15 releases (5 breaking)

Uses new Rust 2021

0.8.0 May 6, 2022
0.5.3 Nov 17, 2021
0.4.3 Nov 12, 2021
0.3.0 Nov 4, 2021
0.1.1 Oct 22, 2021

#661 in Development tools

22 downloads per month


1.5K SLoC



lintrunner is a tool that runs linters. It is responsible for:

  • Deciding which files need to be linted.
  • Invoking linters according to a common protocol.
  • Gathering results and presenting them to users.

The intention is to provide a universal way to configure and invoke linters, which is useful on large polyglot projects.

The design of lintrunner is heavily inspired by linttool, a project that exists internally at Meta.


pip install lintrunner


First, you need to add a configuration file to your repo. See the Linter configuration section for more info.

Then, simply run lintrunner to lint your changes!

How to control what paths to lint lintrunner

When run with no arguments, lintrunner will check:

  • The files changed in the HEAD commit.
  • The files changed in the user’s working tree.

It does not check:

  • Any files not tracked by git; git add them to lint them.

There are multiple ways to customize how paths are checked:

Pass paths as positional arguments

For example:

lintrunner foo.py bar.cpp

This naturally composes with xargs, for example the canonical way to check every path in the repo is:

git grep -Il . | xargs lintrunner


Some ways to invoke xargs will cause multiple lintrunner processes to be run, increasing lint time (especially on huge path sets). As an alternative that gives lintrunner control of parallelization, you can use --paths-cmd. If --paths-cmd is specified lintrunner will execute that command and consider each line of its stdout to be a file to lint.

For example, the same command above would be:

lintrunner --paths-cmd='git grep -Il .'


If this is specified, lintrunner will read paths from the given file, one per line, and check those. This can be useful if you have some really complex logic to determine which paths to check.


This value can be any <tree-ish> accepted by git diff-tree, like a commit hash or revspec. If this is specified, lintrunner will check:

  • All paths changed from <tree-ish> to HEAD
  • All paths changed in the user's working tree.


Like --revision, except the revision is determined by computing the merge-base of HEAD and the provided <tree-ish>. This is useful for linting all commits in a specific pull request. For example, for a pull request targeting master, you can run:

lintrunner -m master

Linter configuration

lintrunner knows which linters to run and how by looking at a configuration file, conventionally named .lintrunner.toml.

Here is an example linter configuration:

name = 'FLAKE8'
include_patterns = [
  'src/**/*.py',  # unix-style globs supported
exclude_patterns = ['src/my_bad_file.py']
command = [
  # {{PATHSFILE}} gets rewritten to a tmpfile containing all paths to lint

A complete description of the configuration schema can be found here.

Linter protocol

Most linters have their own output format and arguments. In order to impose consistency on linter invocation and outputs, lintrunner implements a protocol that it expects linters to fulfill. In most cases, a small script (called a linter adapter) is required to implement the protocol for a given external linter. You can see some example adapters in examples/ .


Linters will be invoked according to the command specified by their configuration. They will be called once per lint run.

If a linter needs to know which paths to run on, it should take a {{PATHSFILE}} argument. During invocation, the string {{PATHSFILE}} will be replaced with the name of a temporary file containing which paths the linter should run on, one path per line.

A common way to implement this in a linter adapter is to use argparse’s fromfile_prefix_chars feature. In the Flake8 example above, we use @ as the fromfile_prefix_chars argument, so argparse will automatically read the {{PATHSFILE}} and supply its contents as a list of arguments.


Any lint messages a linter would like to communicate the user must be represented as a LintMessage. The linter, must print LintMessages as JSON Lines to stdout, one message per line. Output to stderr will be ignored.

A complete description of the LintMessage schema can be found here.


Linters should always exit with code 0. This is true even if lint errors are reported; lintrunner itself will determine how to exit based on what linters report.

To signal a general linter failure (which should ideally never happen!), linters can return a LintMessage with path = None.

In the event a linter exits non-zero, it will be caught by lintrunnerand presented as a “general linter failure” with stdout/stderr shown to the user. This should be considered a bug in the linter’s implementation of this protocol.


~175K SLoC