We define MessageState, which is essentially an atomically set *MessageInfo.
By nesting this as the first field in every generated message, we can
implement the reflective methods on a *MessageState when obtained by
unsafe casting a concrete message pointer as a *MessageState.
The MessageInfo held by MessageState provides additional Go type information
to interpret the memory that comes after the contents of the MessageState.
Since we are nesting a MessageState in every message,
the memory use of every message instance grows by 8B.
On average, the body of ProtoReflect grows from 133B to 202B (+50%).
However, this is offset by XXX_Methods, which is 108B and
will be removed in a future CL. Taking into account the eventual removal
of XXX_Methods, this is a net reduction of 25%.
name old time/op new time/op delta
Name/Value-4 70.3ns ± 2% 17.5ns ± 6% -75.08% (p=0.000 n=10+10)
Name/Nil-4 70.6ns ± 3% 33.4ns ± 2% -52.66% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
Name/Value-4 16.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10)
Name/Nil-4 16.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
Name/Value-4 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
Name/Nil-4 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
Change-Id: I92bd58dc681c57c92612fd5ba7fc066aea34e95a
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/185460
Reviewed-by: Damien Neil <dneil@google.com>
The encoding/prototext and encoding/protojson are implemented entirely
in terms of protobuf reflection, which side-steps this information.
Remove the hacks in the generator to special-case MessageSet.
Change-Id: I708c4636b77672545a103b7ab686f103b9dfc514
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/185240
Reviewed-by: Herbie Ong <herbie@google.com>
We modify protoc-gen-go to stop generating exported XXX fields.
The unsafe implementation is unaffected by this change since unsafe
can access fields regardless of visibility. However, for the purego
implementation, we need to respect Go visibility rules as enforced
by the reflect package.
We work around this by generating a exporter function that given
a reference to the message and the field to export, returns a reference
to the unexported field value. This exporter function is protected by
a constant such that it is not linked into the final binary in non-purego
build environment.
Updates golang/protobuf#276
Change-Id: Idf5c1f158973fa1c61187ff41440acb21c5dac94
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/185141
Reviewed-by: Damien Neil <dneil@google.com>
Hyrum's Law dictates that if we do not prevent naughty behavior,
people will rely on it. If we do not validate that the provided
file descriptor is correct today, it will be near impossible
to add proper validation checks later on.
The logic added validates that the provided file descriptor is
correct according to the same semantics as protoc,
which was reversed engineered to derive the set of rules implemented here.
The rules are unfortunately complicated because protobuf is a language
full of many non-orthogonal features. While our logic is complicated,
it is still 1/7th the size of the equivalent C++ code!
Change-Id: I6acc5dc3bd2e4c6bea6cd9e81214f8104402602a
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/184837
Reviewed-by: Damien Neil <dneil@google.com>
This is a breaking change.
The replacement is the Files.FindDescriptorByName method,
which is more flexible as it handles all descriptor types.
Change-Id: I2ccd544a7630396a2428b1d41f836c5246070912
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/183700
Reviewed-by: Damien Neil <dneil@google.com>
This change makes it such that Files now functionally registers all
descriptors in a file (not just enums, messages, extensions, and services),
but also including enum values, messages fields/oneofs, and service methods.
The ability to look up any descriptor by full name is needed to:
1) properly detect namespace conflicts on enum values
2) properly implement the relative name lookup logic in reflect/protodesc
The approach taken:
1) Assumes that a FileDescriptor has no internal name conflicts.
This will (in a future CL) be guaranteed by reflect/protodesc and
is guaranteed today by protoc for generated descriptors.
2) Observes that the only declarations that can possibly conflict
with another file are top-level declarations (i.e., enums, enum values,
messages, extensions, and services). Enum values are annoying
since they live in the same scope as the parent enum, rather than
being under the enum.
For the internal data structure of Files, we only register the top-level
declarations. This is the bare minimum needed to detect whether the file
being registered has any namespace conflicts with previously registered files.
We shift the effort to lookups, where we now need to peel off the end fragments
of a full name until we find a match in the internal registry. If a match
is found, we may need to descend into that declaration to find a nested
declaration by name.
For initialization, we modify internal/filedesc to initialize the
enum values for all top-level enums. This performance cost is offsetted
by the fact that Files.Register now avoids internally registering
nested enums, messages, and extensions.
For lookup, the cost has shifted from O(1) to O(N),
where N is the number of segments in the full name.
Top-level descriptors still have O(1) lookup times.
Nested descriptors have O(M) lookup times,
where M is the level of nesting within a single file.
Change-Id: I950163423431f04a503b6201ddcc20a62ccba017
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/183697
Reviewed-by: Damien Neil <dneil@google.com>
If we were starting from scratch, we would not have added the enum maps.
However, they already exist and see a fair amount of usage.
The effort to remove them is not worth it. Thus, remove the deprecation
warning since they are here to stay.
Note that the generated code does not refer to the generated enum maps.
One day, the linker should be able to elide them if unused by the user.
However, https://golang.org/issue/2559 would need to be resolved first.
Change-Id: Ia8b9b1812b5d8462ca2fa1d543170e4a09ff9e4f
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/183177
Reviewed-by: Damien Neil <dneil@google.com>
The internal/fileinit package is split apart into two packages:
* internal/filedesc constructs descriptors from the raw proto.
It is very similar to the previous internal/fileinit package.
* internal/filetype wraps descriptors with Go type information
Overview:
* The internal/fileinit package will be deleted in a future CL.
It is kept around since the v1 repo currently depends on it.
* The internal/prototype package is deleted. All former usages of it
are now using internal/filedesc instead. Most significantly,
the reflect/protodesc package was almost entirely re-written.
* The internal/impl package drops support for messages that do not
have a Descriptor method (pre-2016). This removes a significant amount
of technical debt.
filedesc.Builder to parse raw descriptors.
* The internal/encoding/defval package now handles enum values by name.
Change-Id: I3957bcc8588a70470fd6c7de1122216b80615ab7
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/182360
Reviewed-by: Damien Neil <dneil@google.com>
This is a breaking change in light of new methods added in CL/174918.
This removes:
Enum.Type: replacement is Descriptor
Message.Type: replacement is Descriptor and New
Change-Id: Iaa5328795407c8401ef14ed038bd5ace19d8e03b
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/174938
Reviewed-by: Damien Neil <dneil@google.com>
This is a breaking change in light of API added in CL/182497.
This removes:
Files.RangeFilesByPath: replacement is Files.FindFileByPath
Change-Id: I47bf59b37c355844984661056212953853a0db51
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/182537
Reviewed-by: Damien Neil <dneil@google.com>
Previously, we liberally permitted mutiple files to be registered that
have the same path. However, doing so causes complexity in various places
that need to assume that file paths are unique. Since unique paths are
the intention of the proto language, we strictly enforce that now.
Change-Id: Ie8fdd57c824c9809a51859cf20c4bc477b6871be
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/182497
Reviewed-by: Damien Neil <dneil@google.com>
Generate the needed infrastructure to ensure that we can statically
enforce minimum and maximum versions. This enables us to have a policy
when we release v2 where it fails to build for:
* new generated code with really old runtimes
* new runtimes with really old generated code
Change-Id: Ib699ad62c06dff8f9285806394a741c18db00288
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/178546
Reviewed-by: Damien Neil <dneil@google.com>
Instead of accepting a concrete protoregistry.Types type,
accept an interface that provides the necessary functionality
to perform the serialization.
The advantages of this approach:
* There is no need for complex logic to allow a Parent or custom
Resolver on the protoregistry.Types type.
* Users can pass their own custom resolver implementations directly
to the serialization functions.
* This is a more principled approach to plumbing custom resolvers
than the previous approach of overloading behavior on the concrete
Types type.
The disadvantages of this approach:
* A pointer to a concrete type is 8B, while an interface is 16B.
However, the expansion of the {Marshal,Unmarshal}Options structs
should be a concern solved separately from how to plumb custom resolvers.
* The resolver interfaces as defined today may be insufficient to
provide functionality needed in the future if protobuf expands its
feature set. For example, let's suppose the Any message permits
directly representing a enum by name. This would require the ability
to lookup an enum by name. To support that hypothetical need,
we can document that the serializers type-assert the provided Resolver
to a EnumTypeResolver and use that if possible. There is some loss
of type safety with this approach, but provides a clear path forward.
Change-Id: I81ca80e59335d36be6b43d57ec8e17abfdfa3bad
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/177044
Reviewed-by: Damien Neil <dneil@google.com>
The name MessageType is easily confused with protoreflect.MessageType.
Rename it as MessageInfo, which follows the pattern set by v1,
where the equivalent data structure is called InternalMessageInfo.
Change-Id: I535956e1f7c6e9b07e9585e889d5e93388d0d2ce
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/178478
Reviewed-by: Damien Neil <dneil@google.com>
Temporarily remove go.mod, since we can't generate an accurate one until
the corresponding v1 change is submitted.
Change-Id: I1e1ad97f2b455e33f61ffaeb8676289795e47e72
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/177000
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Added API:
FieldDescriptor.IsExtension
FieldDescriptor.IsList
FieldDescriptor.MapKey
FieldDescriptor.MapValue
FieldDescriptor.ContainingOneof
FieldDescriptor.ContainingMessage
Deprecated API (to be removed in subsequent CL):
FieldDescriptor.Oneof
FieldDescriptor.Extendee
These methods help cleanup several common usage patterns.
Change-Id: I9a3ffabc2edb2173c536509b22f330f98bba7cf3
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/176977
Reviewed-by: Damien Neil <dneil@google.com>
Added methods:
Enum.Descriptor
Message.Descriptor
EnumType.Descriptor
MessageType.Descriptor
ExtensionType.Descriptor
Message.New
All functionality is switched over to use those methods instead of
implicitly relying on the fact that {Enum,Message}Type implicitly
implement the associated descriptor interface.
This CL does not yet remove {Enum,Message}.Type or prevent
{Enum,Message,Extension}Type from implementating a descriptor.
That is a subsequent CL.
The Message.New method is also added to replace functionality
that will be lost when the Type methods are removed.
Change-Id: I7fefde1673bbd40bfdac489aca05cec9a6c98eb1
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/174918
Reviewed-by: Damien Neil <dneil@google.com>
Reviewed-by: Herbie Ong <herbie@google.com>
The protobuf type system uses the word "descriptor" instead of "type".
We should avoid the "type" verbage when we aren't talking about Go types.
The old names are temporarily kept around for compatibility reasons.
Change-Id: Icc99c913528ead011f7a74aa8399d9c5ec6dc56e
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/172238
Reviewed-by: Herbie Ong <herbie@google.com>
Reviewed-by: Damien Neil <dneil@google.com>
Using a named fields gives us the flexibility to change the underlying
representation of special fields without needing to regenerate user code.
We add a named type for ExtensionFields, UnknownFields, and SizeCache.
Change-Id: I107cf82899850ea76665310ce79def60f0f7ab97
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/172402
Reviewed-by: Damien Neil <dneil@google.com>
Lower-casing the FileDescriptor variable provides for shorter and
more readable global variable names. This has a very minor benefit
that the read-only data section of a binary is slightly smaller
since function names do end up in the final binary.
Change-Id: I077364905e5c9adea69873b3ea580fddc68b9eb8
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/172119
Reviewed-by: Herbie Ong <herbie@google.com>
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>
Go1.12 on GoLLVM with LLVM 9.0.0 has a bug where the simpler but
equivalent expression results in memory corruption.
Revert to the more complex expression to avoid breaking GoLLVM and
it is not worth waiting for a GoLLVM fix.
Change-Id: I8b42965ca9bd333deb8693021e7b22f966e13521
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/171463
Reviewed-by: Damien Neil <dneil@google.com>
Allow message implementations to provide optimized versions of standard
operations. Generated messages now include a ProtoReflectMethods method,
returning a protoiface.Methods struct containing pointers to assorted
optional functions.
The Methods struct also includes a Flags field indicating support for
optional features such as deterministic marshaling.
Implementation of the fast paths (and tests) will come in later CLs.
Change-Id: Idd1beed0ecf43ec5e5e7b8da2ee1e08d3ce32213
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/170340
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Minor changes:
* Use x as the receiver since "e" and "m" are meaningless in the presence
of user-defined enum and message names.
* Consistently keep enum methods together, rather awkwardly split apart
by the value maps.
Change-Id: I68e5666efb56ac7a4d062fb223b9f826dc72aba9
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/170357
Reviewed-by: Herbie Ong <herbie@google.com>
This reduces the init-time cost slightly since the GZIP'd
raw descriptor is constructed lazily on demand.
Change-Id: I482c6a2201b8786e425d7dee5612fdfd60ab1500
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/169917
Reviewed-by: Herbie Ong <herbie@google.com>
This removes yet another set of dependencies of v2 on v1.
The only remaining dependency are in the _test.go files,
primarily for proto.Equal.
Changes made:
* cmd/protoc-gen-go no longer generates any functionality that depends
on the v1 package, and instead only depends on v2.
* internal/fileinit.FileBuilder.MessageOutputTypes is switched from
protoreflect.MessageType to protoimpl.MessageType since the
implementation must be fully inialized before registration occurs.
* The test for internal/legacy/file_test.go is switched to a legacy_test
package to avoid a cyclic dependency.
This requires Load{Enum,Message,File}Desc to be exported.
Change-Id: I43e2fe64cff4eea204258ce11e791aca5eb6e569
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/169298
Reviewed-by: Damien Neil <dneil@google.com>
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>
The v1 registration leaks the message types out to the proto package.
When doing that, it must ensure that the reflection data structures
for those types are properly initialized first. We achieve that by
doing v1 registration at the end of the reflection init function.
Change-Id: If6df18df59d05bad50ff39c2eff6beb19e7466cc
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/168348
Reviewed-by: Damien Neil <dneil@google.com>
The EnumName, UnmarshalJSONEnum, and CompressGZIP helpers currently live
in v1 protoapi, which would cause all generated messages to depend on v1.
In an effort to break the dependency of v2 on v1, we move these helper
functions to v2 (and re-written to take advantage of protobuf reflection).
These helpers are unfortunate, but we cannot eliminate the functionality
that they implement since they are exposed in the publicly generated API.
Since EnumName does not rely on the enum maps, it removes another dependency
on those variables. Eventually, we can get to the point where these variables
(though declared) are not linked into the binary if the user does not use them.
Also, we rely on the v1 proto package for registration instead of v1 protoapi.
This may re-introduce a cyclic dependency on descriptor proto again in the
future, but the better approach is to just start registering with v2.
Change-Id: Id755585a7a1df14e4a6a2dfa650df221a3c153fb
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/167921
Reviewed-by: Damien Neil <dneil@google.com>
Unfortunately a good amount of code uses pointer comparisons on the
v1 ExtensionDesc to determine exactly which extension field is set,
rather than checking whether the extension descriptor semantically
describes the field that they are interested in.
To preserve this behavior in v1, we need a 1:1 mapping between
a v2 ExtensionType and a specific v1 ExtensionDesc.
Change-Id: I852b3cefb4585bd656e48e5adad6cc28795d02df
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/167759
Reviewed-by: Damien Neil <dneil@google.com>
The prototype package was initially used by generated reflection support,
but has now been replaced by internal/fileinit.
Eventually, this functionality should be deleted and re-written in terms
of other components in the repo.
Usages that prototype currently provides (but should be moved) are:
* Constructing standalone messages and enums, which is behavior we should
provide in reflect/protodesc. The google.protobuf.{Enum,Type} are well-known
proto messages designed for this purpose.
* Constructing placeholder files, enums, and messages.
* Consructing protoreflect.{Message,Enum,Extension}Types, which are protobuf
descriptors with associated Go type information.
Change-Id: Id7dbefff952682781b439aa555508c59b2629f9e
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/167383
Reviewed-by: Damien Neil <dneil@google.com>
Ensure that the init funcs for files within a Go package run in the
dependency order of the source .proto files. That is, if a.proto and b.proto
are part of the same Go package, and a.proto imports b.proto, then b.pb.go's
init funcs must run before a.pb.go's.
Change-Id: I0e86ff22e5c4cab9df7a73fe4805390fadd34b0d
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/166419
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Reviewed-by: Herbie Ong <herbie@google.com>
Using arrays in the generated reflection information adds unnecessary
eq and hash functions being added to the package. Change to slices
to reduce bloat.
Change-Id: I1a4f6d59021644d93dd6c24679b9233141e89a75
Reviewed-on: https://go-review.googlesource.com/c/164640
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Create a single binary for handling generation of protos.
This replaces previous logic spread throughout the repo in:
* regenerate.bash
* cmd/protoc-gen-go/golden_test.go
* cmd/protoc-gen-go-grpc/golden_test.go
* (indirectly) internal/protogen/goldentest
One of the problems with the former approaches is that they relied on
a version of protoc that was specific to a developer's workstation.
This meant that the result of generation was not hermetic.
To address this, we rely on the hard-coded version of protobuf specified
in the test.bash script.
A summary of changes in this CL are:
* The internal_gengo.GenerateFile and internal_gengogrpc.GenerateFile
functions are unified to have consistent signatures. It seems that the
former accepted a *protogen.GeneratedFile to support v1 where gRPC code
was generated into the same file as the base .pb.go file. However, the
same functionality can be achieved by having the function return
the generated file object.
* The test.bash script patches the protobuf toolchain to have properly
specified go_package options in each proto source file.
* The test.bash script accepts a "-regenerate" argument.
* Add generation for the well-known types. Contrary to how these were
laid out in the v1 repo, all the well-known types are placed in the
same Go package.
* Add generation for the conformance proto.
* Remove regenerate.bash
* Remove internal/protogen
* Remove cmd/protoc-gen-go/golden_test.go
* Remove cmd/protoc-gen-go-grpc/golden_test.go
* Add cmd/protoc-gen-go/annotation_test.go
Change-Id: I4a1a97ae6f66e2fabcf4e4d292c95ab2a2db0248
Reviewed-on: https://go-review.googlesource.com/c/164477
Reviewed-by: Damien Neil <dneil@google.com>
Use the full path (including the extension) for the generation of
the per-file variable name. Several reasons for this:
* The current logic is buggy in the case where pathType == pathTypeImport
since the prefix variable will be mangled with the Go import path.
* The extension is technically part of the path.
Thus, "path/to/foo.proto" and "path/to/foo.protodevel" are two
distinctly different imports.
* Style-wise, it subjectively looks better. Rather than being a mixture
of camelCase and snake_case, it is all snake_case for the common case:
before: ProtoFile_google_protobuf_any
after: File_google_protobuf_any_proto
* Since the extension is almost always ".proto", this results in a
suffix of "_proto", which provides an additional layer of protection
against possible name conflicts. The previous approach could possibly
have a conflict between "Foo.proto" and a message named ProtoFile
with a sub-message called Foo.
Also, use the per-file variable name for the raw descriptor variables
instead of the hashed version.
Change-Id: Ic91e326b7593e5985cee6ececc60539c27fe32fe
Reviewed-on: https://go-review.googlesource.com/c/164379
Reviewed-by: Damien Neil <dneil@google.com>
Calling a helper function directly should reduce binary bloat slightly.
Change-Id: I6068dc4cd00c8d90d2e6e6d99633b81388bc8781
Reviewed-on: https://go-review.googlesource.com/c/164679
Reviewed-by: Damien Neil <dneil@google.com>
For a file "foo/bar.proto", put the FileDescriptor in "ProtoFile_foo_bar"
rather than "Bar_fileDescriptor".
Avoid name clashes when a package contains "a/foo.proto" and "b/foo.proto".
Don't camelcase the filename: These vars weren't fully camelcased to begin
with, and leaving the filename relatively unchanged is clearer and more
predictable.
Move "ProtoFile" from the end of the var name to the start, so that vars
will sort better in packages with multiple descriptors.
These changes do add a chance of name collision when the input filename
begins with an uppercase letter: Foo.proto becomes "ProtoFile_Foo", which
could be the result of camelcasing "proto_file.foo". The readability
benefits seem worth it.
Change-Id: If27d3a0d7b5bf3535aa1607a8579eb057c74d2dc
Reviewed-on: https://go-review.googlesource.com/c/163199
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Reviewed-by: Herbie Ong <herbie@google.com>
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>
A "_ProtoFile" suffix can potentially conflict with a sub-message named
"ProtoFile" nested within a message that matches the camel-cased
form of the basename of the .proto source file.
Avoid unlikely conflicts and rename this to use a "_protoFile" suffix,
which can never conflict except with an enum value that is also named
"protoFile" (which is a violation of the style guide).
Change-Id: Ie9d22f9f741a63021b8f76906b20c6c2f599885b
Reviewed-on: https://go-review.googlesource.com/c/157218
Reviewed-by: Damien Neil <dneil@google.com>
Most usages of New actually prefer to interact with the reflective view
rather than the native Go type. Thus, change New to return that instead.
This parallels reflect.New, which returns the reflective view
(i.e., reflect.Value) instead of native type (i.e., interface{}).
We make the equivalent change to KnownFields.NewMessage, List.NewMessage,
and Map.NewMessage for consistency.
Since this is a subtle change where the type system will not always
catch the changed type, this change was made by both changing the type
and renaming the function to NewXXX and manually looking at every usage
of the the function to ensure that the usage correctly operates
on either the native Go type or the reflective view of the type.
After the entire codebase was cleaned up, a rename was performed to convert
NewXXX back to New.
Change-Id: I153fef627b4bf0a427e4039ce0aaec52e20c7950
Reviewed-on: https://go-review.googlesource.com/c/157077
Reviewed-by: Damien Neil <dneil@google.com>
Drop the protoreflect.ProtoEnum type (containing a single method
returning a protoreflect.Enum) and make generated enum types
directly implement protoreflect.Enum instead.
Messages have a two-level type split (ProtoMessage and Message) to
minimize conflicts between reflection methods and field names. Enums
need no such split, since enums do not have fields and therefore have
no source of conflicts.
Change-Id: I2b6222e9404253e6bfef2217859e1b760ffcd29b
Reviewed-on: https://go-review.googlesource.com/c/156902
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Reviewed-by: Damien Neil <dneil@google.com>
The first commit of protoregistry only added a registry for files.
However, a separate type of registry is needed to provide a mapping between
protobuf names and actual Go types representing those names.
Additional high-level API:
var GlobalTypes = new(Types)
type Type interface{ ... }
type Types struct{ ... }
func NewTypes(...Type) *Types
func (*Types) Register(...Type) error
func (*Types) FindEnumByName(pref.FullName) (pref.EnumType, error)
func (*Types) FindMessageByName(pref.FullName) (pref.MessageType, error)
func (*Types) FindMessageByURL(string) (pref.MessageType, error)
func (*Types) FindExtensionByName(pref.FullName) (pref.ExtensionType, error)
func (*Types) FindExtensionByNumber(pref.FullName, pref.FieldNumber) (pref.ExtensionType, error)
func (*Types) RangeEnums(func(pref.EnumType) bool)
func (*Types) RangeMessages(func(pref.MessageType) bool)
func (*Types) RangeExtensions(func(pref.ExtensionType) bool)
func (*Types) RangeExtensionsByMessage(pref.FullName, func(pref.ExtensionType) bool)
Change-Id: I0d07705801684a1eb5853bcd05fcce12598a0047
Reviewed-on: https://go-review.googlesource.com/c/131345
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
This change was created by running:
git ls-files | xargs sed -i "s|google.golang.org/proto|github.com/golang/protobuf/v2|g"
This change is *not* an endorsement of "github.com/golang/protobuf/v2" as the
final import path when the v2 API is eventually released as stable.
We continue to reserve the right to make breaking changes as we see fit.
This change enables us to host the v2 API on a repository that is go-gettable
(since go.googlesource.com is not a known host by the "go get" tool;
and google.golang.org/proto was just a stub URL that is not currently served).
Thus, we can start work on a forked version of the v1 API that explores
what it would take to implement v1 in terms of v2 in a backwards compatible way.
Change-Id: Ia3ebc41ac4238af62ee140200d3158b53ac9ec48
Reviewed-on: https://go-review.googlesource.com/136736
Reviewed-by: Damien Neil <dneil@google.com>
Implement NewFileFromDescriptorProto, which constructs a
protoreflect.FileDescriptor from a provided descriptor.FileDescriptorProto.
Some other minor changes:
* Allow calling Find and Range methods on nil *protoregistry.Files to match
the behavior of maps where index operations are permitted, but store panics.
* Switch protoregistry test to be protoregistry_test to avoid cyclic dependency.
Change-Id: I5536901ef5096014a3421e63bc4a9dfcad335ea4
Reviewed-on: https://go-review.googlesource.com/132455
Reviewed-by: Damien Neil <dneil@google.com>
