GitHub Action to lint Python code with ruff

[cclauss]

Ruff supports over 500 lint rules and can be used to replace Flake8 (plus dozens of plugins), isort, pydocstyle, yesqa, eradicate, pyupgrade, and autoflake, all while executing (in Rust) tens or hundreds of times faster than any individual tool.

The ruff Action uses minimal steps to run in ~5 seconds, rapidly providing intuitive GitHub Annotations to contributors.

---

📚 Documentation preview 📚: https://datasette--2056.org.readthedocs.build/en/2056/

[simonw]

https://github.com/simonw/datasette/actions/runs/4664796647/jobs/8300596121?pr=2056 it's pretty fast - that finished in 9s.

[simonw]

Here are the failures: https://github.com/simonw/datasette/actions/runs/4684460653/jobs/8300630794?pr=2056

[simonw]

Cool - and now https://github.com/simonw/datasette/pull/2056/files is showing me those inline annotations:

[simonw]

OK, I'm sold - this is a really neat improvement.

One thing to change in the PR: right now it runs pip install --user ruff on every commit, which hits PyPI to install the package.

I prefer to avoid hitting PyPI every time, so I like to use the GitHub Actions cache. My usual pattern for that looks like this:

datasette/.github/workflows/test-pyodide.yml

Lines 16 to 20 in 5890a20

	- name: Set up Python 3.10
	uses: actions/setup-python@v3
	with:
	python-version: "3.10"
	cache: 'pip'

Then a separate command that runs pip install ... will benefit from that cache.

Are you OK to make that change?

[cclauss]

Ruff (written in Rust, not Python) is a 23MB executable so the time to download and pip install it dwarfs its runtime.

Let's run ruff with and without GitHub Actions pip cache side-by-side to see the relative performance.

Once you approve the workflows below, ruff_with_cache should echo cache-hit = false but if you rerun that job hopefully it should echo cache-hit = true. That will be the execution time that we are interested to compare.

There are two great problems in computer science: ;-)

1. Naming things
2. Cache invalidation
3. Off-by-one errors

For 2., https://github.com/actions/setup-python#caching-packages-dependencies is vital reading.
Only exactly pinned requirements can be cached. Currently in setup.py the only pinned dependencies are:

1. Sphinx==6.1.3
2. furo==2023.3.27
3. black==23.3.0
4. blacken-docs==1.13.0 # but unpinned elsewhere in setup.py

This means that there will be very few cache hits in the current actions. See the link below to print out cache hits:
https://github.com/actions/setup-python/blob/main/docs/advanced-usage.md#cache-hit

PyPI uses Fastly's CDN to quickly serve content to end-users, allowing us to minimize our hosting infrastructure and obscure possible downtime. -- https://pypi.org/sponsors

I would be shocked if Fastly does not have beefy CDN nodes in the same datacenters where GitHub Actions run so GHA requests to download ruff probably never hit a PyPI server.

[cclauss]

Status?