_$3_size(msg_); }
|
|
|
|
inline void clear_$0() { $1_clear_$3(msg_); }
|
|
|
|
void delete_$0($2 key);
|
|
|
|
)cc",
|
|
|
|
resolved_field_name, MessageName(desc), CppConstType(key),
|
|
|
|
resolved_upbc_name);
|
|
|
|
|
|
|
|
if (val->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_MESSAGE) {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
bool set_$0($1 key, $3 value);
|
|
|
|
bool set_$0($1 key, $4 value);
|
|
|
|
absl::StatusOr<$3> get_$0($1 key);
|
|
|
|
)cc",
|
|
|
|
resolved_field_name, CppConstType(key), CppConstType(val),
|
|
|
|
MessagePtrConstType(val, /* is_const */ true),
|
|
|
|
MessagePtrConstType(val, /* is_const */ false));
|
|
|
|
} else {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
bool set_$0($1 key, $2 value);
|
|
|
|
absl::StatusOr<$2> get_$0($1 key);
|
|
|
|
)cc",
|
|
|
|
resolved_field_name, CppConstType(key), CppConstType(val));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
void WriteAccessorsInSource(const protobuf::Descriptor* desc, Output& output) {
|
|
|
|
std::string class_name = ClassName(desc);
|
|
|
|
absl::StrAppend(&class_name, "Access");
|
|
|
|
output("namespace internal {\n");
|
|
|
|
const char arena_expression[] = "arena_";
|
|
|
|
auto field_names = CreateFieldNameMap(desc);
|
|
|
|
auto upbc_field_names = upb::generator::CreateFieldNameMap(desc);
|
|
|
|
|
|
|
|
// Generate const methods.
|
|
|
|
OutputIndenter i(output);
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
for (const auto* field : FieldNumberOrder(desc)) {
|
|
|
|
std::string resolved_field_name = ResolveFieldName(field, field_names);
|
|
|
|
std::string resolved_upbc_name =
|
|
|
|
upb::generator::ResolveFieldName(field, upbc_field_names);
|
|
|
|
if (field->is_map()) {
|
|
|
|
WriteMapAccessorDefinitions(desc, field, resolved_field_name, class_name,
|
|
|
|
output);
|
|
|
|
} else if (desc->options().map_entry()) {
|
|
|
|
// TODO Implement map entry
|
|
|
|
} else if (field->is_repeated()) {
|
|
|
|
if (field->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_MESSAGE) {
|
|
|
|
WriteRepeatedMessageAccessor(desc, field, resolved_field_name,
|
|
|
|
class_name, output);
|
|
|
|
} else if (field->cpp_type() ==
|
|
|
|
protobuf::FieldDescriptor::CPPTYPE_STRING) {
|
|
|
|
WriteRepeatedStringAccessor(desc, field, resolved_field_name,
|
|
|
|
class_name, output);
|
|
|
|
} else {
|
|
|
|
WriteRepeatedScalarAccessor(desc, field, resolved_field_name,
|
|
|
|
class_name, output);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// non-repeated field.
|
|
|
|
if (field->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_STRING) {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
$1 $0::$2() const {
|
|
|
|
return ::hpb::UpbStrToStringView($3_$4(msg_));
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, CppConstType(field), resolved_field_name,
|
|
|
|
MessageName(desc), resolved_upbc_name);
|
|
|
|
// Set string.
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
void $0::set_$2($1 value) {
|
|
|
|
$4_set_$3(msg_, ::hpb::UpbStrFromStringView(value, $5));
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, CppConstType(field), resolved_field_name,
|
|
|
|
resolved_upbc_name, MessageName(desc), arena_expression);
|
|
|
|
} else if (field->cpp_type() ==
|
|
|
|
protobuf::FieldDescriptor::CPPTYPE_MESSAGE) {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
$1 $0::$2() const {
|
|
|
|
if (!has_$2()) {
|
|
|
|
return $4::default_instance();
|
|
|
|
}
|
|
|
|
return ::hpb::internal::CreateMessage<$4>(
|
|
|
|
(upb_Message*)($3_$5(msg_)), arena_);
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, MessagePtrConstType(field, /* is_const */ true),
|
|
|
|
resolved_field_name, MessageName(desc),
|
|
|
|
MessageBaseType(field, /* maybe_const */ false),
|
|
|
|
resolved_upbc_name);
|
|
|
|
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
$1 $0::mutable_$2() {
|
|
|
|
return ::hpb::internal::CreateMessageProxy<$4>(
|
|
|
|
(upb_Message*)($3_mutable_$5(msg_, $6)), $6);
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, MessagePtrConstType(field, /* is_const */ false),
|
|
|
|
resolved_field_name, MessageName(desc),
|
|
|
|
MessageBaseType(field, /* maybe_const */ false), resolved_upbc_name,
|
|
|
|
arena_expression);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
output("\n");
|
|
|
|
output("} // namespace internal\n\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
void WriteMapAccessorDefinitions(const protobuf::Descriptor* message,
|
|
|
|
const protobuf::FieldDescriptor* field,
|
|
|
|
const absl::string_view resolved_field_name,
|
|
|
|
const absl::string_view class_name,
|
|
|
|
Output& output) {
|
|
|
|
const protobuf::Descriptor* entry = field->message_type();
|
|
|
|
const protobuf::FieldDescriptor* key = entry->FindFieldByNumber(1);
|
|
|
|
const protobuf::FieldDescriptor* val = entry->FindFieldByNumber(2);
|
|
|
|
absl::string_view upbc_name = field->name();
|
|
|
|
absl::string_view converted_key_name = "key";
|
|
|
|
absl::string_view optional_conversion_code = "";
|
|
|
|
|
|
|
|
if (key->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_STRING) {
|
|
|
|
// Insert conversion from absl::string_view to upb_StringView.
|
|
|
|
// Creates upb_StringView on stack to prevent allocation.
|
|
|
|
converted_key_name = "upb_key";
|
|
|
|
optional_conversion_code =
|
|
|
|
"upb_StringView upb_key = {key.data(), key.size()};\n";
|
|
|
|
}
|
|
|
|
if (val->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_MESSAGE) {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
bool $0::set_$1($2 key, $3 value) {
|
|
|
|
upb_Message* clone = upb_Message_DeepClone(
|
|
|
|
::hpb::internal::PrivateAccess::GetInternalMsg(value), &$9,
|
|
|
|
arena_);
|
|
|
|
$6return $4_$8_set(msg_, $7, ($5*)clone, arena_);
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key),
|
|
|
|
MessagePtrConstType(val, /* is_const */ true), MessageName(message),
|
|
|
|
MessageName(val->message_type()), optional_conversion_code,
|
|
|
|
converted_key_name, upbc_name,
|
|
|
|
::upb::generator::MessageInit(val->message_type()->full_name()));
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
bool $0::set_$1($2 key, $3 value) {
|
|
|
|
upb_Message* clone = upb_Message_DeepClone(
|
|
|
|
::hpb::internal::PrivateAccess::GetInternalMsg(value), &$9,
|
|
|
|
arena_);
|
|
|
|
$6return $4_$8_set(msg_, $7, ($5*)clone, arena_);
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key),
|
|
|
|
MessagePtrConstType(val, /* is_const */ false), MessageName(message),
|
|
|
|
MessageName(val->message_type()), optional_conversion_code,
|
|
|
|
converted_key_name, upbc_name,
|
|
|
|
::upb::generator::MessageInit(val->message_type()->full_name()));
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
absl::StatusOr<$3> $0::get_$1($2 key) {
|
|
|
|
$5* msg_value;
|
|
|
|
$7bool success = $4_$9_get(msg_, $8, &msg_value);
|
|
|
|
if (success) {
|
|
|
|
return ::hpb::internal::CreateMessage<$6>(UPB_UPCAST(msg_value), arena_);
|
|
|
|
}
|
|
|
|
return absl::NotFoundError("");
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key),
|
|
|
|
MessagePtrConstType(val, /* is_const */ true), MessageName(message),
|
|
|
|
MessageName(val->message_type()),
|
|
|
|
QualifiedClassName(val->message_type()), optional_conversion_code,
|
|
|
|
converted_key_name, upbc_name);
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
void $0::delete_$1($2 key) { $6$4_$8_delete(msg_, $7); }
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key),
|
|
|
|
MessagePtrConstType(val, /* is_const */ false), MessageName(message),
|
|
|
|
MessageName(val->message_type()), optional_conversion_code,
|
|
|
|
converted_key_name, upbc_name);
|
|
|
|
} else if (val->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_STRING) {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
bool $0::set_$1($2 key, $3 value) {
|
|
|
|
$5return $4_$7_set(
|
|
|
|
msg_, $6, ::hpb::UpbStrFromStringView(value, arena_), arena_);
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key), CppConstType(val),
|
|
|
|
MessageName(message), optional_conversion_code, converted_key_name,
|
|
|
|
upbc_name);
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
absl::StatusOr<$3> $0::get_$1($2 key) {
|
|
|
|
upb_StringView value;
|
|
|
|
$5bool success = $4_$7_get(msg_, $6, &value);
|
|
|
|
if (success) {
|
|
|
|
return absl::string_view(value.data, value.size);
|
|
|
|
}
|
|
|
|
return absl::NotFoundError("");
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key), CppConstType(val),
|
|
|
|
MessageName(message), optional_conversion_code, converted_key_name,
|
|
|
|
upbc_name);
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
void $0::delete_$1($2 key) { $5$4_$7_delete(msg_, $6); }
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key), CppConstType(val),
|
|
|
|
MessageName(message), optional_conversion_code, converted_key_name,
|
|
|
|
upbc_name);
|
|
|
|
} else {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
bool $0::set_$1($2 key, $3 value) {
|
|
|
|
$5return $4_$7_set(msg_, $6, value, arena_);
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key), CppConstType(val),
|
|
|
|
MessageName(message), optional_conversion_code, converted_key_name,
|
|
|
|
upbc_name);
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
absl::StatusOr<$3> $0::get_$1($2 key) {
|
|
|
|
$3 value;
|
|
|
|
$5bool success = $4_$7_get(msg_, $6, &value);
|
|
|
|
if (success) {
|
|
|
|
return value;
|
|
|
|
}
|
|
|
|
return absl::NotFoundError("");
|
|
|
|
}
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key), CppConstType(val),
|
|
|
|
MessageName(message), optional_conversion_code, converted_key_name,
|
|
|
|
upbc_name);
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
void $0::delete_$1($2 key) { $5$4_$7_delete(msg_, $6); }
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name, CppConstType(key), CppConstType(val),
|
|
|
|
MessageName(message), optional_conversion_code, converted_key_name,
|
|
|
|
upbc_name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void WriteUsingAccessorsInHeader(const protobuf::Descriptor* desc,
|
|
|
|
MessageClassType handle_type, Output& output) {
|
|
|
|
bool read_only = handle_type == MessageClassType::kMessageCProxy;
|
|
|
|
|
|
|
|
// Generate const methods.
|
|
|
|
OutputIndenter i(output);
|
|
|
|
std::string class_name = ClassName(desc);
|
|
|
|
auto field_names = CreateFieldNameMap(desc);
|
|
|
|
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
for (const auto* field : FieldNumberOrder(desc)) {
|
|
|
|
std::string resolved_field_name = ResolveFieldName(field, field_names);
|
|
|
|
// Generate hazzer (if any).
|
upb is self-hosting!
This CL changes the upb compiler to no longer depend on C++ protobuf libraries. upb now uses its own reflection libraries to implement its code generator.
# Key Benefits
1. upb can now use its own reflection libraries throughout the compiler. This makes upb more consistent and principled, and gives us more chances to dogfood our own C++ reflection API. This highlighted several parts of the C++ reflection API that were incomplete.
2. This CL removes code duplication that previously existed in the compiler. The upb reflection library has code to build MiniDescriptors and MiniTables out of descriptors, but prior to this CL the upb compiler could not use it. The upb compiler had a separate copy of this logic, and the compiler's copy of this logic was especially tricky and hard to maintain. This CL removes the separate copy of that logic.
3. This CL (mostly) removes upb's dependency on the C++ protobuf library. We still depend on `protoc` (the binary), but the runtime and compiler no longer link against C++'s libraries. This opens up the possibility of speeding up some builds significantly if we can use a prebuilt `protoc` binary.
# Bootstrap Stages
To bootstrap, we check in a copy of our generated code for `descriptor.proto` and `plugin.proto`. This allows the compiler to depend on the generated code for these two protos without creating a circular dependency. This code is checked in to the `stage0` directory.
The bootstrapping process is divided into a few stages. All `cc_library()`, `upb_proto_library()`, and `cc_binary()` targets that would otherwise be circular participate in this staging process. That currently includes:
* `//third_party/upb:descriptor_upb_proto`
* `//third_party/upb:plugin_upb_proto`
* `//third_party/upb:reflection`
* `//third_party/upb:reflection_internal`
* `//third_party/upbc:common`
* `//third_party/upbc:file_layout`
* `//third_party/upbc:plugin`
* `//third_party/upbc:protoc-gen-upb`
For each of these targets, we produce a rule for each stage (the logic for this is nicely encapsulated in Blaze/Bazel macros like `bootstrap_cc_library()` and `bootstrap_upb_proto_library()`, so the `BUILD` file remains readable). For example:
* `//third_party/upb:descriptor_upb_proto_stage0`
* `//third_party/upb:descriptor_upb_proto_stage1`
* `//third_party/upb:descriptor_upb_proto`
The stages are:
1. `stage0`: This uses the checked-in version of the generated code. The stage0 compiler is correct and outputs the same code as all other compilers, but it is unnecessarily slow because its protos were compiled in bootstrap mode. The stage0 compiler is used to generate protos for stage1.
2. `stage1`: The stage1 compiler is correct and fast, and therefore we use it in almost all cases (eg. `upb_proto_library()`). However its own protos were not generated using `upb_proto_library()`, so its `cc_library()` targets cannot be safely mixed with `upb_proto_library()`, as this would lead to duplicate symbols.
3. final (no stage): The final compiler is identical to the `stage1` compiler. The only difference is that its protos were built with `upb_proto_library()`. This doesn't matter very much for the compiler binary, but for the `cc_library()` targets like `//third_party/upb:reflection`, only the final targets can be safely linked in by other applications.
# "Bootstrap Mode" Protos
The checked-in generated code is generated in a special "bootstrap" mode that is a bit different than normal generated code. Bootstrap mode avoids depending on the internal representation of MiniTables or the messages, at the cost of slower runtime performance.
Bootstrap mode only interacts with MiniTables and messages using public APIs such as `upb_MiniTable_Build()`, `upb_Message_GetInt32()`, etc. This is very important as it allows us to change the internal representation without needing to regenerate our bootstrap protos. This will make it far easier to write CLs that change the internal representation, because it avoids the awkward dance of trying to regenerate the bootstrap protos when the compiler itself is broken due to bootstrap protos being out of date.
The bootstrap generated code does have two downsides:
1. The accessors are less efficient, because they look up MiniTable fields by number instead of hard-coding the MiniTableField into the generated code.
2. It requires runtime initialization of the MiniTables, which costs CPU cycles at startup, and also allocates memory which is never freed. Per google3 rules this is not really a leak, since this memory is still reachable via static variables, but it is undesirable in many contexts. We could fix this part by introducing the equivalent of `google::protobuf::ShutdownProtobufLibrary()`).
These downsides are fine for the bootstrapping process, but they are reason enough not to enable bootstrap mode in general for all protos.
# Bootstrapping Always Uses OSS Protos
To enable smooth syncing between Google3 and OSS, we always use an OSS version of the checked in generated code for `stage0`, even in google3.
This requires that the google3 code can be switched to reference the OSS proto names using a preprocessor define. We introduce the `UPB_DESC(xyz)` macro for this, which will expand into either `proto2_xyz` or `google_protobuf_xyz`. Any libraries used in `stage0` must use `UPB_DESC(xyz)` rather than refer to the symbol names directly.
PiperOrigin-RevId: 501458451
2 years ago
|
|
|
if (field->has_presence()) {
|
|
|
|
output("using $0Access::has_$1;\n", class_name, resolved_field_name);
|
|
|
|
if (!read_only) {
|
|
|
|
output("using $0Access::clear_$1;\n", class_name, resolved_field_name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (field->is_map()) {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
using $0Access::$1_size;
|
|
|
|
using $0Access::get_$1;
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name);
|
|
|
|
if (!read_only) {
|
|
|
|
output(
|
|
|
|
R"cc(
|
|
|
|
using $0Access::clear_$1;
|
|
|
|
using $0Access::delete_$1;
|
|
|
|
using $0Access::set_$1;
|
|
|
|
)cc",
|
|
|
|
class_name, resolved_field_name);
|
|
|
|
}
|
|
|
|
} else if (desc->options().map_entry()) {
|
|
|
|
// TODO Implement map entry
|
|
|
|
} else if (field->is_repeated()) {
|
|
|
|
WriteRepeatedFieldUsingAccessors(field, class_name, resolved_field_name,
|
|
|
|
output, read_only);
|
|
|
|
} else {
|
|
|
|
if (field->cpp_type() == protobuf::FieldDescriptor::CPPTYPE_MESSAGE) {
|
|
|
|
output("using $0Access::$1;\n", ClassName(desc), resolved_field_name);
|
|
|
|
if (!read_only) {
|
|
|
|
output("using $0Access::mutable_$1;\n", class_name,
|
|
|
|
resolved_field_name);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
output("using $0Access::$1;\n", class_name, resolved_field_name);
|
|
|
|
if (!read_only) {
|
|
|
|
output("using $0Access::set_$1;\n", class_name, resolved_field_name);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (int i = 0; i < desc->real_oneof_decl_count(); ++i) {
|
|
|
|
const protobuf::OneofDescriptor* oneof = desc->oneof_decl(i);
|
|
|
|
output("using $0Access::$1_case;\n", class_name, oneof->name());
|
|
|
|
output("using $0Access::$1Case;\n", class_name,
|
|
|
|
ToCamelCase(oneof->name(), /*lower_first=*/false));
|
|
|
|
for (int j = 0; j < oneof->field_count(); ++j) {
|
|
|
|
const protobuf::FieldDescriptor* field = oneof->field(j);
|
|
|
|
output("using $0Access::k$1;\n", class_name,
|
|
|
|
ToCamelCase(field->name(), /*lower_first=*/false),
|
|
|
|
field->number());
|
|
|
|
}
|
|
|
|
output("using $0Access::$1_NOT_SET;\n", class_name,
|
|
|
|
absl::AsciiStrToUpper(oneof->name()));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void WriteOneofAccessorsInHeader(const protobuf::Descriptor* desc,
|
|
|
|
Output& output) {
|
|
|
|
// Generate const methods.
|
|
|
|
OutputIndenter i(output);
|
|
|
|
std::string class_name = ClassName(desc);
|
|
|
|
auto field_names = CreateFieldNameMap(desc);
|
|
|
|
for (int i = 0; i < desc->real_oneof_decl_count(); ++i) {
|
|
|
|
const protobuf::OneofDescriptor* oneof = desc->oneof_decl(i);
|
|
|
|
output("enum $0Case {\n",
|
|
|
|
ToCamelCase(oneof->name(), /*lower_first=*/false));
|
|
|
|
for (int j = 0; j < oneof->field_count(); ++j) {
|
|
|
|
const protobuf::FieldDescriptor* field = oneof->field(j);
|
|
|
|
output(" k$0 = $1,\n", ToCamelCase(field->name(), /*lower_first=*/false),
|
|
|
|
field->number());
|
|
|
|
}
|
|
|
|
output(" $0_NOT_SET = 0,\n", absl::AsciiStrToUpper(oneof->name()));
|
|
|
|
output("};\n\n");
|
|
|
|
output("$0Case $1_case() const {\n",
|
|
|
|
ToCamelCase(oneof->name(), /*lower_first=*/false), oneof->name());
|
|
|
|
for (int j = 0; j < oneof->field_count(); ++j) {
|
|
|
|
const protobuf::FieldDescriptor* field = oneof->field(j);
|
|
|
|
std::string resolved_field_name = ResolveFieldName(field, field_names);
|
|
|
|
output(" if (has_$0()) { return k$1; }\n", resolved_field_name,
|
|
|
|
ToCamelCase(field->name(), /*lower_first=*/false));
|
|
|
|
}
|
|
|
|
output(" return $0_NOT_SET;\n", absl::AsciiStrToUpper(oneof->name()));
|
|
|
|
output("}\n;");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
std::string ResolveFieldName(const protobuf::FieldDescriptor* field,
|
|
|
|
const NameToFieldDescriptorMap& field_names) {
|
|
|
|
// C++ implementation specific reserved names.
|
|
|
|
static const auto& kReservedNames =
|
|
|
|
*new absl::flat_hash_set<absl::string_view>({
|
|
|
|
"msg",
|
|
|
|
"msg_",
|
|
|
|
"arena",
|
|
|
|
"arena_",
|
|
|
|
});
|
|
|
|
|
|
|
|
// C++ specific prefixes used by code generator for field access.
|
|
|
|
static constexpr absl::string_view kClearMethodPrefix = "clear_";
|
|
|
|
static constexpr absl::string_view kSetMethodPrefix = "set_";
|
|
|
|
static constexpr absl::string_view kHasMethodPrefix = "has_";
|
|
|
|
static constexpr absl::string_view kDeleteMethodPrefix = "delete_";
|
|
|
|
static constexpr absl::string_view kAddToRepeatedMethodPrefix = "add_";
|
|
|
|
static constexpr absl::string_view kResizeArrayMethodPrefix = "resize_";
|
|
|
|
|
|
|
|
// List of generated accessor prefixes to check against.
|
|
|
|
// Example:
|
|
|
|
// optional repeated string phase = 236;
|
|
|
|
// optional bool clear_phase = 237;
|
|
|
|
static constexpr absl::string_view kAccessorPrefixes[] = {
|
|
|
|
kClearMethodPrefix, kDeleteMethodPrefix, kAddToRepeatedMethodPrefix,
|
|
|
|
kResizeArrayMethodPrefix, kSetMethodPrefix, kHasMethodPrefix};
|
|
|
|
|
|
|
|
absl::string_view field_name = field->name();
|
|
|
|
if (kReservedNames.count(field_name) > 0) {
|
|
|
|
if (absl::EndsWith(field_name, "_")) {
|
|
|
|
return absl::StrCat(field_name, "_");
|
|
|
|
} else {
|
|
|
|
return absl::StrCat(field_name, "__");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (const auto prefix : kAccessorPrefixes) {
|
|
|
|
// If field name starts with a prefix such as clear_ and the proto
|
|
|
|
// contains a field name with trailing end, depending on type of field
|
|
|
|
// (repeated, map, message) we have a conflict to resolve.
|
|
|
|
if (absl::StartsWith(field_name, prefix)) {
|
|
|
|
auto match = field_names.find(field_name.substr(prefix.size()));
|
|
|
|
if (match != field_names.end()) {
|
|
|
|
const auto* candidate = match->second;
|
|
|
|
if (candidate->is_repeated() || candidate->is_map() ||
|
|
|
|
(candidate->cpp_type() ==
|
|
|
|
protobuf::FieldDescriptor::CPPTYPE_STRING &&
|
|
|
|
prefix == kClearMethodPrefix) ||
|
|
|
|
prefix == kSetMethodPrefix || prefix == kHasMethodPrefix) {
|
|
|
|
return absl::StrCat(field_name, "_");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return upb::generator::ResolveKeywordConflict(std::string(field_name));
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace protobuf
|
|
|
|
} // namespace google::hpb_generator
|