Please open an issue and explain the problem.
Just start a discussion.
Building and testing this project requires node
(14.5.0), npm
, protoc
(3.12.3),
make
, bazel
, git
, go
.
go
is required for the Twirp transport client compatibility test suiteprotoc
is required for test fixturesgit
andbazel
are required to checkout and compile the protobuf conformance test suite
The entire project can be built by running make
, but if you do not have all required
tooling installed, you can still build all of the TypeScript packages.
Not all packages are public, or even JavaScript projects. There are also examples,
benchmarks and fixtures in the packages/
directory. They can be built with their
respective Makefiles.
The runtime is tested without generating any code. For example, to test the
ReflectionJsonReader
, we simply pass reflection field information to the
constructor of a ReflectionJsonReader
, and can test whether it reads and
writes JSON as expected.
This even works with the MessageType
, which implements the public API of
each protobuf message in TypeScript. We just don't have a TypeScript interface
for the message in question.
To test only the runtime, run:
cd packages/runtime
npm install
make
The make
compiles TypeScript to Javascript and runs all test cases
(in spec/
) in nodejs (using jasmine) and in a headless browser (using karma
and jasmine).
Some test cases use test fixtures from packages/test-fixtures
.
packages/test-fixtures/index.ts
exports a FixtureRegistry
. This registry
provides access to a collection of message type fixtures.
A message type fixture is basically a set of test data for a specific protobuf message. It has a type name (the qualified protobuf message type name) and reflection field information. It can also provide message instances.
The only purpose of the fixtures is to isolate test data from test code to make test cases more readable.
packages/test-fixtures
also contains a large number of .proto files. These
files are used to test code generation and functionality of the generated code.
For each fixture, there should also be a corresponding .proto file. This makes it possible to compare the field information generated by the plugin to the field information in the fixture data, asserting that the plugin generates expected data.
packages/test-fixtures/all.descriptorset
is a binary protobuf message of type
FileDescriptorSet
(defined in google/protobuf/descriptor.proto
). It contains
file descriptors for all .proto files in test-fixtures
. The descriptor set
is generated using protoc (see packages/plugin/Makefile
and
packages/plugin-framework/Makefile
). A FileDescriptorSet
can be used to
run the plugin during testing without invoking protoc.
All protobuf plugins work with CodeGeneratorRequest
and CodeGeneratorResponse
messages (defined in google/protobuf/compiler/plugin.proto
).
The protocol buffer compiler parses .proto files and creates a CodeGeneratorRequest
,
then passes it to a plugin. Because invoking protoc during testing is difficult,
we generate a FileDescriptorSet
in packages/test-fixtures/all.descriptorset
ahead of time. The FileDescriptorSet
contains all information necessary to
create the CodeGeneratorRequest
s we need for testing the plugin.
packages/plugin
and packages/plugin-framework
are both tested using the
FileDescriptorSet
. In spec/helpers.ts
, the function getFixtureCodeGeneratorRequest
can be used to create a CodeGeneratorRequest
from the fixture file descriptors.
The plugin itself has only very basic test coverage. We generate TypeScript
(in memory) for all .proto files in packages/test-fixture
and compile the
generated code using the TypeScript Compiler API, checking for static errors.
Adding a feature or fixing a bug in the plugin can cumbersome. Instead of
building the plugin and running it with protoc, you can let a test case spit
out the generated code for you. spec/protobufts-plugin.spec.ts
contains the
necessary code to do this (commented out). If you enable the code, you can simply
run make test
after your change to see the generated code. There should probably
be a separate script for this, but at the moment, we do not have one. Please
remember not to commit unintentional changes to the spec file.
The plugin generates speed optimized methods as well as custom method for well-known types. This code is not part of the runtime and can only be tested by testing the actual generated code.
packages/test-generated/
is responsible to test the generated code.
See the README.md for details.