Add suffix function to the fs module

[joukewitteveen]

When iterating over a list of files, it can be convenient to check suffixes. You could already do fs.name(file).endswith('.txt'), but fs.suffix(file) == '.txt' is just that tiny bit nicer. We also get the invariant fs.name(file) == fs.stem(file) + fs.suffix(file).

[joukewitteveen]

The corner case of an 'empty-but-present suffix' (e.g. foo/bar.) will change behavior in Python 3.14. I've removed our test for it.

[dcbaker]

Everything looks good to me other than the one very small nit, so this should be good to go.

[bonzini]

The corner case of an 'empty-but-present suffix' (e.g. foo/bar.) will change behavior in Python 3.14.

Should Meson instead use already os.path.splitext so that it already uses the 3.14 behavior?

[nirbheek]

Should Meson instead use already os.path.splitext so that it already uses the 3.14 behavior?

I agree, and also we should have a test for this so the behaviour stays the same regardless of Python version.

[dcbaker]

pathlib for concrete paths has a lot of ugly and painful corners (the purepaths are much better), I'd prefer to use os.path in general if possible

[joukewitteveen]

I'll try to find some time end of next week to redo the module in terms of os.path 👍.

[joukewitteveen]

There was a single call to mesonlib.path_is_in_root that still required importing pathlib.Path. I don't think this PR is the place to change the signature of path_is_in_root, so I kept it. Otherwise, the module now directly calls into os.path. I used the opportunity to unquote all type specifiers (all function bodies were touched).

[joukewitteveen]

The failing checks are due to a bug in ntpath.isabs that was fixed in Python 3.13.

I'm not sure what to do. We cannot use os.path.splitroot, because it is only available from Python 3.12 onward. Using a PurePath for just this function seems excessive. I'll need to think a bit, suggestions are welcome!

[bonzini]

Using a PurePath for just this function seems excessive

Maybe but I don't see any alternative. Just use a PurePath and add a link to python/cpython#44626 in a comment. Thanks for looking into the failures!

[joukewitteveen]

How about this fix? We would otherwise (and currently are) be affected by a related bug where incomplete UNC paths (which are absolute because they don't depend on the current working directory) would not be considered absolute.

[joukewitteveen]

The previous test failure was due to an implicit assumption that replace_suffix changed the alternate path separator on Windows (/) to \. Regardless of whether we want that, I think the assumption should not be part of the specific test that failed, so I watered that test down (by using \ both before and after replacement, the test passes with both implementations).

Let me know if we want to reinstate the separator-changing behavior of replace_suffix. I think we shouldn't.

[joukewitteveen]

The CI failures seem unrelated. Before my last push, where I corrected some authorship details, all checks passed. Maybe someone can restart the unsuccessful checks?