Have relative_to return PosixUPath

[CompRhys]

Close #214

Test that should pass if the local path code worked as expected

[ap--]

Thank you so much for the PR! 🙏 ❤️

I made a few comments. Let me know what you think.

Cheers,
Andreas

[ap--]

It would be great if you could add a test that checks if a ValueError is raised if a UPath of a different protocol is provided as argument.

And a test that checks the cases for the walk_up keyword argument.

[CompRhys]

added a test for s3, not sure what walk_up allows will need more time to look at that to write a test case

[CompRhys]

added a test for walk_up seems like it's broken for azure

[CompRhys]

seemingly its because UPath("az:///other_test_bucket").parents is an empty list which I don't understand

[ap--]

You can prevent some of the failing tests, by editing

universal_pathlib/setup.cfg

Lines 22 to 28 in 380144c

	[options]
	python_requires = >=3.8
	zip_safe = False
	packages = find:
	install_requires=
	fsspec>=2022.1.0

to use: fsspec >=2022.1.0,!=2024.3.1

This is due to a incompatibility of UPath's test suite with the newest fsspec.

[bolkedebruin]

Where are we with this? I would be nice to have this in, because the current behavior is broken :-).

[ap--]

With the new fsspec release, this is ready to be merged. I'm on holiday until the end of May and will prepare a new release when I'm back 😊

[bolkedebruin]

Great!

[bolkedebruin]

Thinking and testing this a little bit further (from an Airflow perspective, but I don't think that matters). The following does not seem to make sense:

o = ObjectStoragePath("s3://tmp/test")
t = ObjectStoragePath("s3://tmp/test/subdir/bla/bla.txt")

t.relative_to(o).resolve()

prints PosixUPath('/Users/bolke/dev/airflow/subdir/bla/bla.txt')

which doesn't make sense either way. If posix it should resolve to tmp/test/subdir/bla/bla.txt or if s3 (that should make more sense I guess?) to s3://tmp/test/subdir/bla/bla.txt.

Isn't it better to return a PurePath or keep a reference to the original?

[bolkedebruin]

What I am considering is to have

ObjectStoragePath.relative_to return itself (e.g. an ObjectStoragePath) with a storage_option of parent set to the self.

__str__ then reproduces the relative path by checking if parent was set, but all other mechanisms intact.

[CompRhys]

coming back to clean up some old PRs, completely no idea what I was attempting to do here as was for an old job but have merged the recent changes and rebased.

Let me know if I can help

[scholtalbers]

Would be nice to have this resolved. I agree that returning PosixUPath is not ideal, but at least its better than the current state.

[ap--]

Closing in favour of #405.

Similar to what @bolkedebruin suggested, but currently with a private base path.

Could think about exposing it via a special storage option as suggested.