BUG: officially drop Python <3.12

[lucascolley]

gh-160

[lucascolley]

could you review @crusaderky ?

[betatim]

LGTM

It would be good to have a short summary and such in the description of the PR. I assume people who discover that they can't use Python<3.12 anymore will find this PR and wonder "what the heck?" without one. Yes, we link to the issue where this was discussed but it is a long one, with back and forth, etc. So as a service to people from the future we should have a short summary here that explains why this change was made (not why this bug fix, but why drop support for old Python versions)

[lucascolley]

as a service to people from the future we should have a short summary here that explains why this change was made (not why this bug fix, but why drop support for old Python versions)

This may have to wait until after the meeting on Thursday, or whenever @ev-br is well and able to help. Personally I do not have good justification for dropping these versions instead of keeping __array__ around for a bit to support 3.10 and 3.11.

Of course feel free to write something yourself if you do have that justification @betatim .

[crusaderky]

LGTM, but let's wait until the meeting this Thursday to merge

[ev-br]

LGTM.

[lucascolley]

It was decided in the community meeting to restore support for Python 3.10+ by bringing back __array__ for the versions that need it.

@crusaderky how should we proceed here? Do we want to make a 2.4.1 release which fixes the metadata before yanking 2.4? Or can we just jump straight to 2.5?

[crusaderky]

It was decided in the community meeting to restore support for Python 3.10+ by bringing back __array__ for the versions that need it.

@crusaderky how should we proceed here? Do we want to make a 2.4.1 release which fixes the metadata before yanking 2.4? Or can we just jump straight to 2.5?

I think we should just issue a 2.4.1 that reintroduces __array__ on 3.10/11. There haven't been enough other changes since 2.4 to justify a minor version bump IMHO.
Then we give some time to scikit-learn to upgrade to 2.4.1 and finally yank 2.4.

[lucascolley]

Sounds good to me. Are you happy to make those PRs?