Python GN templates

The Python build is implemented with GN templates defined in pw_build/python.gni. See the .gni file for complete usage documentation.


The main Python template is pw_python_package. Each pw_python_package target represents a Python package. As described in pw_python_package targets, each pw_python_package expands to several subtargets. In summary, these are:

  • <name> - Represents the files themselves

  • <name>.lint - Runs static analysis

  • <name>.tests - Runs all tests for this package

  • <name>.install - Installs the package

  • <name>.wheel - Builds a Python wheel

GN permits using abbreviated labels when the target name matches the directory name (e.g. //foo for //foo:foo). For consistency with this, Python package subtargets are aliased to the directory when the target name is the same as the directory. For example, these two labels are equivalent:


The actions in a pw_python_package (e.g. installing packages and running Pylint) are done within a single GN toolchain to avoid duplication in multi-toolchain builds. This toolchain can be set with the pw_build_PYTHON_TOOLCHAIN GN arg, which defaults to $dir_pw_build/python_toolchain:python.


  • setup - List of setup file paths ( or pyproject.toml & setup.cfg), which must all be in the same directory.

  • generate_setup: As an alternative to setup, generate setup files with the keywords in this scope. name is required. This follows the same structure as a setup.cfg file’s declarative config For example:

    generate_setup = {
      metadata = {
        name = "a_nifty_package"
        version = "1.2a"
      options = {
        install_requires = [ "a_pip_package" ]
  • sources - Python sources files in the package.

  • tests - Test files for this Python package.

  • python_deps - Dependencies on other pw_python_packages in the GN build.

  • python_test_deps - Test-only pw_python_package dependencies.

  • other_deps - Dependencies on GN targets that are not pw_python_packages.

  • inputs - Other files to track, such as package_data.

  • proto_library - A pw_proto_library target to embed in this Python package. generate_setup is required in place of setup if proto_library is used. See Adding Python proto modules to an existing package.

  • static_analysis List of static analysis tools to run; "*" (default) runs all tools. The supported tools are "mypy" and "pylint".

  • pylintrc - Optional path to a pylintrc configuration file to use. If not provided, Pylint’s default rcfile search is used. Pylint is executed from the package’s setup directory, so pylintrc files in that directory will take precedence over others.

  • mypy_ini - Optional path to a mypy configuration file to use. If not provided, mypy’s default configuration file search is used. mypy is executed from the package’s setup directory, so mypy.ini files in that directory will take precedence over others.


This is an example Python package declaration for a pw_my_module module.



pw_python_package("py") {
  setup = [
  sources = [
  tests = [
  python_deps = [
  python_test_deps = [ "$dir_pw_build/py" ]
  pylintrc = "$dir_pigweed/.pylintrc"


A pw_python_script represents a set of standalone Python scripts and/or tests. These files support all of the arguments of pw_python_package except those setup. These targets can be installed, but this only installs their dependencies.

pw_python_script allows creating a pw_python_action associated with the script. To create an action, pass an action scope to pw_python_script. If there is only a single source file, it serves as the action’s script by default.

An action in pw_python_script can always be replaced with a standalone pw_python_action, but using the embedded action has some advantages:

  • The embedded action target bridges the gap between actions and Python targets. A Python script can be expressed in a single, concise GN target, rather than in two overlapping, dependent targets.

  • The action automatically depends on the pw_python_script. This ensures that the script’s dependencies are installed and the action automatically reruns when the script’s sources change, without needing to specify a dependency, a step which is easy to forget.

  • Using a pw_python_script with an embedded action is a simple way to check an existing action’s script with Pylint or Mypy or to add tests.


Represents a group of pw_python_package and pw_python_script targets. These targets do not add any files. Their subtargets simply forward to those of their dependencies.


Represents a set of local and PyPI requirements, with no associated source files. These targets serve the role of a requirements.txt file.

When packages are installed by Pigweed, additional version constraints can be provided using the pw_build_PIP_CONSTRAINTS GN arg. This option should contain a list of paths to pass to the --constraint option of pip install. This can be used to synchronize dependency upgrades across a project which facilitates reproducibility of builds.

Note using multiple pw_python_requirements that install different versions of the same package will currently cause unpredictable results, while using constraints should have correct results (which may be an error indicating a conflict).

Python Distributables

Pigweed also provides some templates to make it easier to bundle Python packages for deployment. These templates are found in pw_build/python_dist.gni. See the .gni file for complete documentation on building distributables.


Collects Python wheels for one or more pw_python_package targets, plus any additional pw_python_package targets they depend on, directly or indirectly. Note that this does not include Python dependencies that come from outside the GN build, like packages from PyPI, for example. Those should still be declared in the package’s file as usual.


  • packages - List of pw_python_package targets whose wheels should be included; their dependencies will be pulled in as wheels also.

  • directory - Output directory for the collected wheels. Defaults to $target_out_dir/$target_name.

Wheel collection under the hood

The .wheel subtarget of every pw_python_package generates a wheel (.whl) for the Python package. The pw_python_wheels template figures out which wheels to collect by traversing the pw_python_package_wheels GN metadata key, which lists the output directory for each wheel.


Generates a .zip archive suitable for deployment outside of the project’s developer environment. The generated .zip contains Python wheels (.whl files) for one or more pw_python_package targets, plus wheels for any additional pw_python_package targets in the GN build they depend on, directly or indirectly. Dependencies from outside the GN build, such as packages from PyPI, must be listed in packages’ or setup.cfg files as usual.

The .zip also includes simple setup scripts for Linux, MacOS, and Windows. The setup scripts automatically create a Python virtual environment and install the whole collection of wheels into it using pip.

Optionally, additional files and directories can be included in the archive. One common example of an additional file to include is a README file with setup and usage instructions for the distributable. A simple ready-to-use README file is available at pw_build/py_dist/


  • packages - A list of pw_python_package targets whose wheels should be included; their dependencies will be pulled in as wheels also.

  • inputs - An optional list of extra files to include in the generated .zip, formatted the same way as the inputs argument to pw_zip targets.

  • dirs - An optional list of directories to include in the generated .zip, formatted the same was as the dirs argument to pw_zip targets.




pw_python_zip_with_setup("my_tools") {
  packages = [ ":some_python_package" ]
  inputs = [ "$dir_pw_build/python_dist/ > /${target_name}/" ]


Generates a directory of Python packages from source files suitable for deployment outside of the project developer environment. The resulting directory contains only files mentioned in each package’s setup.cfg file. This is useful for bundling multiple Python packages up into a single package for distribution to other locations like


  • packages - A list of pw_python_package targets to be installed into the build directory. Their dependencies will be pulled in as wheels also.

  • include_tests - If true, copy Python package tests to a tests subdir.

  • extra_files - A list of extra files that should be included in the output. The format of each item in this list follows this convention:

    //some/nested/source_file > nested/destination_file
    • Source and destination file should be separated by >.

    • The source file should be a GN target label (starting with //).

    • The destination file will be relative to the generated output directory. Parent directories are automatically created for each file. If a file would be overwritten an error is raised.

  • generate_setup_cfg - If included, create a merged setup.cfg for all python Packages using a common_config_file as a base. That file should contain the required fields in the metadata and options sections as shown in Configuring setup() using setup.cfg files. append_git_sha_to_version and append_date_to_version will optionally append the current git SHA or date to the package version string after a + sign.

    generate_setup_cfg = {
      common_config_file = "pypi_common_setup.cfg"
      append_git_sha_to_version = true
      append_date_to_version = true





pw_create_python_source_tree("build_python_source_tree") {
  packages = [
  include_tests = true
  extra_files = [
    "// > ./",
    "//some_python_package/py/BUILD.bazel > some_python_package/BUILD.bazel",
    "//another_python_package/py/BUILD.bazel > another_python_package/BUILD.bazel",
  generate_setup_cfg = {
    common_config_file = "pypi_common_setup.cfg"
    append_git_sha_to_version = true
    append_date_to_version = true


$ tree ./out/obj/pw_env_setup/build_python_source_tree/
├── setup.cfg
├── some_python_package
│   ├── BUILD.bazel
│   ├──
│   ├── py.typed
│   ├──
│   └── tests
│       └──
└── another_python_package
    ├── BUILD.bazel
    ├── py.typed
    └── tests