hatch-protobuf 0.5.0

A Hatch build plugin to generate Python files from Protocol Buffers .proto files

Hatch Protocol Buffers Generator

This Hatch plugin uses grpcio to generate Python files from Protocol Buffers .proto files.

Usage

You will need to add hatch-protobuf to your project's pyproject.toml:

[tool.hatch.build.hooks.protobuf]
dependencies = ["hatch-protobuf"]

There are a few options that can be set in the [tool.hatch.build.hooks.protobuf] section:

Key	Default	Description
generate_grpc	true	Whether to generate gRPC output files.
generate_pyi	true	Whether to generate .pyi output files. Note that these are not generated for the gRPC output. You may want to use mypy-protobuf instead.
import_site_packages	false	Adds your Python site-packages directory to --proto_path, so you can import .proto files from installed Python packages. This does not add individual .proto files in site-packages as arguments to protoc.
proto_paths	["."] or ["src"]	An array of paths to search for .proto files. Also passed as --proto_path arguments to protoc. This does not follow symlinks.
output_path	"." or "src"	The default output directory. This can be overridden on a per-generator basis for custom generators.
library_paths	[]	Similar to proto_paths, but without building the _pb2.py files, allowing imports from .protos not included in the Python site-packages directory.

Hatch-protobuf will guess whether to use "src" as the default input/output directory in a similar way to the wheel builder. If src/<NAME>/__init__.py exists, but <NAME>/__init__.py does not, it will use "src" as the default for proto_paths and output_path. Otherwise it will use "." as the default for both.

Custom generators

If you want to use custom generators (not just the python, gRPC and pyi ones built in to the version of protoc that ships with grpcio-tools), you can add them in [[tool.hatch.build.hooks.protobuf.generators]] sections. You will also need to add the plugin to the list of dependencies.

Options that can be set in generator sections:

Key	Default	Description
name	required	The name of the plugin. The argument passed to protoc will be --<name>_out.
outputs	required	A list of paths (relative to output_path). See below for more information.
output_path	same as output_path from the main protobuf config section	Where to write generated files to. This is the value passed to the --<name>_out argument.
protoc_plugin	None	The protoc plugin to use for this generator, if any. Will be passed as --plugin to protoc. This is useful for plugins that are not installed in the Python environment.
options	[]	Extra parameters to be passed to the protoc plugin using the --<name>_opt argument.

Each entry in the outputs field is a template that depends on the .proto file being processed. The string {proto_name} will be replaced with the base filename of each input .proto file, and {proto_path} will be replaced with the path (relative to the proto_paths) of the input proto files. For example, if proto_paths is set to ["src"], for the input file src/foo/bar/test.proto "{proto_name}" will expand to "test" and "{proto_path}" will expand to "foo/bar".

Here is an example using the TypeScript generator:

[[tool.hatch.build.hooks.protobuf.generators]]
name = "ts"
outputs = ["{proto_path}/{proto_name}.ts"]
output_path = "./frontend/src"
protoc_plugin = "./frontend/node_modules/.bin/protoc-gen-ts_proto"

Mypy output

The mypy-protobuf package provides mypy stub files with comments copied from the input .proto files. Here is an example of how to use it in your pyproject.toml:

[tool.hatch.build.hooks.protobuf]
dependencies = [
    "hatch-protobuf",
    "mypy-protobuf~=3.0",
]
generate_pyi = false  # we'll let mypy-protobuf do this

[[tool.hatch.build.hooks.protobuf.generators]]
name = "mypy"
outputs = ["{proto_path}/{proto_name}_pb2.pyi"]

[[tool.hatch.build.hooks.protobuf.generators]]
name = "mypy_grpc"
outputs = ["{proto_path}/{proto_name}_pb2_grpc.pyi"]

Import .proto files from site-packages

Setting include_site_packages = true causes the plugin to add your current site-packages directory as a --proto_path when running protoc, without trying to build a _pb2.py for every .proto file.

This allows your package to consume Protocol Buffer definitions from published Python packages which include both .proto and _pb2.py files.

As an example, consider a gRPC service definition which includes a Protocol Buffer message definiton from Google API common protos:

// proto/example/v1/example.proto
syntax = "proto3";

package example.v1;

import "google/api/annotations.proto";

// The greeting service definition.
service Greeter {
  // Sends a greeting
  rpc SayHello(HelloRequest) returns (HelloReply) {
    option (google.api.http) = {
      get: "/say"
    };
  }
}

// The request message containing the user's name.
message HelloRequest {
  string name = 1;
}

// The response message containing the greetings
message HelloReply {
  string message = 1;
}

In this case, the googleapis-common-protos package contains annotations.proto and a pre-built version of annotations_pb2.py, which is normally installed in a directory tree layout which can be used by both Python and protoc:

site-packages/google/api/
├── annotations_pb2.py
├── annotations.proto
├── auth_pb2.py
├── auth.proto
...

Setting include_site_packages = true makes your generated code contain imports that reference the already-built bindings, and not rebuild them:

# example/v1/example_pb2.py
from google.api import annotations_pb2 as google_dot_api_dot_annotations__pb2

If you had added the directory containing google/api/annotations.proto to this plugin's proto_paths option, this would cause it to re-build Python files for all .proto files in that directory, not just the things used for your package, and stomp all over your site-packages directory.

[!NOTE] You can only import Protocol Buffer definitions in .proto files from non-editable dependencies (ie: from ordinary packages published on PyPI or private registries).

Editable dependencies (eg: installed with pip install -e or using uv workspaces) use a different directory layout which can't be imported from by protoc.

To consume Protocol Buffer definitions from published Python packages which include only _pb2.py files, library_paths can be set to the directory containing the .proto files. For example, You can add opentelemetry-proto as a submodule in a project, and you will be able to import files from it after including the following config:

[tool.hatch.build.hooks.protobuf]
library_paths = ["./libs/opentelemetry-proto"]