✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: cec9b9b. Link to the linter CI: here}

OK, so I could spot two architectures of CPU:

CPU Model: AMD EPYC 7763 64-Core Processor                
Architecture: 9
Physical Cores: 1
Logical Processors: 2
Max Clock Speed: 2445 MHz
Current Clock Speed: 2445 MHz
L2 Cache Size:  KB
L3 Cache Size: 0 KB

and

CPU Model: Intel(R) Xeon(R) Platinum 8272CL CPU @ 2.60GHz
Architecture: 9
Physical Cores: 2
Logical Processors: 2
Max Clock Speed: 2594 MHz
Current Clock Speed: 2594 MHz
L2 Cache Size:  KB
L3 Cache Size: 0 KB

The failure seems therefore related to MKL and the AMD CPU. I'll switch to openblas to see if the tests are passing and we might want to try reporting the issue.

Could you please update the lock file only for the windows build instead? There is a commandline option in the script to do so. Use --help to find its name.

I pushed a commit changing only the Windows lock-file and renaming the lock-file to have openblas rather than mkl.

Let's see what the CI says 🤞.

Windows CI passed and it ran on the problematic AMD EPYC 7763 64-Core Processor build log where we have seen failures and timeout occur consistently.

Merging this to be able to make other PRs go forward.

One possible improvement for a later PR would be to put the CPU info on Windows in the same place as we do lscpu in build_tools/test_script.sh.

In an ideal world, we would also further understand the issue but this seems time-consuming (would first need a reproducer with scipy/numpy only, and maybe eventually a reproducer in C) and if in the end if this is indeed a MKL issue, I am not sure where to report.

Note for completeness: I was able to reproduce by using action-tmate to have an interactive SSH session in a GHA runner (same AMD CPU as in Azure), see https://github.com/mxschmitt/action-tmate?rgh-link-date=2025-06-24T14%3A01%3A52Z.

Thanks @lesteve to have finish this one.

After some time of silence, the jury of the maintainability award judges this PR worth of the 5th trophy 🏆 (after #25069 (comment)). As the Nobel committee does, the price is split between @glemaitre and @lesteve.

Race:
@glemaitre and @lesteve both 1.5 , @jeremiedbb 1, @thomasjpfan 1.

The jury kindly asks for support about future suggestions of nominees.

Not sure about a maintainability award for this one, maybe more of a firefighting mode award 😅.

This may blow up in our face one day ¹ but oh well, for now things are back to normal, let's hope for some time 🤞.