`uv pip install` fails to install package while `pip install` succeeds

[arpitg-sh]

Description: I'm encountering an issue when using uv pip to install Python packages. Specifically, the build process for jsmin==2.2.2 fails, while doing the same with native pip passes.

Steps to Reproduce

Using uv:

uv pip install setuptools
uv pip install azure-cli

This results in a failure:

9.923   × Failed to download and build `jsmin==2.2.2`
9.923   ╰─▶ Build backend failed to determine requirements with `build_wheel()`
9.923       (exit status: 1)
9.923
9.923       [stderr]
9.923       /root/.cache/uv/builds-v0/.tmpdkIMOq/lib/python3.9/site-packages/setuptools/_distutils/dist.py:261:        
9.923       UserWarning: Unknown distribution option: 'test_suite'
9.923         warnings.warn(msg)
9.923       error in jsmin setup command: `use_2to3` is invalid.

Using regular pip(v24.0):

pip install setuptools
pip install azure-cli

This works successfully without errors.
Additional Information

uv Platform: "Debian GNU/Linux 10 (buster)"
uv Version: uv 0.5.1

[dimbleby]

You will want to allow prereleases when adding azure-cli

[charliermarsh]

Yeah adding --pre seems to get this to work. I think this is correct as-is -- without --pre, we end up trying to build a dependency but failing.

[charliermarsh]

Actually, I think this is failing during pre-fetching, maybe? I have a separate solution to this, so I'll leave it open.

[dimbleby]

It's failing because when pre-releases are disallowed, uv searches way back to old versions of azure-cli that no-one should want.

I think it's understood (by those who know!) that uv takes a different view of pre-releases than other installers, and different also than the specs. Of course there are reasons, but this is an example of the downsides.

[charliermarsh]

Yeah -- I generally agree. The thing I noticed in this specific case though is that we don't actually seem to ever consider jsmin, we just "pre-fetch" it when looking at certain azure-cli versions. So we should be able to delay the error here such that the resolution still succeeds (but would fail, as today, if the resolution proceeded differently and we did end up looking at jsmin as a candidate).

[arpitg-sh]

@charliermarsh Do you know when will this be released?