Let's just wait for the CI to complete and then merge.

There were some unexpected failures:

• nogil build with segmentation faults in sklearn/metrics/tests/test_pairwise_distances_reduction.py test_memmap_backed_data, no idea why they did not show up in the last scheduled nightly build. I am not updating any dependencies in nogil and the Python nogil branch has no recent changes.
  build log: https://dev.azure.com/scikit-learn/scikit-learn/_build/results?buildId=43845&view=logs&j=67fbb25f-e417-50be-be55-3b1e9637fce5&t=653e8520-555e-5c53-897b-d98df273bc1b. The latest scheduled nogil nightly build with no failure: https://dev.azure.com/scikit-learn/scikit-learn/_build/results?buildId=43840&view=logs&j=67fbb25f-e417-50be-be55-3b1e9637fce5&t=653e8520-555e-5c53-897b-d98df273bc1b
• build with latest dependencies with documentation doctest failure. Fixed by @ogrisel

I also triggered full doc builds to make sure they pass.

Let me try to see if I can reproduce the nogil crash locally to report upstream, independently of this PR.

After the last CI runs:

• the full doc builds passed
• no segmentation fault in nogil this time ...
• the scipy-dev failures are known already and being worked on

Maybe good to merge?

Maybe good to merge?

Yes ;)

no segmentation fault in nogil this time ...

I could not reproduce the segfault on my local nogil setup on scikit-learn main even after updating everything (CPython, numpy, scipy). I will try another run.

Merged!

For information, I ran:

for i in {1..100}; do pytest -vlk test_memmap_backed_data sklearn/metrics/tests/test_pairwise_distances_reduction.py ; done

on my local nogil env and I could not reproduce the segfault...

I could not reproduce the segfault either locally ... let's wait and see if we this failure happens again in one of the nightly build