internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
// Copyright 2018 The Go Authors. All rights reserved.
|
|
|
|
// Use of this source code is governed by a BSD-style
|
|
|
|
// license that can be found in the LICENSE file.
|
|
|
|
|
|
|
|
// Package fileinit constructs protoreflect.FileDescriptors from the encoded
|
|
|
|
// file descriptor proto messages. This package uses a custom proto unmarshaler
|
|
|
|
// 1) to avoid a dependency on the descriptor proto 2) for performance to keep
|
|
|
|
// the initialization cost as low as possible.
|
|
|
|
package fileinit
|
|
|
|
|
|
|
|
import (
|
|
|
|
"fmt"
|
|
|
|
"reflect"
|
|
|
|
"sync"
|
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
descopts "github.com/golang/protobuf/v2/internal/descopts"
|
2019-03-25 14:41:32 -07:00
|
|
|
pimpl "github.com/golang/protobuf/v2/internal/impl"
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
pragma "github.com/golang/protobuf/v2/internal/pragma"
|
|
|
|
pfmt "github.com/golang/protobuf/v2/internal/typefmt"
|
|
|
|
"github.com/golang/protobuf/v2/proto"
|
|
|
|
pref "github.com/golang/protobuf/v2/reflect/protoreflect"
|
2019-03-18 13:46:39 -07:00
|
|
|
preg "github.com/golang/protobuf/v2/reflect/protoregistry"
|
all: move v1 types over to the v2 repository
As a goal, v2 should not depend on v1. As another step towards that end,
we move all the types that used to be in the v1 protoapi package over to v2.
For now, we place MessageV1, ExtensionRangeV1, and ExtensionDescV1
in runtime/protoiface since these are types that generated messages will
probably have to reference forever. An alternative location could be
reflect/protoreflect, but it seems unfortunate to have to dirty the
namespace of that package with these types.
We move ExtensionFieldV1, ExtensionFieldsV1, and ExtensionFieldsOf
to internal/impl, since these are related to the implementation of a
generated message.
Since moving these types from v1 to v2 implies that the v1 protoapi
package is useless, we update all usages of v1 protoapi in the v2
repository to point to the relevant v2 type or functionality.
CL/168538 is the corresponding change to alter v1.
There will be a temporary build failure as it is not possible
to submit CL/168519 and CL/168538 atomically.
Change-Id: Ide4025c1b6af5b7f0696f4b65b988b4d10a50f0b
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/168519
Reviewed-by: Herbie Ong <herbie@google.com>
2019-03-20 18:29:32 -07:00
|
|
|
piface "github.com/golang/protobuf/v2/runtime/protoiface"
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
)
|
|
|
|
|
|
|
|
// FileBuilder construct a protoreflect.FileDescriptor from the
|
|
|
|
// raw file descriptor and the Go types for declarations and dependencies.
|
|
|
|
//
|
|
|
|
//
|
|
|
|
// Flattened Ordering
|
|
|
|
//
|
|
|
|
// The protobuf type system represents declarations as a tree. Certain nodes in
|
|
|
|
// the tree require us to either associate it with a concrete Go type or to
|
|
|
|
// resolve a dependency, which is information that must be provided separately
|
|
|
|
// since it cannot be derived from the file descriptor alone.
|
|
|
|
//
|
|
|
|
// However, representing a tree as Go literals is difficult to simply do in a
|
|
|
|
// space and time efficient way. Thus, we store them as a flattened list of
|
|
|
|
// objects where the serialization order from the tree-based form is important.
|
|
|
|
//
|
|
|
|
// The "flattened ordering" is defined as a tree traversal of all enum, message,
|
|
|
|
// extension, and service declarations using the following algorithm:
|
|
|
|
//
|
|
|
|
// def VisitFileDecls(fd):
|
|
|
|
// for e in fd.Enums: yield e
|
|
|
|
// for m in fd.Messages: yield m
|
|
|
|
// for x in fd.Extensions: yield x
|
|
|
|
// for s in fd.Services: yield s
|
|
|
|
// for m in fd.Messages: yield from VisitMessageDecls(m)
|
|
|
|
//
|
|
|
|
// def VisitMessageDecls(md):
|
|
|
|
// for e in md.Enums: yield e
|
|
|
|
// for m in md.Messages: yield m
|
|
|
|
// for x in md.Extensions: yield x
|
|
|
|
// for m in md.Messages: yield from VisitMessageDecls(m)
|
|
|
|
//
|
|
|
|
// The traversal starts at the root file descriptor and yields each direct
|
|
|
|
// declaration within each node before traversing into sub-declarations
|
|
|
|
// that children themselves may have.
|
|
|
|
type FileBuilder struct {
|
|
|
|
// RawDescriptor is the wire-encoded bytes of FileDescriptorProto.
|
|
|
|
RawDescriptor []byte
|
|
|
|
|
|
|
|
// GoTypes is a unique set of the Go types for all declarations and
|
|
|
|
// dependencies. Each type is represented as a zero value of the Go type.
|
|
|
|
//
|
|
|
|
// Declarations are Go types generated for enums and messages directly
|
|
|
|
// declared (not publicly imported) in the proto source file.
|
|
|
|
// Messages for map entries are included, but represented by nil.
|
|
|
|
// Enum declarations in "flattened ordering" come first, followed by
|
|
|
|
// message declarations in "flattened ordering". The length of each sub-list
|
|
|
|
// is len(EnumOutputTypes) and len(MessageOutputTypes), respectively.
|
|
|
|
//
|
|
|
|
// Dependencies are Go types for enums or messages referenced by
|
|
|
|
// message fields (excluding weak fields), for parent extended messages of
|
|
|
|
// extension fields, for enums or messages referenced by extension fields,
|
|
|
|
// and for input and output messages referenced by service methods.
|
|
|
|
// Dependencies must come after declarations, but the ordering of
|
|
|
|
// dependencies themselves is unspecified.
|
|
|
|
GoTypes []interface{}
|
|
|
|
|
|
|
|
// DependencyIndexes is an ordered list of indexes into GoTypes for the
|
|
|
|
// dependencies of messages, extensions, or services. There are 4 sub-lists
|
|
|
|
// each in "flattened ordering" concatenated back-to-back:
|
|
|
|
// * Extension field targets: list of the extended parent message of
|
|
|
|
// every extension. Length is len(ExtensionOutputTypes).
|
|
|
|
// * Message field dependencies: list of the enum or message type
|
|
|
|
// referred to by every message field.
|
|
|
|
// * Extension field dependencies: list of the enum or message type
|
|
|
|
// referred to by every extension field.
|
|
|
|
// * Service method dependencies: list of the input and output message type
|
|
|
|
// referred to by every service method.
|
|
|
|
DependencyIndexes []int32
|
|
|
|
|
|
|
|
// TODO: Provide a list of imported files.
|
|
|
|
// FileDependencies []pref.FileDescriptor
|
|
|
|
|
|
|
|
// TODO: Provide a list of extension types for options extensions.
|
|
|
|
// OptionDependencies []pref.ExtensionType
|
|
|
|
|
2019-03-14 16:08:22 -07:00
|
|
|
// LegacyExtensions are a list of legacy extension descriptors.
|
|
|
|
// If provided, the pointer to the v1 ExtensionDesc will be stored into the
|
|
|
|
// associated v2 ExtensionType and accessible via a pseudo-internal API.
|
|
|
|
// Also, the v2 ExtensionType will be stored into each v1 ExtensionDesc.
|
|
|
|
// If non-nil, len(LegacyExtensions) must equal len(ExtensionOutputTypes).
|
all: move v1 types over to the v2 repository
As a goal, v2 should not depend on v1. As another step towards that end,
we move all the types that used to be in the v1 protoapi package over to v2.
For now, we place MessageV1, ExtensionRangeV1, and ExtensionDescV1
in runtime/protoiface since these are types that generated messages will
probably have to reference forever. An alternative location could be
reflect/protoreflect, but it seems unfortunate to have to dirty the
namespace of that package with these types.
We move ExtensionFieldV1, ExtensionFieldsV1, and ExtensionFieldsOf
to internal/impl, since these are related to the implementation of a
generated message.
Since moving these types from v1 to v2 implies that the v1 protoapi
package is useless, we update all usages of v1 protoapi in the v2
repository to point to the relevant v2 type or functionality.
CL/168538 is the corresponding change to alter v1.
There will be a temporary build failure as it is not possible
to submit CL/168519 and CL/168538 atomically.
Change-Id: Ide4025c1b6af5b7f0696f4b65b988b4d10a50f0b
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/168519
Reviewed-by: Herbie Ong <herbie@google.com>
2019-03-20 18:29:32 -07:00
|
|
|
LegacyExtensions []piface.ExtensionDescV1
|
2019-03-14 16:08:22 -07:00
|
|
|
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
// EnumOutputTypes is where Init stores all initialized enum types
|
|
|
|
// in "flattened ordering".
|
|
|
|
EnumOutputTypes []pref.EnumType
|
|
|
|
// MessageOutputTypes is where Init stores all initialized message types
|
2019-03-19 17:04:06 -07:00
|
|
|
// in "flattened ordering". This includes slots for map entry messages,
|
|
|
|
// which are skipped over.
|
2019-03-25 14:41:32 -07:00
|
|
|
MessageOutputTypes []pimpl.MessageType
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
// ExtensionOutputTypes is where Init stores all initialized extension types
|
|
|
|
// in "flattened ordering".
|
|
|
|
ExtensionOutputTypes []pref.ExtensionType
|
|
|
|
|
2019-03-18 13:46:39 -07:00
|
|
|
// FilesRegistry is the file registry to register the file descriptor.
|
|
|
|
// If nil, no registration occurs.
|
|
|
|
FilesRegistry *preg.Files
|
|
|
|
// TypesRegistry is the types registry to register each type descriptor.
|
|
|
|
// If nil, no registration occurs.
|
|
|
|
TypesRegistry *preg.Types
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
|
|
|
|
// Init constructs a FileDescriptor given the parameters set in FileBuilder.
|
|
|
|
// It assumes that the inputs are well-formed and panics if any inconsistencies
|
|
|
|
// are encountered.
|
|
|
|
func (fb FileBuilder) Init() pref.FileDescriptor {
|
|
|
|
fd := newFileDesc(fb)
|
|
|
|
|
2019-03-18 13:46:39 -07:00
|
|
|
// Keep v1 and v2 extension descriptors in sync.
|
2019-03-14 16:08:22 -07:00
|
|
|
if fb.LegacyExtensions != nil {
|
|
|
|
for i := range fd.allExtensions {
|
|
|
|
fd.allExtensions[i].legacyDesc = &fb.LegacyExtensions[i]
|
|
|
|
fb.LegacyExtensions[i].Type = &fd.allExtensions[i]
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-03-18 13:46:39 -07:00
|
|
|
// Copy type descriptors to the output.
|
2019-03-25 14:41:32 -07:00
|
|
|
//
|
|
|
|
// While iterating over the messages, we also determine whether the message
|
|
|
|
// is a map entry type.
|
|
|
|
messageGoTypes := fb.GoTypes[len(fd.allEnums):][:len(fd.allMessages)]
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
for i := range fd.allEnums {
|
|
|
|
fb.EnumOutputTypes[i] = &fd.allEnums[i]
|
|
|
|
}
|
|
|
|
for i := range fd.allMessages {
|
2019-03-25 14:41:32 -07:00
|
|
|
if messageGoTypes[i] == nil {
|
|
|
|
fd.allMessages[i].isMapEntry = true
|
|
|
|
} else {
|
|
|
|
fb.MessageOutputTypes[i].GoType = reflect.TypeOf(messageGoTypes[i])
|
|
|
|
fb.MessageOutputTypes[i].PBType = fd.allMessages[i].asDesc().(pref.MessageType)
|
2019-03-19 17:04:06 -07:00
|
|
|
}
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
for i := range fd.allExtensions {
|
|
|
|
fb.ExtensionOutputTypes[i] = &fd.allExtensions[i]
|
|
|
|
}
|
2019-03-18 13:46:39 -07:00
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
// As a special-case for descriptor.proto,
|
|
|
|
// locally register concrete message type for the options.
|
|
|
|
if fd.Path() == "google/protobuf/descriptor.proto" && fd.Package() == "google.protobuf" {
|
|
|
|
for i := range fd.allMessages {
|
|
|
|
switch fd.allMessages[i].Name() {
|
|
|
|
case "FileOptions":
|
|
|
|
descopts.File = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "EnumOptions":
|
|
|
|
descopts.Enum = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "EnumValueOptions":
|
|
|
|
descopts.EnumValue = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "MessageOptions":
|
|
|
|
descopts.Message = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "FieldOptions":
|
|
|
|
descopts.Field = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "OneofOptions":
|
|
|
|
descopts.Oneof = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "ExtensionRangeOptions":
|
|
|
|
descopts.ExtensionRange = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "ServiceOptions":
|
|
|
|
descopts.Service = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
case "MethodOptions":
|
|
|
|
descopts.Method = messageGoTypes[i].(pref.ProtoMessage)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-03-18 13:46:39 -07:00
|
|
|
// Register file and type descriptors.
|
|
|
|
if fb.FilesRegistry != nil {
|
|
|
|
if err := fb.FilesRegistry.Register(fd); err != nil {
|
|
|
|
panic(err)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if fb.TypesRegistry != nil {
|
|
|
|
for i := range fd.allEnums {
|
|
|
|
if err := fb.TypesRegistry.Register(&fd.allEnums[i]); err != nil {
|
|
|
|
panic(err)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for i := range fd.allMessages {
|
2019-03-19 17:04:06 -07:00
|
|
|
if mt, _ := fd.allMessages[i].asDesc().(pref.MessageType); mt != nil {
|
|
|
|
if err := fb.TypesRegistry.Register(mt); err != nil {
|
|
|
|
panic(err)
|
|
|
|
}
|
2019-03-18 13:46:39 -07:00
|
|
|
}
|
|
|
|
}
|
|
|
|
for i := range fd.allExtensions {
|
|
|
|
if err := fb.TypesRegistry.Register(&fd.allExtensions[i]); err != nil {
|
|
|
|
panic(err)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
return fd
|
|
|
|
}
|
|
|
|
|
|
|
|
type (
|
|
|
|
// fileInit contains a copy of certain fields in FileBuilder for use during
|
|
|
|
// lazy initialization upon first use.
|
|
|
|
fileInit struct {
|
|
|
|
GoTypes []interface{}
|
|
|
|
DependencyIndexes []int32
|
|
|
|
}
|
|
|
|
fileDesc struct {
|
|
|
|
fileInit
|
2019-04-16 01:46:14 -07:00
|
|
|
rawDesc []byte
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
|
|
|
|
path string
|
|
|
|
protoPackage pref.FullName
|
|
|
|
|
|
|
|
fileDecls
|
|
|
|
|
|
|
|
enums enumDescs
|
|
|
|
messages messageDescs
|
|
|
|
extensions extensionDescs
|
|
|
|
services serviceDescs
|
|
|
|
|
|
|
|
once sync.Once
|
|
|
|
lazy *fileLazy // protected by once
|
|
|
|
}
|
|
|
|
fileDecls struct {
|
|
|
|
allEnums []enumDesc
|
|
|
|
allMessages []messageDesc
|
|
|
|
allExtensions []extensionDesc
|
|
|
|
}
|
|
|
|
fileLazy struct {
|
|
|
|
syntax pref.Syntax
|
|
|
|
imports fileImports
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
2019-05-12 02:27:46 -07:00
|
|
|
func (fd *fileDesc) ParentFile() pref.FileDescriptor { return fd }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (fd *fileDesc) Parent() (pref.Descriptor, bool) { return nil, false }
|
|
|
|
func (fd *fileDesc) Index() int { return 0 }
|
|
|
|
func (fd *fileDesc) Syntax() pref.Syntax { return fd.lazyInit().syntax }
|
|
|
|
func (fd *fileDesc) Name() pref.Name { return fd.Package().Name() }
|
|
|
|
func (fd *fileDesc) FullName() pref.FullName { return fd.Package() }
|
|
|
|
func (fd *fileDesc) IsPlaceholder() bool { return false }
|
2019-04-10 16:45:04 -07:00
|
|
|
func (fd *fileDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.File, fd.lazyInit().options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
reflect/protoreflect, reflect/protoregistry: move descriptor lookup to registry
Drop the protoreflect.FileDescriptor.DescriptorByName method.
Descriptor lookup will always happen through a protoregistry.Files, which
is more generally useful (it's rare that you want to find a descriptor in a
specific file, as opposed to a package which may be composed of multiple files).
Split protoregistry.Files descriptor lookup into individual per-type functions
(enum, message, extension, service), matching the preg.Types API.
Drop the ability to look up enum values, message fields, and service methods
for now. This can be easily added later if needed, and is trivial to implement
in user code. (e.g., look up the service and then consult sd.Methods.ByName().)
Change-Id: I2b3d8ef888921a8464ba1434eddab20c7d3a458e
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/172118
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-04-15 13:05:13 -07:00
|
|
|
func (fd *fileDesc) Path() string { return fd.path }
|
|
|
|
func (fd *fileDesc) Package() pref.FullName { return fd.protoPackage }
|
|
|
|
func (fd *fileDesc) Imports() pref.FileImports { return &fd.lazyInit().imports }
|
|
|
|
func (fd *fileDesc) Enums() pref.EnumDescriptors { return &fd.enums }
|
|
|
|
func (fd *fileDesc) Messages() pref.MessageDescriptors { return &fd.messages }
|
|
|
|
func (fd *fileDesc) Extensions() pref.ExtensionDescriptors { return &fd.extensions }
|
|
|
|
func (fd *fileDesc) Services() pref.ServiceDescriptors { return &fd.services }
|
|
|
|
func (fd *fileDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, fd) }
|
|
|
|
func (fd *fileDesc) ProtoType(pref.FileDescriptor) {}
|
|
|
|
func (fd *fileDesc) ProtoInternal(pragma.DoNotImplement) {}
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
|
2019-04-16 01:46:14 -07:00
|
|
|
// ProtoLegacyRawDesc is a pseudo-internal API for allowing the v1 code
|
|
|
|
// to be able to retrieve the raw descriptor.
|
|
|
|
//
|
|
|
|
// WARNING: This method is exempt from the compatibility promise and may be
|
|
|
|
// removed in the future without warning.
|
|
|
|
func (fd *fileDesc) ProtoLegacyRawDesc() []byte {
|
|
|
|
return fd.rawDesc
|
|
|
|
}
|
|
|
|
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
type (
|
|
|
|
enumDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
lazy *enumLazy // protected by fileDesc.once
|
|
|
|
}
|
|
|
|
enumLazy struct {
|
|
|
|
typ reflect.Type
|
|
|
|
new func(pref.EnumNumber) pref.Enum
|
|
|
|
|
|
|
|
values enumValueDescs
|
|
|
|
resvNames names
|
|
|
|
resvRanges enumRanges
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
enumValueDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
number pref.EnumNumber
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
2019-05-01 12:29:25 -07:00
|
|
|
func (ed *enumDesc) Descriptor() pref.EnumDescriptor { return ed }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (ed *enumDesc) GoType() reflect.Type { return ed.lazyInit().typ }
|
|
|
|
func (ed *enumDesc) New(n pref.EnumNumber) pref.Enum { return ed.lazyInit().new(n) }
|
2019-04-10 16:45:04 -07:00
|
|
|
func (ed *enumDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.Enum, ed.lazyInit().options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
func (ed *enumDesc) Values() pref.EnumValueDescriptors { return &ed.lazyInit().values }
|
|
|
|
func (ed *enumDesc) ReservedNames() pref.Names { return &ed.lazyInit().resvNames }
|
|
|
|
func (ed *enumDesc) ReservedRanges() pref.EnumRanges { return &ed.lazyInit().resvRanges }
|
|
|
|
func (ed *enumDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, ed) }
|
|
|
|
func (ed *enumDesc) ProtoType(pref.EnumDescriptor) {}
|
|
|
|
func (ed *enumDesc) lazyInit() *enumLazy {
|
|
|
|
ed.parentFile.lazyInit() // implicitly initializes enumLazy
|
|
|
|
return ed.lazy
|
|
|
|
}
|
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
func (ed *enumValueDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.EnumValue, ed.options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
func (ed *enumValueDesc) Number() pref.EnumNumber { return ed.number }
|
|
|
|
func (ed *enumValueDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, ed) }
|
|
|
|
func (ed *enumValueDesc) ProtoType(pref.EnumValueDescriptor) {}
|
|
|
|
|
|
|
|
type (
|
2019-03-19 17:04:06 -07:00
|
|
|
messageType struct{ *messageDesc }
|
|
|
|
messageDescriptor struct{ *messageDesc }
|
|
|
|
|
|
|
|
// messageDesc does not implement protoreflect.Descriptor to avoid
|
|
|
|
// accidental usages of it as such. Use the asDesc method to retrieve one.
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
messageDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
enums enumDescs
|
|
|
|
messages messageDescs
|
|
|
|
extensions extensionDescs
|
|
|
|
|
2019-03-19 17:04:06 -07:00
|
|
|
isMapEntry bool
|
|
|
|
lazy *messageLazy // protected by fileDesc.once
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
messageLazy struct {
|
|
|
|
typ reflect.Type
|
|
|
|
new func() pref.Message
|
|
|
|
|
2019-03-20 09:46:22 -07:00
|
|
|
isMessageSet bool
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
fields fieldDescs
|
|
|
|
oneofs oneofDescs
|
|
|
|
resvNames names
|
|
|
|
resvRanges fieldRanges
|
|
|
|
reqNumbers fieldNumbers
|
|
|
|
extRanges fieldRanges
|
|
|
|
extRangeOptions [][]byte
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
fieldDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
number pref.FieldNumber
|
|
|
|
cardinality pref.Cardinality
|
|
|
|
kind pref.Kind
|
|
|
|
hasJSONName bool
|
|
|
|
jsonName string
|
|
|
|
hasPacked bool
|
|
|
|
isPacked bool
|
|
|
|
isWeak bool
|
|
|
|
isMap bool
|
|
|
|
defVal defaultValue
|
|
|
|
oneofType pref.OneofDescriptor
|
|
|
|
enumType pref.EnumDescriptor
|
|
|
|
messageType pref.MessageDescriptor
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
oneofDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
fields oneofFields
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
func (md *messageDesc) options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.Message, md.lazyInit().options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
2019-03-19 17:04:06 -07:00
|
|
|
func (md *messageDesc) IsMapEntry() bool { return md.isMapEntry }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (md *messageDesc) Fields() pref.FieldDescriptors { return &md.lazyInit().fields }
|
|
|
|
func (md *messageDesc) Oneofs() pref.OneofDescriptors { return &md.lazyInit().oneofs }
|
|
|
|
func (md *messageDesc) ReservedNames() pref.Names { return &md.lazyInit().resvNames }
|
|
|
|
func (md *messageDesc) ReservedRanges() pref.FieldRanges { return &md.lazyInit().resvRanges }
|
|
|
|
func (md *messageDesc) RequiredNumbers() pref.FieldNumbers { return &md.lazyInit().reqNumbers }
|
|
|
|
func (md *messageDesc) ExtensionRanges() pref.FieldRanges { return &md.lazyInit().extRanges }
|
2019-04-10 16:45:04 -07:00
|
|
|
func (md *messageDesc) ExtensionRangeOptions(i int) pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.ExtensionRange, md.lazyInit().extRangeOptions[i])
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
func (md *messageDesc) Enums() pref.EnumDescriptors { return &md.enums }
|
|
|
|
func (md *messageDesc) Messages() pref.MessageDescriptors { return &md.messages }
|
|
|
|
func (md *messageDesc) Extensions() pref.ExtensionDescriptors { return &md.extensions }
|
|
|
|
func (md *messageDesc) ProtoType(pref.MessageDescriptor) {}
|
2019-03-19 17:04:06 -07:00
|
|
|
func (md *messageDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, md.asDesc()) }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (md *messageDesc) lazyInit() *messageLazy {
|
|
|
|
md.parentFile.lazyInit() // implicitly initializes messageLazy
|
|
|
|
return md.lazy
|
|
|
|
}
|
|
|
|
|
2019-03-20 09:46:22 -07:00
|
|
|
// IsMessageSet is a pseudo-internal API for checking whether a message
|
|
|
|
// should serialize in the proto1 message format.
|
2019-04-10 17:11:20 -07:00
|
|
|
//
|
|
|
|
// WARNING: This method is exempt from the compatibility promise and may be
|
|
|
|
// removed in the future without warning.
|
2019-03-20 09:46:22 -07:00
|
|
|
func (md *messageDesc) IsMessageSet() bool {
|
|
|
|
return md.lazyInit().isMessageSet
|
|
|
|
}
|
|
|
|
|
2019-03-19 17:04:06 -07:00
|
|
|
// asDesc returns a protoreflect.MessageDescriptor or protoreflect.MessageType
|
|
|
|
// depending on whether the message is a map entry or not.
|
|
|
|
func (mb *messageDesc) asDesc() pref.MessageDescriptor {
|
|
|
|
if !mb.isMapEntry {
|
|
|
|
return messageType{mb}
|
|
|
|
}
|
|
|
|
return messageDescriptor{mb}
|
|
|
|
}
|
2019-05-01 12:29:25 -07:00
|
|
|
func (mt messageType) Descriptor() pref.MessageDescriptor { return messageDescriptor{mt.messageDesc} }
|
|
|
|
func (mt messageType) GoType() reflect.Type { return mt.lazyInit().typ }
|
|
|
|
func (mt messageType) New() pref.Message { return mt.lazyInit().new() }
|
|
|
|
func (mt messageType) Options() pref.ProtoMessage { return mt.options() }
|
|
|
|
func (md messageDescriptor) Options() pref.ProtoMessage { return md.options() }
|
2019-03-19 17:04:06 -07:00
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
func (fd *fieldDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.Field, fd.options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
func (fd *fieldDesc) Number() pref.FieldNumber { return fd.number }
|
|
|
|
func (fd *fieldDesc) Cardinality() pref.Cardinality { return fd.cardinality }
|
|
|
|
func (fd *fieldDesc) Kind() pref.Kind { return fd.kind }
|
|
|
|
func (fd *fieldDesc) HasJSONName() bool { return fd.hasJSONName }
|
|
|
|
func (fd *fieldDesc) JSONName() string { return fd.jsonName }
|
|
|
|
func (fd *fieldDesc) IsPacked() bool { return fd.isPacked }
|
|
|
|
func (fd *fieldDesc) IsWeak() bool { return fd.isWeak }
|
|
|
|
func (fd *fieldDesc) IsMap() bool { return fd.isMap }
|
|
|
|
func (fd *fieldDesc) HasDefault() bool { return fd.defVal.has }
|
|
|
|
func (fd *fieldDesc) Default() pref.Value { return fd.defVal.get() }
|
|
|
|
func (fd *fieldDesc) DefaultEnumValue() pref.EnumValueDescriptor { return fd.defVal.enum }
|
2019-04-15 23:39:09 -07:00
|
|
|
func (fd *fieldDesc) Oneof() pref.OneofDescriptor { return fd.oneofType }
|
|
|
|
func (fd *fieldDesc) Extendee() pref.MessageDescriptor { return nil }
|
|
|
|
func (fd *fieldDesc) Enum() pref.EnumDescriptor { return fd.enumType }
|
|
|
|
func (fd *fieldDesc) Message() pref.MessageDescriptor { return fd.messageType }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (fd *fieldDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, fd) }
|
|
|
|
func (fd *fieldDesc) ProtoType(pref.FieldDescriptor) {}
|
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
func (od *oneofDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.Oneof, od.options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
func (od *oneofDesc) Fields() pref.FieldDescriptors { return &od.fields }
|
|
|
|
func (od *oneofDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, od) }
|
|
|
|
func (od *oneofDesc) ProtoType(pref.OneofDescriptor) {}
|
|
|
|
|
|
|
|
type (
|
|
|
|
extensionDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
number pref.FieldNumber
|
|
|
|
extendedType pref.MessageDescriptor
|
|
|
|
|
all: move v1 types over to the v2 repository
As a goal, v2 should not depend on v1. As another step towards that end,
we move all the types that used to be in the v1 protoapi package over to v2.
For now, we place MessageV1, ExtensionRangeV1, and ExtensionDescV1
in runtime/protoiface since these are types that generated messages will
probably have to reference forever. An alternative location could be
reflect/protoreflect, but it seems unfortunate to have to dirty the
namespace of that package with these types.
We move ExtensionFieldV1, ExtensionFieldsV1, and ExtensionFieldsOf
to internal/impl, since these are related to the implementation of a
generated message.
Since moving these types from v1 to v2 implies that the v1 protoapi
package is useless, we update all usages of v1 protoapi in the v2
repository to point to the relevant v2 type or functionality.
CL/168538 is the corresponding change to alter v1.
There will be a temporary build failure as it is not possible
to submit CL/168519 and CL/168538 atomically.
Change-Id: Ide4025c1b6af5b7f0696f4b65b988b4d10a50f0b
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/168519
Reviewed-by: Herbie Ong <herbie@google.com>
2019-03-20 18:29:32 -07:00
|
|
|
legacyDesc *piface.ExtensionDescV1
|
2019-03-14 16:08:22 -07:00
|
|
|
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
lazy *extensionLazy // protected by fileDesc.once
|
|
|
|
}
|
|
|
|
extensionLazy struct {
|
|
|
|
typ reflect.Type
|
|
|
|
new func() pref.Value
|
|
|
|
valueOf func(interface{}) pref.Value
|
|
|
|
interfaceOf func(pref.Value) interface{}
|
|
|
|
|
|
|
|
cardinality pref.Cardinality
|
|
|
|
kind pref.Kind
|
|
|
|
// Extensions should not have JSON names, but older versions of protoc
|
|
|
|
// used to set one on the descriptor. Preserve it for now to maintain
|
|
|
|
// the property that protoc 3.6.1 descriptors can round-trip through
|
|
|
|
// this package losslessly.
|
|
|
|
//
|
|
|
|
// TODO: Consider whether to drop JSONName parsing from extensions.
|
|
|
|
hasJSONName bool
|
|
|
|
jsonName string
|
|
|
|
isPacked bool
|
|
|
|
defVal defaultValue
|
2019-05-01 12:29:25 -07:00
|
|
|
enumType pref.EnumDescriptor
|
|
|
|
messageType pref.MessageDescriptor
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
2019-05-01 12:29:25 -07:00
|
|
|
func (xd *extensionDesc) Descriptor() pref.ExtensionDescriptor { return xd }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (xd *extensionDesc) GoType() reflect.Type { return xd.lazyInit().typ }
|
|
|
|
func (xd *extensionDesc) New() pref.Value { return xd.lazyInit().new() }
|
|
|
|
func (xd *extensionDesc) ValueOf(v interface{}) pref.Value { return xd.lazyInit().valueOf(v) }
|
|
|
|
func (xd *extensionDesc) InterfaceOf(v pref.Value) interface{} { return xd.lazyInit().interfaceOf(v) }
|
2019-04-10 16:45:04 -07:00
|
|
|
func (xd *extensionDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.Field, xd.lazyInit().options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
func (xd *extensionDesc) Number() pref.FieldNumber { return xd.number }
|
|
|
|
func (xd *extensionDesc) Cardinality() pref.Cardinality { return xd.lazyInit().cardinality }
|
|
|
|
func (xd *extensionDesc) Kind() pref.Kind { return xd.lazyInit().kind }
|
|
|
|
func (xd *extensionDesc) HasJSONName() bool { return xd.lazyInit().hasJSONName }
|
|
|
|
func (xd *extensionDesc) JSONName() string { return xd.lazyInit().jsonName }
|
|
|
|
func (xd *extensionDesc) IsPacked() bool { return xd.lazyInit().isPacked }
|
|
|
|
func (xd *extensionDesc) IsWeak() bool { return false }
|
|
|
|
func (xd *extensionDesc) IsMap() bool { return false }
|
|
|
|
func (xd *extensionDesc) HasDefault() bool { return xd.lazyInit().defVal.has }
|
|
|
|
func (xd *extensionDesc) Default() pref.Value { return xd.lazyInit().defVal.get() }
|
|
|
|
func (xd *extensionDesc) DefaultEnumValue() pref.EnumValueDescriptor { return xd.lazyInit().defVal.enum }
|
2019-04-15 23:39:09 -07:00
|
|
|
func (xd *extensionDesc) Oneof() pref.OneofDescriptor { return nil }
|
|
|
|
func (xd *extensionDesc) Extendee() pref.MessageDescriptor { return xd.extendedType }
|
|
|
|
func (xd *extensionDesc) Enum() pref.EnumDescriptor { return xd.lazyInit().enumType }
|
|
|
|
func (xd *extensionDesc) Message() pref.MessageDescriptor { return xd.lazyInit().messageType }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (xd *extensionDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, xd) }
|
|
|
|
func (xd *extensionDesc) ProtoType(pref.FieldDescriptor) {}
|
|
|
|
func (xd *extensionDesc) ProtoInternal(pragma.DoNotImplement) {}
|
|
|
|
func (xd *extensionDesc) lazyInit() *extensionLazy {
|
|
|
|
xd.parentFile.lazyInit() // implicitly initializes extensionLazy
|
|
|
|
return xd.lazy
|
|
|
|
}
|
|
|
|
|
2019-03-14 16:08:22 -07:00
|
|
|
// ProtoLegacyExtensionDesc is a pseudo-internal API for allowing the v1 code
|
|
|
|
// to be able to retrieve a v1 ExtensionDesc.
|
2019-04-10 17:11:20 -07:00
|
|
|
//
|
|
|
|
// WARNING: This method is exempt from the compatibility promise and may be
|
|
|
|
// removed in the future without warning.
|
|
|
|
func (xd *extensionDesc) ProtoLegacyExtensionDesc() *piface.ExtensionDescV1 {
|
|
|
|
return xd.legacyDesc
|
|
|
|
}
|
2019-03-14 16:08:22 -07:00
|
|
|
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
type (
|
|
|
|
serviceDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
lazy *serviceLazy // protected by fileDesc.once
|
|
|
|
}
|
|
|
|
serviceLazy struct {
|
|
|
|
methods methodDescs
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
methodDesc struct {
|
|
|
|
baseDesc
|
|
|
|
|
|
|
|
inputType pref.MessageDescriptor
|
|
|
|
outputType pref.MessageDescriptor
|
|
|
|
isStreamingClient bool
|
|
|
|
isStreamingServer bool
|
|
|
|
options []byte
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
func (sd *serviceDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.Service, sd.lazyInit().options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
|
|
|
func (sd *serviceDesc) Methods() pref.MethodDescriptors { return &sd.lazyInit().methods }
|
|
|
|
func (sd *serviceDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, sd) }
|
|
|
|
func (sd *serviceDesc) ProtoType(pref.ServiceDescriptor) {}
|
|
|
|
func (sd *serviceDesc) ProtoInternal(pragma.DoNotImplement) {}
|
|
|
|
func (sd *serviceDesc) lazyInit() *serviceLazy {
|
|
|
|
sd.parentFile.lazyInit() // implicitly initializes serviceLazy
|
|
|
|
return sd.lazy
|
|
|
|
}
|
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
func (md *methodDesc) Options() pref.ProtoMessage {
|
|
|
|
return unmarshalOptions(descopts.Method, md.options)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
}
|
2019-04-15 23:39:09 -07:00
|
|
|
func (md *methodDesc) Input() pref.MessageDescriptor { return md.inputType }
|
|
|
|
func (md *methodDesc) Output() pref.MessageDescriptor { return md.outputType }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (md *methodDesc) IsStreamingClient() bool { return md.isStreamingClient }
|
|
|
|
func (md *methodDesc) IsStreamingServer() bool { return md.isStreamingServer }
|
|
|
|
func (md *methodDesc) Format(s fmt.State, r rune) { pfmt.FormatDesc(s, r, md) }
|
|
|
|
func (md *methodDesc) ProtoType(pref.MethodDescriptor) {}
|
|
|
|
func (md *methodDesc) ProtoInternal(pragma.DoNotImplement) {}
|
|
|
|
|
|
|
|
type baseDesc struct {
|
|
|
|
parentFile *fileDesc
|
|
|
|
parent pref.Descriptor
|
|
|
|
index int
|
|
|
|
fullName
|
|
|
|
}
|
|
|
|
|
2019-05-12 02:27:46 -07:00
|
|
|
func (d *baseDesc) ParentFile() pref.FileDescriptor { return d.parentFile }
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
func (d *baseDesc) Parent() (pref.Descriptor, bool) { return d.parent, true }
|
|
|
|
func (d *baseDesc) Index() int { return d.index }
|
|
|
|
func (d *baseDesc) Syntax() pref.Syntax { return d.parentFile.Syntax() }
|
|
|
|
func (d *baseDesc) IsPlaceholder() bool { return false }
|
|
|
|
func (d *baseDesc) ProtoInternal(pragma.DoNotImplement) {}
|
|
|
|
|
|
|
|
type fullName struct {
|
|
|
|
shortLen int
|
|
|
|
fullName pref.FullName
|
|
|
|
}
|
|
|
|
|
|
|
|
func (s *fullName) Name() pref.Name { return pref.Name(s.fullName[len(s.fullName)-s.shortLen:]) }
|
|
|
|
func (s *fullName) FullName() pref.FullName { return s.fullName }
|
|
|
|
|
2019-04-10 16:45:04 -07:00
|
|
|
func unmarshalOptions(p pref.ProtoMessage, b []byte) pref.ProtoMessage {
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
if b != nil {
|
|
|
|
// TODO: Consider caching the unmarshaled options message.
|
2019-04-10 16:45:04 -07:00
|
|
|
p = reflect.New(reflect.TypeOf(p).Elem()).Interface().(pref.ProtoMessage)
|
internal/fileinit: generate reflect data structures from raw descriptors
This CL takes a significantly different approach to generating support
for protobuf reflection. The previous approach involved generating a
large number of Go literals to represent the reflection information.
While that approach was correct, it resulted in too much binary bloat.
The approach taken here initializes the reflection information from
the raw descriptor proto, which is a relatively dense representation
of the protobuf reflection information. In order to keep initialization
cost low, several measures were taken:
* At program init, the bare minimum is parsed in order to initialize
naming information for enums, messages, extensions, and services declared
in the file. This is done because those top-level declarations are often
relevant for registration.
* Only upon first are most of the other data structures for protobuf
reflection actually initialized.
* Instead of using proto.Unmarshal, a hand-written unmarshaler is used.
This allows us to avoid a dependendency on the descriptor proto and also
because the API for the descriptor proto is fundamentally non-performant
since it requires an allocation for every primitive field.
At a high-level, the new implementation lives in internal/fileinit.
Several changes were made to other parts of the repository:
* cmd/protoc-gen-go:
* Stop compressing the raw descriptors. While compression does reduce
the size of the descriptors by approximately 2x, it is a pre-mature
optimization since the descriptors themselves are around 1% of the total
binary bloat that is due to generated protobufs.
* Seeding protobuf reflection from the raw descriptor significantly
simplifies the generator implementation since it is no longer responsible
for constructing a tree of Go literals to represent the same information.
* We remove the generation of the shadow types and instead call
protoimpl.MessageType.MessageOf. Unfortunately, this incurs an allocation
for every call to ProtoReflect since we need to allocate a tuple that wraps
a pointer to the message value, and a pointer to message type.
* internal/impl:
* We add a MessageType.GoType field and make it required that it is
set prior to first use. This is done so that we can avoid calling
MessageType.init except for when it is actually needed. The allows code
to call (*FooMessage)(nil).ProtoReflect().Type() without fearing that the
init code will run, possibly triggering a recursive deadlock (where the
init code depends on getting the Type of some dependency which may be
declared within the same file).
* internal/cmd/generate-types:
* The code to generate reflect/prototype/protofile_list_gen.go was copied
and altered to generated internal/fileinit.desc_list_gen.go.
At a high-level this CL adds significant technical complexity.
However, this is offset by several possible future changes:
* The prototype package can be drastically simplified. We can probably
reimplement internal/legacy to use internal/fileinit instead, allowing us
to drop another dependency on the prototype package. As a result, we can
probably delete most of the constructor types in that package.
* With the prototype package significantly pruned, and the fact that generated
code no longer depend on depends on that package, we can consider merging
what's left of prototype into protodesc.
Change-Id: I6090f023f2e1b6afaf62bd3ae883566242e30715
Reviewed-on: https://go-review.googlesource.com/c/158539
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
2019-01-18 09:32:24 -08:00
|
|
|
if err := proto.Unmarshal(b, p.(proto.Message)); err != nil {
|
|
|
|
panic(err)
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return p.(proto.Message)
|
|
|
|
}
|