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>
This change preserves consistency with CL/157077,
where New returns a value closer to the reflective type.
Change-Id: I85bfdae24e1ce1a10c3c7b939420fa1043bff743
Reviewed-on: https://go-review.googlesource.com/c/157078
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>
Remove the Mutable methods from KnownFields, List, and Map, replacing
them with methods which return a new, empty message value without adding
that value to the collection.
The new API is simpler, since it clearly applies only to message values,
and more orthogonal, since it provides a way to create a value without
mutating the collection. This latter point is particularly useful in
map deserialization, where the key may be unknown at the time the value
is deserialized.
Drop the Mutable interface, since it is no longer necessary.
Change-Id: Ic5f3d06a2aa331a5d5cd2b4e670a3dba4a74f77c
Reviewed-on: https://go-review.googlesource.com/c/153278
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
Similar to how generated messages allow you to call Get methods on a
nil pointer, we permit similar functionality when protobuf reflection
is used on a nil pointer.
Change-Id: Ie2f596d39105c191073b42d7d689525c3b715240
Reviewed-on: https://go-review.googlesource.com/c/152021
Reviewed-by: Damien Neil <dneil@google.com>
The impl package currently supports wrapping legacy v1 enums and messages
so that they implement the v2 reflective APIs. This functionality is necessary
for v1 and v2 to interoperate. However, the existence of this functionality
presents several problems:
* A significant portion of the complexity in impl is for legacy wrapping.
* This complexity is linked into a Go binary even if all the other messages
in the binary natively support v2 reflection.
* It presents a cyclic dependency when trying to generate descriptor proto.
Suppose you are generating descriptor.proto. The generated code would want to
depend on the impl package because impl is the runtime implementation for
protobuf messages. However, impl currently depends depends on descriptor in
order to wrap legacy enum and messages since it needs the ability to dynamically
create new protobuf descriptor types. In the case of descriptor.proto, it would
presumably be generated with native reflection support, so the legacy wrapping
logic is unneccessary.
To break the dependency of impl on descriptor, we move the legacy support logic
to a different package and instead add hooks in impl so that legacy support could
be dynamically registered at runtime. This is dependency injection.
Change-Id: I01a582908ed5629993f6699e9bf2f4bee93857a4
Reviewed-on: https://go-review.googlesource.com/c/151877
Reviewed-by: Herbie Ong <herbie@google.com>
Add a Proto prefix before the Unwrap method to reduce the probability that
it would ever conflict with a method of the same name that a
custom implementation of Enum, Message, List, or Map may have.
Change-Id: I628bf8335583f2747ab4589f3e6ff82e4501ce98
Reviewed-on: https://go-review.googlesource.com/c/151817
Reviewed-by: Herbie Ong <herbie@google.com>
The new v1 protoapi package enables:
* Referencing types in the protoapi package instead of protoV1, which further
reduces the number of situations where we need to depend on protoV1.
This is for the goal of eventually breaking all cases where the v2 implementation
relies on v1, so that in the near future, proto v1 can rely on proto v2 instead.
* Removes the need for legacy_extension_hack.go since that functionality has now
been exported into the protoapi package.
Change-Id: If71002d9ec711bfabfe494636829df9abf19e23e
Reviewed-on: https://go-review.googlesource.com/c/151403
Reviewed-by: Herbie Ong <herbie@google.com>
Clearly specify that Get on an unpopulated field:
* returns the default value for scalars
* returns a mutable (but empty) List for repeated fields
* returns a mutable (but empty) Map for map fields
* returns an invalid value for message fields
The difference in semantics between List+Maps and Messages is because
protobuf semantics provide no distinction between an unpopulated and empty list
or map. On the other hand, there is a semantic difference between an unpopulated
message and an empty message.
Default values for scalars is trivial to implement with FieldDescriptor.Default.
A mutable, but empty List and Map is easy to implement for known fields since
known fields are generated as a slice or map field in a struct.
Since struct fields are addressable, the implementation can just return a
reference to the slice or map.
Repeated, extension fields are a little more tricky since extension fields
are implemented under the hood as a map[FieldNumber]Extension.
Rather than allocating an empty list in KnownFields.Get upon first retrieval
(which presents a race), delegate the work to ExtensionFieldTypes.Register,
which must occur before any Get operation. Register is not a concurrent-safe
operation, so that is an excellent time to initilize empty lists.
The implementation of extensions will need to be careful that Clear on a repeated
field simply truncates it zero instead of deleting the object.
For unpopulated messages, we return an invalid value, instead of the prior
behavior of returning a typed nil-pointer to the Go type for the message.
The approach is problematic because it assumes that
1) all messages are always implemented on a pointer reciever
2) a typed nil-pointer is an appropriate "read-only, but empty" message
These assumptions are not true of all message types (e.g., dynamic messages).
Change-Id: Ie96e6744c890308d9de738b6cf01d3b19e7e7c6a
Reviewed-on: https://go-review.googlesource.com/c/150319
Reviewed-by: Damien Neil <dneil@google.com>
The terminology Vector does not occur in protobuf documentation at all,
so we should rename the Go use of the term to something more recognizable.
As such, all instances that match the regexp "[Vv]ect(or)?" were replaced.
The C++ documentation uses the term "Repeated", which is a reasonable name.
However, the term became overloaded in 2014, when maps were added as a feature
and implementated under the hood as repeated fields. This is confusing as it
means "repeated" could either refer to repeated fields proper (i.e., explicitly
marked with the "repeated" label in the proto file) or map fields. In the case
of the C++ reflective API, this is not a problem since repeated fields proper
and map fields are interacted with through the same RepeatedField type.
In Go, we do not use a single type to handle both types of repeated fields:
1) We are coming up with the Go protobuf reflection API for the first time
and so do not need to piggy-back on the repeated fields API to remain backwards
compatible since no former usages of Go protobuf reflection exists.
2) Map fields are commonly represented in Go as the Go map type, which do not
preserve ordering information. As such it is fundamentally impossible to present
an unordered map as a consistently ordered list. Thus, Go needs two different
interfaces for lists and maps.
Given the above situation, "Repeated" is not a great term to use since it
refers to two different things (when we only want one of the meanings).
To distinguish between the two, we'll use the terms "List" and "Map" instead.
There is some precedence for the term "List" in the protobuf codebase
(e.g., "getRepeatedInt32List").
Change-Id: Iddcdb6b78e1e60c14fa4ca213c15f45e214b967b
Reviewed-on: https://go-review.googlesource.com/c/149657
Reviewed-by: Damien Neil <dneil@google.com>
Implement support for extension fields for messages that use the v1
data structures for extensions. The legacyExtensionFields type wraps a
v1 map to implement the v2 protoreflect.KnownFields interface.
Working on this change revealed a bug in the dynamic construction of
message types for protobuf messages that had cyclic dependencies (e.g.,
message Foo has a sub-field of message Bar, and Bar has a sub-field of Foo).
In such a situation, a deadlock occurs because initialization code depends on
the very initialization code that is currently running. To break these cycles,
we make some systematic changes listed in the following paragraphs.
Generally speaking, we separate the logic for construction and wrapping,
where constuction does not recursively rely on dependencies,
while wrapping may recursively inspect dependencies.
Promote the MessageType.MessageOf method as a standalone MessageOf function
that dynamically finds the proper *MessageType to use. We make it such that
MessageType only supports two forms of messages types:
* Those that fully implement the v2 API.
* Those that do not implement the v2 API at all.
This removes support for the hybrid form that was exploited by message_test.go
In impl/message_test.go, switch each message to look more like how future
generated messages will look like. This is done in reaction to the fact that
MessageType.MessageOf no longer exists.
In value/{map,vector}.go, fix Unwrap to return a pointer since the underlying
reflect.Value is addressable reference value, not a pointer value.
In value/convert.go, split the logic apart so that obtaining a v2 type and
wrapping a type as v2 are distinct operations. Wrapping requires further
initialization than simply creating the initial message type, and calling it
during initial construction would lead to a deadlock.
In protoreflect/go_type.go, we switch back to a lazy initialization of GoType
to avoid a deadlock since the user-provided fn may rely on the fact that
prototype.GoMessage returned.
Change-Id: I5dea00e36fe1a9899bd2ac0aed2c8e51d5d87420
Reviewed-on: https://go-review.googlesource.com/c/148826
Reviewed-by: Herbie Ong <herbie@google.com>
The current logic in this file is temporary and will be removed.
Rename the file as legacy_extension_hack.go so that the future
legacy_extension.go file can contain the actual legacy code that cannot be
removed anytime in the near future.
Change-Id: I7edd9d682836379c4c97d18ea2c2a2e3b15db5a2
Reviewed-on: https://go-review.googlesource.com/c/147918
Reviewed-by: Herbie Ong <herbie@google.com>
The unknown fields in legacy messages is split across the XXX_unrecognized
field and also the XXX_InternalExtensions field. Implement support for
wrapping both fields and presenting it as if it were a unified set of
unknown fields.
Change-Id: If274fae2b48962520edd8a640080b6eced747684
Reviewed-on: https://go-review.googlesource.com/c/146517
Reviewed-by: Damien Neil <dneil@google.com>
