Move more configs to pyproject.toml

[alexfikl]

It's the future! 🎉

This mostly removes pytest.ini and moves most of the things from setup.cfg here.

Flake8 doesn't support pyproject, so it's not moved, but there is a project that hacks support in there if we want:
https://github.com/john-hen/Flake8-pyproject

[alexfikl]

The failures seem to be related to the release of numpy 2.0, but I'm not sure what it doesn't like since pyopencl should be built from source on the CI. Right?

EDIT: For reference, this fails on main too: https://github.com/inducer/pyopencl/actions/runs/9630426577/job/26560978801

[matthiasdiener]

Flake8 doesn't support pyproject, so it's not moved, but there is a project that hacks support in there if we want: https://github.com/john-hen/Flake8-pyproject

FWIW, I've been using flake8-pyproject in some of my projects, and it seems to work fine (see e.g. https://github.com/matthiasdiener/orderedsets/blob/551786a5b254372c097b3f676808d8e49dc0dcc8/pyproject.toml#L33)

[alexfikl]

FWIW, I've been using flake8-pyproject in some of my projects, and it seems to work fine (see e.g. matthiasdiener/orderedsets@551786a/pyproject.toml#L33)

Yeah, I've used it too and it worked without any issues.

One small problem with flake8 at the moment is also that the scripts in ci-support look for some strings in setup.cfg (e.g. # enable-isort), so that requires some more work..

[inducer]

inducer/ci-support@3cbd99a should allow seamless use of flake8-pyproject.

[alexfikl]

I added this too, but I'm not very convinced, more for a small discussion. (I don't recall if you mentioned this at some point?)

I thought about it mainly for two reasons:

• removing the enable-isort (and others) grepping in ci-support: with this it can just install all the plugins it wants and the project will select the ones it wants.
• if we move to ruff, all these are bundled anyway and we'll need to select the ones we want.
• it's unlikely at this point, but this would also guard against new added error codes.

[inducer]

I think I prefer this, even. Getting ci-support to understand this won't be too much of a lift. I like it mostly because it allows us to avoid making up our own idiotic config bits.

[alexfikl]

I like it mostly because it allows us to avoid making up our own idiotic config bits.

Yeah, that too! We can leave it in then, and slowly teach ci-support and everybody else to slowly migrate.

[inducer]

Now that I think of it: Why do we even need this? If we have the tools listed in project.optional-dependencies.test, we could just use that.

[inducer]

Hmm. I guess the "ruff" argument stands. :/ How close is ruff a this point? Should we just switch to it?

[alexfikl]

Hm, yeah, that's a good point. Sounds easier to standardize on some optional dependencies group name and just work with that if it exists!

I kind of like that better.

[alexfikl]

Hmm. I guess the "ruff" argument stands. :/ How close is ruff a this point? Should we just switch to it?

It doesn't have a lot of the fancier pylint checks that actually execute the code (or whatever they do), but everything else is there and seems to be working very nicely!

I would say it's pretty solid at this point (since they got bored and re-implemented black as well :))

[alexfikl]

Here's a list of the pylint checks that are implemented
astral-sh/ruff#970

[inducer]

Alright, let's say ruff is the future then, replacing flake8+plugins. Pylint can live for the moment.

[alexfikl]

With inducer/ci-support#32 we could also remove setup.cfg completely. The only one that remains is pylint with the special .pylintrc-local.yml 😖

[alexfikl]

@inducer This is looking good to me and all the pieces from ci-support seem to be working nicely. Is there anything else missing? I would leave the transition to ruff for another PR.

[inducer]

So I tried moving the whole thing to ruff, and I liked what I saw. 🤷 It's a bit of churn, but it's just really quite nice. I rather like the LSP integration. I imagine we'll do that (with time) across the whole stack.

[inducer]

LMK how you feel about this.

[inducer]

It's certainly much less of a mess than flake8.

[alexfikl]

So I tried moving the whole thing to ruff, and I liked what I saw. 🤷 It's a bit of churn, but it's just really quite nice. I rather like the LSP integration. I imagine we'll do that (with time) across the whole stack.

I'm all for it! It looks very nice and it seems to have caught some extra stuff 😁

It's a bit of churn

Have you tried ruff check --fix? I mostly replaced isort with ruff check --fix --select=I and it works very well, not sure about the other rules..