Table of contents
Scikit-HEP uses pre-commit to check code style. It can be installed through
brew (macOS) or
pip (anywhere). There are two modes to use it locally; you can check manually with
pre-commit run (changes only) or
pre-commit run --all-files (all). You can also run
pre-commit install to add checks as a git pre-commit hook (which is where it gets its name). It’s worth trying, even if you’ve tried and failed to set up a custom pre-commit hook before; it’s quite elegant and does not add or commit the changes, it just makes the changes and allows you to check and add them. You can always override the hook with
Here is a minimal
.pre-commit-config.yaml file with some handy options:
repos: - repo: https://github.com/pre-commit/pre-commit-hooks rev: "v4.3.0" hooks: - id: check-added-large-files - id: check-case-conflict - id: check-merge-conflict - id: check-symlinks - id: check-yaml - id: debug-statements - id: end-of-file-fixer - id: mixed-line-ending - id: name-tests-test args: ["--pytest-test-first"] - id: requirements-txt-fixer - id: trailing-whitespace
Helpful tip: Pre-commit runs top-to-bottom, so put checks that modify content (like the several of the pre-commit-hooks above, or Black) above checks that might be more likely to pass after the modification (like flake8).
Keeping pinned versions fresh: You can use
pre-commit autoupdate to move your tagged versions forward to the latest tags! Due to the design of pre-commit’s caching system, these must point at fixed tags, never put a branch here.
Checking in CI: You can have this checked and often automatically corrected for you using pre-commit.ci. It will even update your
rev: versions every week or so if your checks update!
To use, just go to pre-commit.ci, click “Log in with GitHub”, click “Add an Installation” if adding for the first time for an org or user, or “Manage repos on GitHub” for an existing installation (like Scikit-HEP), then add your repository from the list in GitHub’s interface.
Now there will be a new check, and pre-commit.ci will commit changes if the pre-commit check made any changes. Note that there are a couple of missing features: Docker based checks will not work (pre-commit.ci already runs in docker), you cannot enable a
--manual flag, so extra checks will not run, and jobs should not download packages (use
additional-dependencies: to add what you need).
Black is a popular auto-formatter from the Python Software Foundation. One of the main features of Black is that it is “opinionated”; that is, it is almost completely unconfigurable. Instead of allowing you to come up with your own format, it enforces one on you. While I am quite sure you can come up with a better format, having a single standard makes it possible to learn to read code very fast - you can immediately see nested lists, matching brackets, etc. There also is a faction of developers that dislikes all auto-formatting tools, but inside a system like pre-commit, auto-formatters are ideal. They also speed up the writing of code because you can ignore formatting your code when you write it. By imposing a standard, all Scikit-HEP developers can quickly read any package’s code.
Also, properly formatted code has other benefits, such as if two developers make the same change, they get the same formatting, and merge requests are easier. The style choices in Black were explicitly made to optimize git diffs!
There are a few options, mostly to enable/disable certain files, remove string normalization, and to change the line length, and those go in your
Here is the snippet to add Black to your
- repo: https://github.com/psf/black rev: "22.10.0" hooks: - id: black
You can add a Black badge to your repo as well
[![Code style: black](https://img.shields.io/badge/code%20style-black-000000.svg)](https://github.com/psf/black)
.. image:: https://img.shields.io/badge/code%20style-black-000000.svg :target: https://github.com/psf/black
In very specific situations, you may want to retain special formatting. After carefully deciding that it is a special use case, you can use
# fmt: on and
# fmt: off around a code block to have it keep custom formatting. Always consider refactoring before you try this option! Most of the time, you can find a way to make the Blacked code look better by rewriting your code; factor out long unreadable portions into a variable, avoid writing matrices as 1D lists, etc.
Jupyter notebook support
If you want Black for Jupyter notebooks too, replace
id: black with
id:black-jupyter above. You also might like the following hook, which cleans Jupyter outputs:
- repo: https://github.com/kynan/nbstripout rev: "0.6.1" hooks: - id: nbstripout
Documentation / README snippets support
If you want Black used in your documentation, you can use blacken-docs. This can even catch syntax errors in code snippets! It supports markdown and restructured text. Note that because black is in
additional_dependencies, you’ll have to keep it up to date manually.
- repo: https://github.com/asottile/blacken-docs rev: v1.12.1 hooks: - id: blacken-docs additional_dependencies: [black==22.8.0]
Check-manifest is a fantastic, highly recommended tool that verifies you have working SDists. You can install it from PyPI. Run it on your repository and see what it says. If you want to ignore files (like test folders, example folders, docs, etc) you can add these into your
[tool.check-manifest] ignore = [ ".travis.yml", ]
Add the following to your pre-commit config:
- repo: https://github.com/mgedmin/check-manifest rev: "0.48" hooks: - id: check-manifest
If you use
setuptools_scm, you might want to add:
If this is too slow:
Warning: For a complex package, this may be slow. You can optionally set
stages: [manual] just below the id, and then only run this explicitly (probably in CI only). In GHA, you should enable the manual stage, which will run all checks:
- uses: firstname.lastname@example.org with: extra_args: --show-diff-on-failure --all-files --hook-stage manual
One of the most exciting advancements in Python in the last 10 years has been static type hints. Scikit-HEP is just beginning to make sure packages are type-hint ready. One of the challenges for providing static type hints is that it was developed in the Python 3 era and it really shines in a Python 3.7+ codebase (due to
from __future__ import annotations, which turns annotations into strings and allows you to use future Python features in Python 3.7+ annotations as long as your type checker supports them). For now, it is recommended that you make an attempt to support type checking through your public API in the best way that you can (based on your supported Python versions). Stub files can be used instead for out-of-line typing. MyPy is suggested for type checking, though there are several other good options to try, as well. If you have built-in support for type checking, you need to add empty
py.typed files to all packages/subpackages to indicate that you support it.
Read more about type checking on the dedicated page.
The MyPy addition for pre-commit:
- repo: https://github.com/pre-commit/mirrors-mypy rev: "v0.991" hooks: - id: mypy files: src args: 
You should always specify args, as the hook’s default hides issues - it’s designed to avoid configuration, but you should add configuration. You can also add items to the virtual environment setup for MyPy by pre-commit, for example:
MyPy has a config section in
pyproject.toml that looks like this:
[tool.mypy] files = "src" python_version = "3.7" strict = true show_error_codes = true enable_error_code = ["ignore-without-code", "redundant-expr", "truthy-bool"] warn_unreachable = true # You can disable imports or control per-module/file settings here [[tool.mypy.overrides]] module = [ "numpy.*", ] ignore_missing_imports = true
There are a lot of options, and you can start with only typing global code and functions with at least one type annotation (the default) and enable more checks as you go (possibly by slowly uncommenting items in the list above). You can ignore missing imports on libraries as shown above, one section each. And you can disable MyPy on a line with
# type: ignore. One strategy would be to enable
check_untyped_defs first, followed by
disallow_incomplete_defs. You can add these per file by adding a
# mypy: <option> at the top of a file. You can also pass
--strict on the command line.
strict = true is now allowed in config files, too.
The extra strict options shown above (
truthy-bool) can trigger too often (like on
sys.platform checks) and have to be ignored occasionally, but can find some signifiant logic errors in your typing.
PyCln will clean up your imports if you have any that are not needed. There is a Flake8 check for this, but it’s usually nicer to automatically do the cleanup instead of forcing a user to manually delete unneeded imports. If you use the manual stage, it’s opt-in instead of automatic.
- repo: https://github.com/hadialqattan/pycln rev: "v2.1.2" hooks: - id: pycln args: [--all] stages: [manual]
Flake8 can check a collection of good practices for you, ranging from simple style to things that might confuse or detract users, such as unused imports, named values that are never used, mutable default arguments, and more. Unlike black and some other tools, flake8 does not correct problems, it just reports them. Some of the checks could have had automated fixes, sadly (which is why Black is nice). Here is a suggested
setup.cfg to enable compatibility with Black (flake8 does not support pyproject.toml configuration, sadly):
[flake8] extend-ignore = E203, E501
One recommended plugin for flake8 is
flake8-bugbear, which catches many common bugs. It is highly opinionated and can be made more so with the
B9 setting. You can also set a max complexity, which bugs you when you have complex functions that should be broken up. Here is an opinionated config:
[flake8] max-complexity = 12 extend-select = B9 extend-ignore = E203, E501, E722, B950
(Error E722 is important, but it is identical to the activated B001.) Here is the flake8 addition for pre-commit, with the
- repo: https://github.com/pycqa/flake8 rev: "6.0.0" hooks: - id: flake8 additional_dependencies: [flake8-bugbear]
This will be too much at first, so you can disable or enable any test by it’s label. You can also disable a check or a list of checks inline with
# noqa: X### (where you list the check label(s)). Over time, you can fix and enable more checks. A few interesting plugins:
flake8-bugbear: Fantastic checker that catches common situations that tend to create bugs. Codes:
flake8-docstrings: Docstring checker.
flake8-spellcheck: Spelling checker. Code:
flake8-import-order: Enforces PEP8 grouped imports (you may prefer isort). Code:
pep8-naming: Enforces PEP8 naming rules. Code:
flake8-print: Makes sure you don’t have print statements that sneak in. Code:
Having something verify you don’t add a print statement by mistake is very useful. A common need for the print checker would be to add it to a single directory (
src if you are following the convention recommended). You can do the next best thing by removing directories and file just for this check (
T) in your flake8 config:
[flake8] per-file-ignores = tests/*: T examples/*: T
Over time, you can end up with extra “noqa” comments that are no longer needed. This is a flake8 helper that removes those comments when they are no longer required.
- repo: https://github.com/asottile/yesqa rev: "v1.4.0" hooks: - id: yesqa
You need to have the same extra dependencies as flake8. In YAML, you can save the list given to yesqa and repeat it in flake8 using
*flake8-dependencies after the colon.
You can have your imports sorted automatically by isort. This will sort your imports, and is black compatible. One reason to have sorted imports is to reduce merge conflicts. Another is to clarify where imports come from - standard library imports are in a group above third party imports, which are above local imports. All this is configurable, as well. To use isort, the following pre-commit config will work:
- repo: https://github.com/PyCQA/isort rev: "5.10.1" hooks: - id: isort
In order to use it, you need to add some configuration. You can add it to
pyproject.toml or classic config files:
[tool.isort] profile = "black"
Another useful tool is PyUpgrade, which monitors your codebase for “old” style syntax. Most useful to keep Python 2 outdated constructs out, it can even do some code updates for different versions of Python 3, like adding f-strings when clearly better (please always use them, they are faster) if you set
--py36-plus (for example). This is a recommended addition for any project.
- repo: https://github.com/asottile/pyupgrade rev: "v3.3.0" hooks: - id: pyupgrade args: ["--py37-plus"]
If you set this to
--py37-plus, you can add the annotations import by adding the following line to your isort pre-commit hook configuration:
args: ["-a", "from __future__ import annotations"]
Also make sure isort comes before pyupgrade. Now when you run pre-commit, it will clean up your annotations to 3.7+ style, too!
There is a tool that keeps your
setup.cfg organized, and makes sure that important parts (like Python classifiers) are in sync. This tool,
setup-cfg-fmt, has native support for pre-commit:
- repo: https://github.com/asottile/setup-cfg-fmt rev: "v2.2.0" hooks: - id: setup-cfg-fmt args: [--include-version-classifiers, --max-py-version=3.11]
Make sure you list the highest version of Python you are testing with here.
You can and should check for spelling errors in your code too. If you want to add this, you can use codespell for common spelling mistakes. Unlike most spell checkers, this has a list of mistakes it looks for, rather than a list of “valid” words. To use:
- repo: https://github.com/codespell-project/codespell rev: "v2.2.2" hooks: - id: codespell args: ["-L", "sur,nd"]
You can list allowed spellings in a comma separated string passed to
--ignore-words-list - usually it is better to use long options when you are not typing things live). The example above will allow “Big Sur” and “ND”. You can instead use a comma separated list in
[codespell] ignore-words-list = sur,nd
You can also use a local pygrep check to eliminate common capitalization errors, such as the one below:
- repo: local hooks: - id: disallow-caps name: Disallow improper capitalization language: pygrep entry: PyBind|Numpy|Cmake|CCache|Github|PyTest exclude: .pre-commit-config.yaml
This is a repository with a collection of pre-commit extra hooks that protect against some common, easy to detect, mistakes. You can pick and choose the hooks you want from the repo; here are some common ones:
- repo: https://github.com/pre-commit/pygrep-hooks rev: "v1.9.0" hooks: - id: python-check-blanket-noqa - id: python-check-blanket-type-ignore - id: python-no-log-warn - id: python-no-eval - id: python-use-type-annotations - id: rst-backticks - id: rst-directive-colons - id: rst-inline-touching-normal
If you want to add specific type ignores, see mypy_clean_slate for a tool that will add the specific ignores for you. You’ll need to remove the existing type ignores (
git ls-files '*.py' | xargs sed -i '' 's/ # type: ignore//g'), copy the pre-commit output (with
--show-error-codes in mypy’s args) to a file called
mypy_error_report.txt, then run
pipx run mypy_clean_slate -a.
If you have C++ code, you should have a
.clang-format file and use the following pre-commit config:
- repo: https://github.com/pre-commit/mirrors-clang-format rev: "v15.0.4" hooks: - id: clang-format types_or: [c++, c, cuda]
This will use 1-2 MB binary wheels from PyPI on all common platforms. You can generated such a file using
pipx run clang-format -style=llvm -dump-config > .clang-format.
If you have shell scripts, you can protect against common mistakes using shellcheck.
- repo: https://github.com/shellcheck-py/shellcheck-py rev: "v0.8.0.4" hooks: - id: shellcheck
The prettier tool can format a large number of different file types. An example of usage:
Since this formats a variety of very common file types (like
.json), you will usually want to provide a
types_or setting (shown above) with the files you are interested in auto-formatting. You can try it without the
types_or first to see what it can do. Special markups in html/markdown files might clash with auto-formatting - check to verify your files are supported. This check runs using node, but pre-commit handles this for you.
If you have a .editor-config file, prettier will respect the rules in it. You can also specify a config file for prettier, or pass options to
args: in pre-commit. One such option is
--prose-wrap, which can be set to
"always" to have prettier reflow text. You can turn off prettier for blocks with comments depending on language.
PyLint is very opinionated, with a high signal-to-noise ratio. However, by limiting the default checks or by starting off a new project using them, you can get some very nice linting, including catching some problematic code that otherwise is hard to catch. PyLint is generally not a good candidate for pre-commit, since it needs to have your package installed - it is less static of check than Flake8. Here is a suggested pyproject.toml entry to get you started:
[tool.pylint] master.py-version = "3.7" master.jobs = "0" reports.output-format = "colorized" similarities.ignore-imports = "yes" messages_control.enable = ["useless-suppression"] messages_control.disable = [ "design", "fixme", "line-too-long", "wrong-import-position", ]
And a noxfile entry:
@nox.session def pylint(session: nox.Session) -> None: session.install("-e", ".") session.install("pylint") session.run("pylint", "src", *session.posargs)
And you can add this to your GitHub Actions using
run: pipx run nox -s pylint.