In PyTorch, we enforce lint rules on code in order to help us catch common mistakes and enforce a greater degree of uniformity in our codebase than human reviewers can normally enforce.

PyTorch uses lintrunner to execute lints locally and in CI. This provides developers with a single command to run all the linters, and ensures consistency between the CI and local development environments. See the lintrunner repo for more info.

To get started, run the following to install lintrunner. Make sure you are at the root of the PyTorch repo:

pip install lintrunner
lintrunner init

This will install lintrunner on your system and download all the necessary dependencies to run linters locally. Note that this will install new packages with pip and download binaries maintained by the PyTorch team. If you want to see what lintrunner init will install, run lintrunner init --dry-run.

lintrunner

to lint your local changes (by default, this is your working tree changes and the HEAD commit).

lintrunner f

to format locally changed files.

If, say, your PR is targeting main, run:

lintrunner -m main

lintrunner torch/jit/_script.py torch/jit/_trace.py

xargs also works fine:

find -type f torch/onnx | xargs lintrunner

lintrunner --all-files

Warning: this may take a while, PyTorch has a lot of files.

lintrunner --take BLACK,CLANGFORMAT

To see what linter names to pass, consult the config. By default, lintrunner runs all linters. To skip a linter, use --skip.

For more docs run lintrunner --help, or see the lintrunner repo.