@echedey-ls do you have any idea why the failing py 3.10-min job is trying to install numpy==1.19.3 instead of 1.21.2? It makes no sense to me. https://github.com/pvlib/pvlib-python/actions/runs/17867208211/job/50812293647?pr=2547#step:4:213

Definitely black magic. Citing the logs:

Installing pip packages: h5py==3.0.0, numpy==1.21.2, pandas==1.3.3, scipy==1.7.2, pytest-rerunfailures, pytest-remotedata, requests-mock
    error: subprocess-exited-with-error
    
    × pip subprocess to install backend dependencies did not run successfully.
    │ exit code: 1
    ╰─> [956 lines of output]
        Collecting numpy==1.19.3
          Downloading numpy-1.19.3.zip (7.3 MB)

In https://github.com/mamba-org/setup-micromamba
There is this link in the readme, to redirect users that want to know caching better (both for downloading and environment): mamba-org/provision-with-micromamba#38 (comment)
In the end, a user reports his issue is the same as this one: mamba-org/setup-micromamba#50 (short conversation, must read all through it).

I know we have:

		uses: mamba-org/setup-micromamba@v2
        with:
          cache-downloads: false
          cache-environment: false

However, I guess it is caching things, what a coincidence that it's the last numpy version... I won't be able to trace the problem down, so I suggest we wait for 8 days till the cache gets deleted, or just merge as is. I believe in this diff more than that action docs. We could also try to use bare python for the minimal CI - that would require translating the ci files into requirements.txt syntax thou.

It makes no sense to me.

Neither to me. Can we delete .github? I beg.

Ah ha, it was an indirect pin via h5py==3.0.0. Increasing to a newer h5py version fixed it. Thanks for taking a look!

Well done and thanks to you Kevin, I thought newer pip reported these incoherencies...

Merging since this is just a routine update.