✔️ Linting Passed

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

_{Generated for commit: ce6bae0. Link to the linter CI: here}

The failure is recurrent (see #28391). It seems to be an excessive memory usage.

Then what do we do with this PR? also cc @lesteve

I just triggered the PyPy build via the web UI, and I guess this is the way forward until it passes. The changes in this PR are unlikely to make the situation worse.

In an ideal world, someone would do something similar as #27662 to find some way to reduce memory usage of the PyPy tests. I don't plan to do this personally in the foreseeable future.

If someone cares strongly about scikit-learn on PyPy, another thing to look at is conda-forge/scikit-learn-feedstock#248 since there seems to be worse issues than only memory (IIRC there were also test failures).

I guess that begs the question: do we really want to support pypy then?

cc @rth maybe?

I guess that begs the question: do we really want to support pypy then?

Let's discuss this point at the next monthly meeting.

Something we could do is officially state that we provide downgraded PyPy support with minimal automated testing: we could only run the test for a small subset of the tests, e.g. just the tests of the sklearn.linear_model module for instance, or just the common tests. This would be a short term workaround for PyPy related maintenance problems.

And we can simultaneously issue a call for volunteer among scikit-learn x pypy users to help us investigate how to reduce memory usage when running the tests on the PyPy runtime similarly to what Loïc did in #27662.

+1 for Oliver's points above. I certainly won't be able to provide support for it :)