Setuptools version ceilings?

After the release of setuptools 60.0.3 yesterday, all the Python 3.7 and 3.8 tests started failing on Dask. This is because the versions of pandas and numpy in those envs hadn’t yet moved to using packaging.version. This is the error:

py.test dask --runslow --cov=dask --cov-report=xml -n4 
ImportError while loading conftest '/home/runner/work/dask/dask/conftest.py'.
conftest.py:28: in <module>
    import pandas  # noqa: F401
/usr/share/miniconda3/envs/test-environment/lib/python3.7/site-packages/pandas/__init__.py:22: in <module>
    from pandas.compat.numpy import (
/usr/share/miniconda3/envs/test-environment/lib/python3.7/site-packages/pandas/compat/numpy/__init__.py:10: in <module>
    _nlv = LooseVersion(_np_version)
/usr/share/miniconda3/envs/test-environment/lib/python3.7/site-packages/setuptools/_distutils/version.py:57: in __init__
    stacklevel=2,
E   DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead.
Error: Process completed with exit code 4.

It feels like this could be fixed on the conda-forge side, but I think the correct fix would be to update all the old recipes for all packages and explicitly add setuptools with a ceiling. But that seems impossible.

Here is the dask PR where I am setting a ceiling on setuptools for these older envs: https://github.com/dask/dask/pull/8509

[beckermr]

The linter runs in our heroku server. Conda build is run in azure etc. So we’d need a secure way to report statuses back to github without putting keys into public forks etc. I think this is just too hard right now.

[jaimergp]

(conda plugins would be sooo useful here)