[3.14] gh-128639: Don't assume one thread in subinterpreter finalization (gh-128640)

[miss-islington]

Incidentally, this also fixed the warning not showing up if a subinterpreter wasn't
cleaned up via _interpreters.destroy. I had to update some of the tests as a result.
(cherry picked from commit 9859791)

Co-authored-by: Peter Bierma zintensitydev@gmail.com

• Issue: Subinterpreters don't properly clean up threads #128639

[hugovk]

Let's check the buildbot failure before merging this: #128640 (comment)

[ZeroIntensity]

That failure looks related, but I have absolutely no idea how.

[vstinner]

I suppose that this change should not be merged, since you proposed a revert for the same change in the main branch: #134256

[ZeroIntensity]

!buildbot iOS

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @ZeroIntensity for commit 76bcc46 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F134254%2Fmerge

The command will test the builders whose names match following regular expression: iOS

The builders matched are:

• iOS ARM64 Simulator PR

[ZeroIntensity]

I'm going to run the buildbots to see if the crash is somehow related to something else on 3.15. I'm still pretty lost on what could be the cause. I can only see two possibilities:

• We revealed an existing bug with _PyThreadState_DeleteList and/or _PyThreadState_RemoveExcept.
• The embedding test revealed a bug with the implementation, but I'm not sure what. Maybe I broke destroying interpreters with only one thread state?

[mhsmith]

I'm not sure if it's related, but the iOS buildbot has also been experiencing another intermittent subinterpreters failure: #127108. The most recent instance was in #125962 (comment).

[ZeroIntensity]

In this specific case, the change was related. It's reverted on main. I haven't had time to look into those other failures yet.

[ZeroIntensity]

Closing in favor of #134606. We'll backport that one instead.