The presubmit module provides Python tools for running presubmit checks and checking and fixing code format. It also includes the presubmit check script for the Pigweed repository,

Presubmit checks are essential tools, but they take work to set up, and projects don’t always get around to it. The pw_presubmit module provides tools for setting up high quality presubmit checks for any project. We use this framework to run Pigweed’s presubmit on our workstations and in our automated building tools.

The pw_presubmit module also includes pw format, a tool that provides a unified interface for automatically formatting code in a variety of languages. With pw format, you can format Bazel, C, C++, Python, GN, and Go code according to configurations defined by your project. pw format leverages existing tools like clang-format, and it’s simple to add support for new languages. (Note: Bazel formatting requires buildifier to be present on your system. If it’s not Bazel formatting passes without checking.)

``pw format`` demo

The pw_presubmit package includes presubmit checks that can be used with any project. These checks include:

  • Check code format of several languages including C, C++, and Python

  • Initialize a Python environment

  • Run all Python tests

  • Run pylint

  • Run mypy

  • Ensure source files are included in the GN and Bazel builds

  • Build and run all tests with GN

  • Build and run all tests with Bazel

  • Ensure all header files contain #pragma once


Python 3

Creating a presubmit check for your project#

Creating a presubmit check for a project using pw_presubmit is simple, but requires some customization. Projects must define their own presubmit check Python script that uses the pw_presubmit package.

A project’s presubmit script can be registered as a pw_cli plugin, so that it can be run as pw presubmit.

Setting up the command-line interface#

The pw_presubmit.cli module sets up the command-line interface for a presubmit script. This defines a standard set of arguments for invoking presubmit checks. Its use is optional, but recommended.


Argument parsing code for presubmit checks.

pw_presubmit.cli.add_arguments(parser: ArgumentParser, programs: Optional[Programs] = None, default: str = '') None#

Adds common presubmit check options to an argument parser. Optional[Program], program: Sequence[Program], step: Sequence[Check], substep: str, output_directory: Optional[Path], package_root: Path, clear: bool, root: Optional[Path] = None, repositories: Collection[Path] = (), only_list_steps=False, list_steps: Optional[Callable[[], None]] = None, **other_args) int#

Processes arguments from add_arguments and runs the presubmit.

  • default_program – program to use if neither –program nor –step is used

  • program – from the –program option

  • step – from the –step option

  • substep – from the –substep option

  • output_directory – from –output-directory option

  • package_root – from –package-root option

  • clear – from the –clear option

  • root – base path from which to run presubmit checks; defaults to the root of the current directory’s repository

  • repositories – roots of Git repositories on which to run presubmit checks; defaults to the root of the current directory’s repository

  • only_list_steps – list the steps that would be executed, one per line, instead of executing them

  • list_steps – list the steps that would be executed with their docstrings

  • **other_args – remaining arguments defined by by add_arguments


exit code for sys.exit; 0 if successful, 1 if an error occurred

Presubmit output directory#

The pw_presubmit command line interface includes an --output-directory option that specifies the working directory to use for presubmits. The default path is out/presubmit. A subdirectory is created for each presubmit step. This directory persists between presubmit runs and can be cleaned by deleting it or running pw presubmit --clean.

Presubmit checks#

A presubmit check is defined as a function or other callable. The function must accept one argument: a PresubmitContext, which provides the paths on which to run. Presubmit checks communicate failure by raising an exception.

Presubmit checks may use the filter_paths decorator to automatically filter the paths list for file types they care about.

Either of these functions could be used as presubmit checks:

def file_contains_ni(ctx: PresubmitContext):
    for path in ctx.paths:
        with open(path) as file:
            contents =
            if 'ni' not in contents and 'nee' not in contents:
                raise PresumitFailure('Files must say "ni"!', path=path)

def run_the_build(_):['make', 'release'], check=True)

Presubmit checks functions are grouped into “programs” – a named series of checks. Projects may find it helpful to have programs for different purposes, such as a quick program for local use and a full program for automated use. The example script uses pw_presubmit.Programs to define quick and full programs.

PresubmitContext has the following members:

  • root: Source checkout root directory

  • repos: Repositories (top-level and submodules) processed by pw presubmit

  • output_dir: Output directory for this specific presubmit step

  • failure_summary_log: File path where steps should write a brief summary of any failures

  • paths: Modified files for the presubmit step to check (often used in formatting steps but ignored in compile steps)

  • all_paths: All files in the repository tree.

  • package_root: Root directory for pw package installations

  • override_gn_args: Additional GN args processed by build.gn_gen()

  • luci: Information about the LUCI build or None if not running in LUCI

  • num_jobs: Number of jobs to run in parallel

  • continue_after_build_error: For steps that compile, don’t exit on the first compilation error

The luci member is of type LuciContext and has the following members:

  • buildbucket_id: The globally-unique buildbucket id of the build

  • build_number: The builder-specific incrementing build number, if configured for this builder

  • project: The LUCI project under which this build is running (often pigweed or pigweed-internal)

  • bucket: The LUCI bucket under which this build is running (often ends with ci or try)

  • builder: The builder being run

  • swarming_server: The swarming server on which this build is running

  • swarming_task_id: The swarming task id of this build

  • cas_instance: The CAS instance accessible from this build

  • pipeline: Information about the build pipeline, if applicable.

  • triggers: Information about triggering commits, if applicable.

The pipeline member, if present, is of type LuciPipeline and has the following members:

  • round: The zero-indexed round number.

  • builds_from_previous_iteration: A list of the buildbucket ids from the previous round, if any, encoded as strs.

The triggers member is a sequence of LuciTrigger objects, which have the following members:

  • number: The number of the change in Gerrit.

  • patchset: The number of the patchset of the change.

  • remote: The full URL of the remote.

  • branch: The name of the branch on which this change is being/was submitted.

  • ref: The refs/changes/.. path that can be used to reference the patch for unsubmitted changes and the hash for submitted changes.

  • gerrit_name: The name of the Gerrit host.

  • submitted: Whether the change has been submitted or is still pending.

Additional members can be added by subclassing PresubmitContext and Presubmit. Then override Presubmit._create_presubmit_context() to return the subclass of PresubmitContext. Finally, add presubmit_class=PresubmitSubClass when calling


Presubmit steps can define substeps that can run independently in other tooling. These steps should subclass SubStepCheck and must define a substeps() method that yields SubStep objects. SubStep objects have the following members:

  • name: Name of the substep

  • _func: Substep code

  • args: Positional arguments for _func

  • kwargs: Keyword arguments for _func

SubStep objects must have unique names. For a detailed example of a SubStepCheck subclass see GnGenNinja in

Existing Presubmit Checks#

A small number of presubmit checks are made available through pw_presubmit modules.

Code Formatting#

Formatting checks for a variety of languages are available from pw_presubmit.format_code. These include C/C++, Java, Go, Python, GN, and others. All of these checks can be included by adding pw_presubmit.format_code.presubmit_checks() to a presubmit program. These all use language-specific formatters like clang-format or black.

These will suggest fixes using pw format --fix.

Options for code formatting can be specified in the pigweed.json file (see also SEED-0101). These apply to both pw presubmit steps that check code formatting and pw format commands that either check or fix code formatting.

  • python_formatter: Choice of Python formatter. Options are black (used by Pigweed itself) and yapf (the default).

  • black_path: If python_formatter is black, use this as the executable instead of black.

Example section from a pigweed.json file:

  "pw": {
    "pw_presubmit": {
      "format": {
        "python_formatter": "black",
        "black_path": "black"
Sorted Blocks#

Blocks of code can be required to be kept in sorted order using comments like the following:

# keep-sorted: start
# keep-sorted: end

This can be included by adding pw_presubmit.keep_sorted.presubmit_check to a presubmit program. Adding ignore-case to the start line will use case-insensitive sorting.

By default, duplicates will be removed. Lines that are identical except in case are preserved, even with ignore-case. To allow duplicates, add allow-dupes to the start line.

Prefixes can be ignored by adding ignore-prefix= followed by a comma-separated list of prefixes. The list below will be kept in this order. Neither commas nor whitespace are supported in prefixes.

# keep-sorted: start ignore-prefix=',"
# keep-sorted: end

Inline comments are assumed to be associated with the following line. For example, the following is already sorted. This can be disabled with sticky-comments=no.

# keep-sorted: start
# TODO(b/1234) Fix this.
# TODO(b/5678) Also fix this.
# keep-sorted: end

By default, the prefix of the keep-sorted line is assumed to be the comment marker used by any inline comments. This can be overridden by adding lines like sticky-comments=%,# to the start line.

Lines indented more than the preceding line are assumed to be continuations. Thus, the following block is already sorted. keep-sorted blocks can not be nested, so there’s no ability to add a keep-sorted block for the sub-items.

# keep-sorted: start
* abc
  * xyz
  * uvw
* def
# keep-sorted: end

The presubmit check will suggest fixes using pw keep-sorted --fix.

Future versions may support additional multiline list items.


Various rules can be applied to .gitmodules files. This check can be included by adding pw_presubmit.gitmodules.create() to a presubmit program. This function takes an optional argument of type pw_presubmit.gitmodules.Config. Config objects have several properties.

  • allow_non_googlesource_hosts: bool = False — If false, all submodules URLs must be on a Google-managed Gerrit server.

  • allowed_googlesource_hosts: Sequence[str] = () — If set, any Google-managed Gerrit URLs for submodules most be in this list. Entries should be like pigweed for

  • require_relative_urls: bool = False — If true, all submodules must be relative to the superproject remote.

  • allow_sso: bool = True — If false, sso:// and rpc:// submodule URLs are prohibited.

  • allow_git_corp_google_com: bool = True — If false, submodule URLs are prohibited.

  • require_branch: bool = False — If True, all submodules must reference a branch.

  • validator: Callable[[PresubmitContext, Path, str, Dict[str, str]], None] = None — A function that can be used for arbitrary submodule validation. It’s called with the PresubmitContext, the path to the .gitmodules file, the name of the current submodule, and the properties of the current submodule.

#pragma once#

There’s a pragma_once check that confirms the first non-comment line of C/C++ headers is #pragma once. This is enabled by adding pw_presubmit.cpp_checks.pragma_once to a presubmit program.

TODO(b/###) Formatting#

There’s a check that confirms TODO lines match a given format. Upstream Pigweed expects these to look like TODO(b/###): Explanation, but makes it easy for projects to define their own pattern instead.

To use this check add todo_check.create(todo_check.BUGS_OR_USERNAMES) to a presubmit program.

Python Checks#

There are two checks in the pw_presubmit.python_checks module, gn_pylint and gn_python_check. They assume there’s a top-level python GN target. gn_pylint runs Pylint and Mypy checks and gn_python_check runs Pylint, Mypy, and all Python tests.

Inclusive Language#

The inclusive language check looks for words that are typical of non-inclusive code, like using “master” and “slave” in place of “primary” and “secondary” or “sanity check” in place of “consistency check”.

These checks can be disabled for individual lines with “inclusive-language: ignore” on the line in question or the line above it, or for entire blocks by using “inclusive-language: disable” before the block and “inclusive-language: enable” after the block.


There’s a check that requires folders matching specific patterns contain OWNERS files. It can be included by adding module_owners.presubmit_check() to a presubmit program. This function takes a callable as an argument that indicates, for a given file, where a controlling OWNERS file should be, or returns None if no OWNERS file is necessary. Formatting of OWNERS files is handled similary to formatting of other source files and is discussed in Code Formatting.

Source in Build#

Pigweed provides checks that source files are configured as part of the build for GN, Bazel, and CMake. These can be included by adding and similar functions to a presubmit check. The CMake check additionally requires a callable that invokes CMake with appropriate options.


The pw_presubmit package provides tools for running presubmit checks.

class pw_presubmit.FileFilter(*, exclude: Iterable[_StrOrPattern] = (), endswith: Iterable[str] = (), name: Iterable[_StrOrPattern] = (), suffix: Iterable[str] = ())#

Allows checking if a path matches a series of filters.

Positive filters (e.g. the file name matches a regex) and negative filters (path does not match a regular expression) may be applied.

__init__(*, exclude: Iterable[_StrOrPattern] = (), endswith: Iterable[str] = (), name: Iterable[_StrOrPattern] = (), suffix: Iterable[str] = ()) None#

Creates a FileFilter with the provided filters.

  • endswith – True if the end of the path is equal to any of the passed strings

  • exclude – If any of the passed regular expresion match return False. This overrides and other matches.

  • name – Regexs to match with file names( True if the resulting regex matches the entire file name.

  • suffix – True if final suffix (as determined by pathlib.Path) is matched by any of the passed str.

matches(path: Union[str, Path]) bool#

Returns true if the path matches any filter but not an exclude.

If no positive filters are specified, any paths that do not match a negative filter are considered to match.

If ‘path’ is a Path object it is rendered as a posix path (i.e. using “/” as the path seperator) before testing with ‘exclude’ and ‘endswith’.

exception pw_presubmit.PresubmitFailure(description: str = '', path: Optional[Path] = None, line: Optional[int] = None)#

Optional exception to use for presubmit failures.

__init__(description: str = '', path: Optional[Path] = None, line: Optional[int] = None)#
class pw_presubmit.Programs(**programs: Sequence)#

A mapping of presubmit check programs.

Use is optional. Helpful when managing multiple presubmit check programs.

__init__(**programs: Sequence)#

Initializes a name: program mapping from the provided keyword args.

A program is a sequence of presubmit check functions. The sequence may contain nested sequences, which are flattened.*args, **kwargs) None#

Optional subprocess wrapper that causes a PresubmitFailure on errors.

pw_presubmit.filter_paths(*, endswith: Iterable[str] = (), exclude: Iterable[Union[Pattern[str], str]] = (), file_filter: Optional[FileFilter] = None, always_run: bool = False) Callable[[Callable], Check]#

Decorator for filtering the paths list for a presubmit check function.

Path filters only apply when the function is used as a presubmit check. Filters are ignored when the functions are called directly. This makes it possible to reuse functions wrapped in @filter_paths in other presubmit checks, potentially with different path filtering rules.

  • endswith – str or iterable of path endings to include

  • exclude – regular expressions of paths to exclude

  • file_filter – FileFilter used to select files

  • always_run – Run check even when no files match


a wrapped version of the presubmit function

Git hook#

You can run a presubmit program or step as a git hook using pw_presubmit.install_hook. This can be used to run certain presubmit checks before a change is pushed to a remote.

We strongly recommend that you only run fast (< 15 seconds) and trivial checks as push hooks, and perform slower or more complex ones in CI. This is because,

  • Running slow checks in the push hook will force you to wait longer for git push to complete, and

  • If your change fails one of the checks at this stage, it will not yet be uploaded to the remote, so you’ll have a harder time debugging any failures (sharing the change with your colleagues, linking to it from an issue tracker, etc).


A simple example presubmit check script follows. This can be copied-and-pasted to serve as a starting point for a project’s presubmit check script.

See for a more complex presubmit check script example.

"""Example presubmit check script."""

import argparse
import logging
import os
from pathlib import Path
import re
import sys
from typing import List, Optional, Pattern

    import pw_cli.log
except ImportError:
    print('ERROR: Activate the environment before running presubmits!',

import pw_presubmit
from pw_presubmit import (
from pw_presubmit.install_hook import install_hook

# Set up variables for key project paths.
PROJECT_ROOT = Path(os.environ['MY_PROJECT_ROOT'])

# Rerun the build if files with these extensions change.
_BUILD_EXTENSIONS = frozenset(
    ['.rst', '.gn', '.gni', *format_code.C_FORMAT.extensions])

# Presubmit checks
def release_build(ctx: PresubmitContext):
    build.gn_gen(ctx, build_type='release')

def host_tests(ctx: PresubmitContext):
    build.gn_gen(ctx, run_host_tests='true')

# Avoid running some checks on certain paths.

# Use the upstream pragma_once check, but apply a different set of path
# filters with @filter_paths.
@filter_paths(endswith='.h', exclude=PATH_EXCLUSIONS)
def pragma_once(ctx: PresubmitContext):

# Presubmit check programs
    # Checks not ran by default but that should be available. These might
    # include tests that are expensive to run or that don't yet pass.

    # List some presubmit checks to run
    # Use the upstream formatting checks, with custom path filters applied.
    # Include the upstream inclusive language check.
    # Include just the lint-related Python checks.

FULL = (
    QUICK,  # Add all checks from the 'quick' program
    # Use the upstream Python checks, with custom path filters applied.
    # Checks listed multiple times are only run once.

PROGRAMS = pw_presubmit.Programs(other=OTHER, quick=QUICK, full=FULL)

# Allowlist of remote refs for presubmit. If the remote ref being pushed to
# matches any of these values (with regex matching), then the presubmits
# checks will be run before pushing.

def run(install: bool, remote_ref: Optional[str],  **presubmit_args) -> int:
    """Process the --install argument then invoke pw_presubmit."""

    # Install the presubmit Git pre-push hook, if requested.
    if install:
        # '$remote_ref' will be replaced by the actual value of the remote ref
        # at runtime.
        install_git_hook('pre-push', [
            'python', '-m', 'tools.presubmit_check', '--base', 'HEAD~',
            '--remote-ref', '$remote_ref'
        return 0

    # Run the checks if either no remote_ref was passed, or if the remote ref
    # matches anything in the allowlist.
    if remote_ref is None or any(
  , remote_ref)
            for pattern in PRE_PUSH_REMOTE_REF_ALLOWLIST):
        return, **presubmit_args)

def main() -> int:
    """Run the presubmit checks for this repository."""
    parser = argparse.ArgumentParser(description=__doc__)
    cli.add_arguments(parser, PROGRAMS, 'quick')

    # Define an option for installing a Git pre-push hook for this script.
        help='Install the presubmit as a Git pre-push hook and exit.')

    # Define an optional flag to pass the remote ref into this script, if it
    # is run as a pre-push hook. The destination variable in the parsed args
    # will be `remote_ref`, as dashes are replaced with underscores to make
    # valid variable names.
        nargs='?',  # Make optional.
        help='Remote ref of the push command, for use by the pre-push hook.')

    return run(**vars(parser.parse_args()))

if __name__ == '__main__':

Code formatting tools#

The pw_presubmit.format_code module formats supported source files using external code format tools. The file can be invoked directly from the command line or from pw as pw format.


A simple example of adding support for a custom format. This code wraps the built in formatter to add a new format. It could also be used to replace a formatter or remove/disable a PigWeed supplied one.

#!/usr/bin/env python
"""Formats files in repository. """

import logging
import sys

import pw_cli.log
from pw_presubmit import format_code
from your_project import presubmit_checks
from your_project import your_check

YOUR_CODE_FORMAT = CodeFormat('YourFormat',
                              filter=FileFilter(suffix=('.your', )),


def _run(exclude, **kwargs) -> int:
    """Check and fix formatting for source files in the repo."""
    return format_code.format_paths_in_repo(exclude=exclude,

def main():
    return _run(**vars(format_code.arguments(git_paths=True).parse_args()))

if __name__ == '__main__':