BLD: revert adding PEP 621 metadata, it confuses setuptools

[rgommers]

See https://github.com/numpy/numpy/pull/22663/files#r1034053192.

The one setup.py tweak is to even make it possible to test newer setuptools versions without build isolation (applies fix suggested in gh-21288).

[rgommers]

Argh, now devpy is unhappy: No project section found in pyproject.toml. @stefanv can that check be removed? There should be no need to use the [project] table, and it's unhelpful to check because it requires adopting PEP 621 support, which then causes probems with setuptools. So during a transition from setuptools to meson, this will be a pain point.

[stefanv]

Argh, now devpy is unhappy: No project section found in pyproject.toml. @stefanv can that check be removed?

Done

[eli-schwartz]

Random bash scripting observation.

[stefanv]

Current test failure is:

 ImportError while loading conftest '/home/runner/work/numpy/numpy/build-install/usr/lib/python3.11/site-packages/numpy/conftest.py'.
numpy/__init__.py:109: in <module>
    from ._globals import (
numpy/_globals.py:20: in <module>
    from ._utils import set_module as _set_module
E   ModuleNotFoundError: No module named 'numpy._utils'

I'm confused; I don't see any mention of _utils in _globals.py.

[seberg]

@stefanv we just added _utils yesterday, sounds like a git clean issue where you got a leftover _utils from the main branch flying around.

[seberg]

Ahh, I guess weird PR timing there, the meson config just doesn't know about _utils yet.

[seberg]

OK, I just added _utils to the meson conf, since it makes CI fail on everything now otherwise.

[mattip]

Can you remove the upload_wheels changes? That will be part of #22690.

[stefanv]

Can you remove the upload_wheels changes? That will be part of #22690.

Done.

[stefanv]

This is probably ready to merge then?

[mattip]

I must be missing something. Where is the information that is commented out going to live? Is it duplicated elsewhere?

Edit: yes, it is in the setup.py

[mattip]

LGTM. Thanks @rgommers and @stefanv

[mroeschke]

So with this change it appears that strictly setuptools==59.2.0 is required now to install the nightly numpy wheel?

We have some jobs in pandas to test with the nightly wheels and are encountering this error

https://github.com/pandas-dev/pandas/actions/runs/3589194341/jobs/6041379775

Successfully installed pip-22.3.1 setuptools-65.6.3 wheel-0.38.4
Looking in indexes: https://pypi.anaconda.org/scipy-wheels-nightly/simple
Collecting numpy
  Downloading https://pypi.anaconda.org/scipy-wheels-nightly/simple/numpy/1.25.0.dev0%2B79.g01d64079b/numpy-1.25.0.dev0%2B79.g01d64079b.tar.gz (10.9 MB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 10.9/10.9 MB 13.6 MB/s eta 0:00:00
  Installing build dependencies: started
  Installing build dependencies: finished with status 'error'
  error: subprocess-exited-with-error
  
  × pip subprocess to install build dependencies did not run successfully.
  │ exit code: 1
  ╰─> [3 lines of output]
      Looking in indexes: https://pypi.anaconda.org/scipy-wheels-nightly/simple
      ERROR: Could not find a version that satisfies the requirement setuptools==59.2.0 (from versions: none)
      ERROR: No matching distribution found for setuptools==59.2.0
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
error: subprocess-exited-with-error

[rgommers]

That action indeed includes the changes from this PR. It's unlikely that this caused the problem though - this reverted an earlier changed, and all that is left in pyproject.toml is a bunch of commented-out code.

==59.2.0 was always required for an isolated build, and never for a non-isolated build. That should not have changed.

The real question here is why you're getting a build from source, rather than picking up a wheel.

[rgommers]

The problem is that some macOS wheel builds failed, but an sdist was uploaded to the nightly bucket. @charris @mattip I don't think we should upload an sdist to nightlies at all, that is going to result in these kinds of problems as Pandas is seeing. If a nightly wheel is missing for some platform, one should get a wheel built a week before.

[rgommers]

@mroeschke I've deleted the most recent sdist, so I think your next CI run should not fail. Could you confirm that after it has run?

[mroeschke]

Ah thanks @rgommers. A recent build has started to pass now! https://github.com/pandas-dev/pandas/actions/runs/3596938344/jobs/6058167508

[charris]

Does this need a backport to 1.24.x?

[rgommers]

Sorry I missed this question. Yes, I think it's a good idea to backport this change. I'll open a PR now.

[rgommers]

(bit late, sorry about this. it can go into the next bug fix release)

[rgommers]

gh-22820