Hey! These docs are for version 2.11, which is no longer officially supported. Click here for the latest version, 2.18!

Protobuf and gRPC

How to generate Python from Protocol Buffers.

When your Python code imports Protobuf generated files, Pants will detect the imports and run the Protoc compiler to generate those files.


Example repository

See the codegen example repository for an example of using Protobuf to generate Python.


Benefit of Pants: generated files are always up-to-date

With Pants, there's no need to manually regenerate your code or check it into version control. Pants will ensure you are always using up-to-date files in your builds.

Thanks to fine-grained caching, Pants will regenerate the minimum amount of code required when you do make changes.

Step 1: Activate the Protobuf Python backend

Add this to your pants.toml:

backend_packages.add = [

This adds the new protobuf_source target, which you can confirm by running ./pants help protobuf_source.

To reduce boilerplate, you can also use the protobuf_sources target, which generates one protobuf_source target per file in the sources field.

protobuf_sources(name="protos", sources=["user.proto", "admin.proto"])

# Spiritually equivalent to:
protobuf_source(name="user", source="user.proto")
protobuf_source(name="admin", source="admin.proto")

# Thanks to the default `sources` value of '*.proto', spiritually equivalent to:


Enable the MyPy Protobuf plugin

The MyPy Protobuf plugin generates .pyi type stubs. If you use MyPy through Pants's check goal, this will ensure MyPy understands your generated code.

To activate, set mypy_plugin = true in the [python-protobuf] scope:

mypy_plugin = true

MyPy will use the generated .pyi type stub file, rather than looking at the .py implementation file.

Step 2: Set up the protobuf and grpcio runtime libraries

Generated Python files require the protobuf dependency for their imports to work properly. If you're using gRPC, you also need the grpcio dependency.

Add protobufβ€”and grpcio, if relevantβ€” to your project, e.g. your requirements.txt (see Third-party dependencies).


Pants will then automatically add these dependencies to your protobuf_source targets created in the next step.

Step 3: Generate protobuf_sources target

Run ./pants tailor for Pants to create a protobuf_sources target wherever you have .proto files:

$ ./pants tailor
Created src/protos/BUILD:
  - Add protobuf_sources target protos

Pants will use dependency inference for any import statements in your .proto files, which you can confirm by running ./pants dependencies path/to/file.proto. You should also see the python_requirement target for the protobuf library from the previous step.

If you want gRPC code generated for all files in the folder, set grpc=True.


If you only want gRPC generated for some files in the folder, you can use the overrides field:

        "admin.proto": {"grpc": True},
        # You can also use a tuple for multiple files.
        ("user.proto", "org.proto"): {"grpc": True},

Step 4: Confirm Python imports are working

Now, you can import the generated Python module in your Python code. For example, to import project/example/f.proto, add import project.example.f_pb2 to your code.

If you have source roots other than the repository root, remove the source root from the import. For example, src/protos/example/f.proto gets stripped to import example.f_pb2. See the below section on source roots for more info.

Pants's dependency inference will detect Python imports of Protobuf modules, which you can confirm by running ./pants dependencies path/to/file.py.

If gRPC is activated, you can also import the module with _pb2_grpc at the end, e.g. project.example.f_pb2_grpc.

from project.example.f_pb2 import HelloReply
from project.example.f_pb2_grcp import GreeterServicer


Run ./pants export-codegen :: to inspect the files

./pants export-codegen :: will run all relevant code generators and write the files to dist/codegen using the same paths used normally by Pants.

You do not need to run this goal for codegen to work when using Pants; export-codegen is only for external consumption outside of Pants.


You likely need to add empty __init__.py files

By default, Pants will generate the Python files in the same directory as the .proto file. To get Python imports working properly, you will likely need to add an empty __init__.py in the same location, and possibly in ancestor directories.

See the below section "Protobuf and source roots" for how to generate into a different directory. If you use this option, you will still likely need an empty __init__.py file in the destination directory.

Protobuf and source roots

By default, generated code goes into the same source root as the .proto file from which it was generated. For example, a file src/proto/example/f.proto will generate src/proto/example/f_pb2.py.

However, this may not always be what you want. In particular, you may not want to have to add __init__py files under src/proto just so you can import Python code generated to that source root.

You can configure a different source root for generated code by setting the python_source_root field:


Now src/proto/example/f.proto will generate src/python/example/f_pb2.py, i.e., the generated files will share a source root with your other Python code.


Set the .proto file's package relative to the source root

Remember that the package directive in your .proto file should be relative to the source root.

For example, if you have a file at src/proto/example/subdir/f.proto, you'd set its package to example.subdir; and in your Python code, from example.subdir import f_pb2.

Multiple resolves

If you're using multiple resolves (i.e. multiple lockfiles), then you may need to set the python_resolve field. protobuf_source targets only work with a single resolve, meaning, for example, that a python_source target that uses the resolve 'a' can only depend on Protobuf targets that also uses this same resolve.

By default, protobuf_source / protobuf_sources targets use the resolve set by the option [python].default_resolve. To use a different resolve, set the field python_resolve: str to one of the values from the option [python].resolves.

You must also make sure that any resolves that use codegen include python_requirement targets for the protobuf and grpcio runtime libraries from Step 2. Pants will eagerly validate this for you.

If the same Protobuf files should work with multiple resolves, you can use the
parametrize mechanism.

For example:

    # Here, we use the same version of Protobuf in both resolves. You could instead create
    # a distinct target per resolve so that they have different versions.
    resolve=parametrize("resolve-a", "resolve-b"),

    python_resolve=parametrize("resolve-a", "resolve-b")

Buf: format and lint Protobuf

Pants integrates with the Buf formatter and linter for Protobuf files.

To activate, add this to pants.toml:

backend_packages = [

Now you can run ./pants fmt and ./pants lint:

❯ ./pants lint src/protos/user.proto

Use ./pants fmt lint dir: to run on all files in the directory, and ./pants fmt lint dir:: to run on all files in the directory and subdirectories.

Temporarily disable Buf with --buf-fmt-skip and --buf-lint-skip:

❯ ./pants --buf-fmt-skip fmt ::

Only run Buf with --lint-only=buf-fmt or --lint-only=buf-lint, and --fmt-only=buf-fmt:

❯ ./pants fmt --only=buf-fmt ::