MAINT: add config for ruff formatter

[tobiasdiez]

Add configuration for the ruff formatter to enforce single quotes for strings, otherwise it constantly reformats the strings to double-quotes.

[dnicolodi]

I'm not a fan of formatters, but if others think that this is useful, I'm not opposed to it.

[rgommers]

I'm fine with adding this setting too. Not a user myself, but if it helps others then a few lines of tool config doesn't hurt.

[eli-schwartz]

Not directly involved in mesonpy development but FWIW...

I'm not a fan of autoformatters either. Actually, when someone does use an autoformatter and that autoformatter is named either "black" or "ruff", I will review their formatting like I review any other code and I will ask them to undo all changes made by those two tools since it looks really ugly.

While Meson does use single quotes, and telling an autoformatter to use single quotes produces output more aligned with Meson's coding style than use of the autoformatter without that setting, it's not enough to make it helpful when developing on Meson.

So, it does depend on whether the meson-python project devs view ruff as "eh, personally I wouldn't bother" or "please revert the style choices the autoformatter just made". In the former case, it may make sense to merge this PR. In the latter case, I'd suggest adding a ruff lint rule under the Q (flake8-quotes) section to check in CI that the correct project-specific quoting style is used, and auto fix them with ruff check --fix.

[eli-schwartz]

It may independently be worth configuring the linter to check the quoting style. Interestingly, ruff doesn't have a consistent configuration schema, so you will have to configure single quotes separately for the linter and the autoformatter or they will produce different results (format will produce code that check errors out on and autofixes to something that format then tries to revert again).

This isn't a bug, it's officially documented behavior that you have to set it twice. :/

[rgommers]

So, it does depend on whether the meson-python project devs view ruff as "eh, personally I wouldn't bother" or "please revert the style choices the autoformatter just made"

What I thought this would be doing is align style choices for code already written by the author of a PR with what pre-commit enforces in CI. I don't touch pre-commit locally, so ruff formatting may cause less conflicts with style choices made in CI.

Either way, yeah the PR diffs should be minimal, that's always important - a formatting should never increase those.

[eli-schwartz]

What I thought this would be doing is align style choices for code already written by the author of a PR with what pre-commit enforces in CI. I don't touch pre-commit locally, so ruff formatting may cause less conflicts with style choices made in CI.

Either way, yeah the PR diffs should be minimal, that's always important - a formatting should never increase those.

Autoformatters traditionally discard all formatting in the file, converting the file to an abstract syntax tree and then re-rendering the entire file from scratch using its own internally preferred style (which I personally believe I have never actually seen anyone organically write -- it will never match any code other than something already processed by the autoformatter).

If any lines in the file still look the same after the autoformatter runs, it's because those lines happened to be what the autoformatter does anyway. Typically this is because the autoformatter is mandatorily enforced on every commit.

It's of course possible to hack around this with janky ecosystem wrappers that use git diff to detect the line number range that the PR modifies, seek out context markers for runs of code that are associated with the modified code (e.g. long tables of dictionary key/value pairs?) and then apply the autoformatter to a buffer. (This is implemented by the formatter as a --line-ranges=1-15 option, and is not idempotent!!! you will get different results depending on the order you format specific sections in, and thus will also produce different results compared to formatting the entire file at once.)

[tobiasdiez]

It may independently be worth configuring the linter to check the quoting style.

That's already check if I'm not mistaken.

Autoformatters traditionally discard all formatting in the file

My strategy is usually to apply the formatter automatically on save but then only stage the lines relevant for the given PR.
I'm also culturally influenced by the js world, where autoformatters are way more common and less controversial.

[eli-schwartz]

It may independently be worth configuring the linter to check the quoting style.

That's already check if I'm not mistaken.

I don't see it in lint.select, are you sure?

I'm also culturally influenced by the js world, where autoformatters are way more common and less controversial.

Sure, but the js world has autoformatters written by people with more taste and less toxic ableism. :)

Black is simply bad at autoformatting, to the people that don't like it. For example, my personal belief is that black is a mixture of:

• changes that autopep8 would already make
• changes where black asked "what's the worst possible way to format", and then implemented that

with no actual exceptions. I wouldn't mind using an autoformatter if the results did what I wanted, rather than made my eyes bleed and destroyed my interest in contributing at all to projects that use that formatting. The point of tools is to automate stuff you want to do, not automate stuff you don't want to do. The JavaScript world understands that -- the python world doesn't (and probably never will).

[dnicolodi]

It may independently be worth configuring the linter to check the quoting style.

That's already check if I'm not mistaken.

I don't see it in lint.select, are you sure?

It's enforced by pre-commit-hook, I think here

meson-python/.pre-commit-config.yaml

Line 21 in 9d46f41

- id: double-quote-string-fixer

Having it enforced by ruff too would not be a bad idea, just to make it easier to check for the ones of us that do not buy into the pre-commit-hook thing.

[dnicolodi]

Please squash the commits.

[eli-schwartz]

It's enforced by pre-commit-hook, I think here

meson-python/.pre-commit-config.yaml

Line 21 in 9d46f41

- id: double-quote-string-fixer

Having it enforced by ruff too would not be a bad idea, just to make it easier to check for the ones of us that do not buy into the pre-commit-hook thing.

Ohhhh I see.

I think it would be worth moving that, yes, given that the pre-commit-hook thing already seems to be running ruff, and that would also speed it up because it doesn't have to run as many distinct shell commands (one per id of that hooks collection thing).

[dnicolodi]

See #757