ruff

An extremely fast Python linter, written in Rust.

Linting the CPython codebase from scratch.

Major features:

• 10-100x faster than existing linters.
• Installable via pip.
• Python 3.10 compatibility.
• ESLint-inspired cache semantics.
• TypeScript-inspired --watch semantics.
• pyproject.toml support.

ruff is a proof-of-concept and not yet intended for production use. It supports only a small subset of the Flake8 rules, and may crash on your codebase.

Installation and usage

Installation

Available as ruff on PyPI:

pip install ruff

For now, wheels are available for Python 3.7, 3.8, 3.9, and 3.10 on macOS, Windows, and Linux. If a wheel isn't available for your Python version or platform, you'll need to install the Rust toolchain prior to running pip install ruff. (This is an effort limitation on my part, not a technical limitation.)

Usage

To run ruff, try any of the following:

ruff path/to/code/to/check.py
ruff path/to/code/
ruff path/to/code/*.py

You can run ruff in --watch mode to automatically re-run on-change:

ruff path/to/code/ --watch

Configuration

ruff is configurable both via pyproject.toml and the command line.

For example, you could configure ruff to only enforce a subset of rules with:

[tool.ruff]
line-length = 88
select = [
    "F401",
    "F403",
]

Alternatively, on the command-line:

ruff path/to/code/ --select F401 F403

See ruff --help for more:

ruff
A Python linter written in Rust

USAGE:
    ruff [OPTIONS] <FILES>...

ARGS:
    <FILES>...

OPTIONS:
    -e, --exit-zero             Exit with status code "0", even upon detecting errors
    -h, --help                  Print help information
        --ignore <IGNORE>...    Comma-separated list of error codes to ignore
    -n, --no-cache              Disable cache reads
    -q, --quiet                 Disable all logging (but still exit with status code "1" upon
                                detecting errors)
        --select <SELECT>...    Comma-separated list of error codes to enable
    -v, --verbose               Enable verbose logging
    -w, --watch                 Run in watch mode by re-running whenever files change

Development

ruff is written in Rust (1.63.0). You'll need to install the Rust toolchain for development.

Assuming you have cargo installed, you can run:

cargo run resources/test/src
cargo fmt
cargo clippy
cargo test

Deployment

ruff is distributed on PyPI, and published via maturin.

See: .github/workflows/release.yaml.

Benchmarking

First, clone CPython. It's a large and diverse Python codebase, which makes it a good target for benchmarking.

git clone --branch 3.10 https://github.com/python/cpython.git resources/test/cpython

Add this pyproject.toml to the CPython directory:

[tool.linter]
line-length = 88
exclude = [
    "Lib/ctypes/test/test_numbers.py",
    "Lib/dataclasses.py",
    "Lib/lib2to3/tests/data/bom.py",
    "Lib/lib2to3/tests/data/crlf.py",
    "Lib/lib2to3/tests/data/different_encoding.py",
    "Lib/lib2to3/tests/data/false_encoding.py",
    "Lib/lib2to3/tests/data/py2_test_grammar.py",
    "Lib/sqlite3/test/factory.py",
    "Lib/sqlite3/test/hooks.py",
    "Lib/sqlite3/test/regression.py",
    "Lib/sqlite3/test/transactions.py",
    "Lib/sqlite3/test/types.py",
    "Lib/test/bad_coding2.py",
    "Lib/test/badsyntax_3131.py",
    "Lib/test/badsyntax_pep3120.py",
    "Lib/test/encoded_modules/module_iso_8859_1.py",
    "Lib/test/encoded_modules/module_koi8_r.py",
    "Lib/test/sortperf.py",
    "Lib/test/test_email/torture_test.py",
    "Lib/test/test_fstring.py",
    "Lib/test/test_genericpath.py",
    "Lib/test/test_getopt.py",
    "Lib/test/test_grammar.py",
    "Lib/test/test_htmlparser.py",
    "Lib/test/test_importlib/stubs.py",
    "Lib/test/test_importlib/test_files.py",
    "Lib/test/test_importlib/test_metadata_api.py",
    "Lib/test/test_importlib/test_open.py",
    "Lib/test/test_importlib/test_util.py",
    "Lib/test/test_named_expressions.py",
    "Lib/test/test_patma.py",
    "Lib/test/test_peg_generator/__main__.py",
    "Lib/test/test_pipes.py",
    "Lib/test/test_source_encoding.py",
    "Lib/test/test_weakref.py",
    "Lib/test/test_webbrowser.py",
    "Lib/tkinter/__main__.py",
    "Lib/tkinter/test/test_tkinter/test_variables.py",
    "Modules/_decimal/libmpdec/literature/fnt.py",
    "Modules/_decimal/tests/deccheck.py",
    "Tools/c-analyzer/c_parser/parser/_delim.py",
    "Tools/i18n/pygettext.py",
    "Tools/test2to3/maintest.py",
    "Tools/test2to3/setup.py",
    "Tools/test2to3/test/test_foo.py",
    "Tools/test2to3/test2to3/hello.py",
]

Next, to benchmark the release build:

cargo build --release

hyperfine --ignore-failure --warmup 1 \
  "./target/release/ruff ./resources/test/cpython/ --no-cache" \
  "./target/release/ruff ./resources/test/cpython/"

Benchmark 1: ./target/release/ruff ./resources/test/cpython/ --no-cache
  Time (mean ± σ):     353.6 ms ±   7.6 ms    [User: 2868.8 ms, System: 171.5 ms]
  Range (min … max):   344.4 ms … 367.3 ms    10 runs

Benchmark 2: ./target/release/ruff ./resources/test/cpython/
  Time (mean ± σ):      59.6 ms ±   2.5 ms    [User: 36.4 ms, System: 345.6 ms]
  Range (min … max):    55.9 ms …  67.0 ms    48 runs

To benchmark against the ecosystem's existing tools:

hyperfine --ignore-failure --warmup 5 \
  "./target/release/ruff ./resources/test/cpython/ --no-cache" \
  "pylint --recursive=y resources/test/cpython/" \
  "pyflakes resources/test/cpython" \
  "autoflake --recursive --expand-star-imports --remove-all-unused-imports --remove-unused-variables --remove-duplicate-keys resources/test/cpython" \
  "pycodestyle resources/test/cpython" \
  "pycodestyle --select E501 resources/test/cpython" \
  "flake8 resources/test/cpython" \
  "flake8 --select=F831,F541,F634,F403,F706,F901,E501 resources/test/cpython" \
  "python -m scripts.run_flake8 resources/test/cpython" \
  "python -m scripts.run_flake8 resources/test/cpython --select=F831,F541,F634,F403,F706,F901,E501"

In order, these evaluate:

• ruff
• Pylint
• PyFlakes
• autoflake
• pycodestyle
• pycodestyle, limited to the checks supported by ruff
• Flake8
• Flake8, limited to the checks supported by ruff
• Flake8, with a hack to enable multiprocessing on macOS
• Flake8, with a hack to enable multiprocessing on macOS, limited to the checks supported by ruff

(You can poetry install from ./scripts to create a working environment for the above.)

Benchmark 1: ./target/release/ruff ./resources/test/cpython/ --no-cache
  Time (mean ± σ):     469.3 ms ±  16.3 ms    [User: 2663.0 ms, System: 972.5 ms]
  Range (min … max):   445.2 ms … 494.8 ms    10 runs

Benchmark 2: pylint --recursive=y resources/test/cpython/
  Time (mean ± σ):     27.211 s ±  0.097 s    [User: 26.405 s, System: 0.799 s]
  Range (min … max):   27.056 s … 27.349 s    10 runs

Benchmark 3: pyflakes resources/test/cpython
  Time (mean ± σ):     27.309 s ±  0.033 s    [User: 27.137 s, System: 0.169 s]
  Range (min … max):   27.267 s … 27.372 s    10 runs

Benchmark 4: autoflake --recursive --expand-star-imports --remove-all-unused-imports --remove-unused-variables --remove-duplicate-keys resources/test/cpython
  Time (mean ± σ):      8.027 s ±  0.024 s    [User: 74.255 s, System: 0.953 s]
  Range (min … max):    7.969 s …  8.052 s    10 runs

Benchmark 5: pycodestyle resources/test/cpython
  Time (mean ± σ):     41.666 s ±  0.266 s    [User: 41.531 s, System: 0.132 s]
  Range (min … max):   41.295 s … 41.980 s    10 runs

Benchmark 6: pycodestyle --select E501 resources/test/cpython
  Time (mean ± σ):     14.547 s ±  0.077 s    [User: 14.466 s, System: 0.079 s]
  Range (min … max):   14.429 s … 14.695 s    10 runs

Benchmark 7: flake8 resources/test/cpython
  Time (mean ± σ):     75.700 s ±  0.152 s    [User: 75.254 s, System: 0.440 s]
  Range (min … max):   75.513 s … 76.014 s    10 runs

Benchmark 8: flake8 --select=F831,F541,F634,F403,F706,F901,E501 resources/test/cpython
  Time (mean ± σ):     75.122 s ±  0.532 s    [User: 74.677 s, System: 0.440 s]
  Range (min … max):   74.130 s … 75.606 s    10 runs

Benchmark 9: python -m scripts.run_flake8 resources/test/cpython
  Time (mean ± σ):     12.794 s ±  0.147 s    [User: 90.792 s, System: 0.738 s]
  Range (min … max):   12.606 s … 13.030 s    10 runs

Benchmark 10: python -m scripts.run_flake8 resources/test/cpython --select=F831,F541,F634,F403,F706,F901,E501
  Time (mean ± σ):     12.487 s ±  0.118 s    [User: 90.052 s, System: 0.714 s]
  Range (min … max):   12.265 s … 12.665 s    10 runs

Summary
  './target/release/ruff ./resources/test/cpython/ --no-cache' ran
   17.10 ± 0.60 times faster than 'autoflake --recursive --expand-star-imports --remove-all-unused-imports --remove-unused-variables --remove-duplicate-keys resources/test/cpython'
   26.60 ± 0.96 times faster than 'python -m scripts.run_flake8 resources/test/cpython --select=F831,F541,F634,F403,F706,F901,E501'
   27.26 ± 1.00 times faster than 'python -m scripts.run_flake8 resources/test/cpython'
   30.99 ± 1.09 times faster than 'pycodestyle --select E501 resources/test/cpython'
   57.98 ± 2.03 times faster than 'pylint --recursive=y resources/test/cpython/'
   58.19 ± 2.02 times faster than 'pyflakes resources/test/cpython'
   88.77 ± 3.14 times faster than 'pycodestyle resources/test/cpython'
  160.06 ± 5.68 times faster than 'flake8 --select=F831,F541,F634,F403,F706,F901,E501 resources/test/cpython'
  161.29 ± 5.61 times faster than 'flake8 resources/test/cpython'

License

MIT