test_import.test_concurrency leaks ref on Windows

[sunmy2019]

On Python 3.12, test_import.test_concurrency() leaks on Windows:

vstinner@WIN C:\victor\python\3.12>python -m test test_import -m test_concurrency -R 3:3
(...)
test_import leaked [22, 36, 28] references, sum=86

bisected to #94504

Potentially Related: #104702

See #102251 (comment) #104796 (comment)

Linked PRs

• gh-106176: Fix reference leak in importlib/_bootstrap.py #106207
• GH-106176, GH-104702: Fix reference leak when importing across multiple threads #108497
• [3.12] GH-106176, GH-104702: Fix reference leak when importing across multiple threads (GH-108497) #108612

[vstinner]

This leak can be seen on the "AMD64 Windows11 Refleaks 3.12" buildbot for example: https://buildbot.python.org/all/#/builders/1103/builds/36

I reproduced it locally.

[vstinner]

cc @exarkun @brettcannon

[Eclips4]

On Python 3.12, test_import.test_concurrency() leaks on Windows:

Did you mean that this test leaks only on 3.12?
I can still reproduce it on current main branch (3.13).

More info

./python -m test test_import -m test_concurrency -R 3:3
Running debug|x64 interpreter...
0:00:00 Run tests sequentially
0:00:00 [1/1] test_import
beginning 6 repetitions
123456
......
test_import leaked [40, 40, 40] references, sum=120
test_import leaked [40, 40, 40] memory blocks, sum=120
test_import failed (reference leak)

== Tests result: FAILURE ==

1 test failed:
    test_import

Total duration: 7.0 sec
Tests result: FAILURE

[sunmy2019]

Did you mean that this test leaks only on 3.12?

No. It also affects 3.12

[exarkun]

I'm not familiar with the leak checking system and I'm not likely to have much time to learn about it in the foreseeable future.

[brettcannon]

@exarkun it just means some things are not getting cleaned up.

[exarkun]

Just trying to signal my likelihood of pitching on here: low.

[vstinner]

Oops, I just reported a duplicate of this bug: #107086 (closed).

I also found the same commit with git bisect: commit 3325f05

commit 3325f054e33b318aa56b74472f76a56b8afc0510
Author: Jean-Paul Calderone <exarkun@twistedmatrix.com>
Date:   Fri Jan 20 19:00:39 2023 -0500

    gh-91351: Fix some bugs in importlib handling of re-entrant imports (GH-94504)

    Co-authored-by: Brett Cannon <brett@python.org>

[vstinner]

A fix was proposed: PR #106207.

[erlend-aasland]

@Yhg1s: IMO, this is also a release blocker (in my opinion, any ref leak is a release blocker).

[Yhg1s]

(Yes, this looks like a real leak, not an artifact of how the refleak checker reruns things, so it should be a release blocker.)

[brettcannon]

We could revert the fix that leads to the leak to unblock the release as I'm not sure if anyone has come up with a good solution to avoid the memory leak and the bugs that get fixed are not fundamental to import (although it would still be nice to have the bugs fixed).

[brettcannon]

I created #108497 to try out an idea that @Eclips4 had for using weakrefs.