With that PR, test_capi is no longer leaking:

admin@Admins-MacBook-Air ~/p/cpython (revert-of-revert)> ./python.exe -m test -R 3:3 test_capi
Using random seed: 992095454
0:00:00 load avg: 26.10 Run 1 test sequentially in a single process
0:00:00 load avg: 26.10 [1/1] test_capi
beginning 6 repetitions. Showing number of leaks (. for 0 or less, X for 10 or more)
123:456
XX. ...
test_capi passed in 43.3 sec

== Tests result: SUCCESS ==

1 test OK.

Total duration: 43.4 sec
Total tests: run=985 skipped=67
Total test files: run=1/1
Result: SUCCESS

gh-125512: Revert #125513 and fix reference cycle #125532

It's not a revert, but a revert of a revert :-D The PR title should be gh-124872: Replace enter/exit events with "switched.

I suggest to wait after Python 3.14.0 alpha1 release to retry this change.

I suggest to wait after Python 3.14.0 alpha1 release to retry this change.

Why? We can debug and fix the ref leak.

@rhansen Please build CPython with ./configure --with-pydebug and run the tests with -R3:3. That will take much longer so I suggest just running asyncio & contextvars & decimal tests with this flag.

me:

I suggest to wait after Python 3.14.0 alpha1 release to retry this change.

@1st1:

Why?

It's just that there are many broken buildbots, I think that this change can wait for 3.14.0a2.

FYI, I ran the test suite in hunt refleaks mode locally and it succeeded:

== Tests result: SUCCESS ==

25 tests skipped:
    test.test_asyncio.test_windows_events
    test.test_asyncio.test_windows_utils test.test_gdb.test_backtrace
    test.test_gdb.test_cfunction test.test_gdb.test_cfunction_full
    test.test_gdb.test_misc test.test_gdb.test_pretty_print
    test.test_multiprocessing_fork.test_manager
    test.test_multiprocessing_fork.test_misc
    test.test_multiprocessing_fork.test_processes
    test.test_multiprocessing_fork.test_threads test_android
    test_dbm_gnu test_devpoll test_epoll test_free_threading
    test_launcher test_msvcrt test_perf_profiler test_perfmaps
    test_startfile test_winapi test_winconsoleio test_winreg test_wmi

11 tests skipped (resource denied):
    test_curses test_peg_generator test_pyrepl test_smtpnet
    test_socketserver test_tkinter test_ttk test_urllib2net
    test_urllibnet test_winsound test_zipfile64

442 tests OK.

Total duration: 20 min 5 sec
Total tests: run=44,347 skipped=2,182
Total test files: run=467/478 skipped=25 resource_denied=11
Result: SUCCESS

@Eclips4 thanks for fixing the leak!

@vstinner

It's just that there are many broken buildbots, I think that this change can wait for 3.14.0a2.

I sympathize, really sorry about breaking them and causing pain :/ Can we try landing this PR since it appears it should solve the issue? I'm afraid if we wait on this for a month to get it merged it will go stale and we just forget. I'd try again and if it fails we wait? If however you really don't want to do it, then fine.

Since we're about to release a1 tonight, I'd wait with landing this until right after a1. We won't forget about it.

Actually, @hugovk should make the call.

Yeah, same call, thanks :)

@ambv @vstinner

Since we're about to release a1 tonight, I'd wait with landing this until right after a1. We won't forget about it.

Oh I was reading the schedule wrong. I didn't realize Victor wants to delay it just one day, essentially, I thought we were talking one month to merge this. Absolutely can wait a few days.

Thanks, a1 is now out, merge whenever you're ready.

I applied @rhansen's change, before that, the test did nothing :-)

Also, would you mind fixing up the commit message, authorship info, etc. on the first commit:

That sounds like a good idea :-) Or at least, make sure that @rhansen is mentioned as a co-author of the change.

Apologies for the breakage, and thank you for debugging!

Is a blurb required now that there's an API change between v3.14.0a1 and v3.14.0a2?

Also, would you mind fixing up the commit message, authorship info, etc. on the first commit:

git -c sequence.editor='sed -i -e "1s/^pick/edit/"' rebase -i main &&
git commit --amend -C 843d28f59d2 &&
git rebase --continue

Sorry about that! I definitely will do it.

@rhansen it seems I have done it. If you agree, I will merge it.

(I'm actually do not understand why for f0b3aa862b629690496d0336904a9bb1ada6dd47 github is treating me as co-author...)

(I'm actually do not understand why for f0b3aa8 github is treating me as co-author...)

GitHub is correct when saying:

rhansen authored and Eclips4 committed

Should this get tested with the refleak buildbots?

Should this get tested with the refleak buildbots?

I don't think so. I tested test_capi manually, and @Eclips4 ran the whole test suite with -R 3:3.