$ ./python -m test -R 3:3 test__xxsubinterpreters -m test_ids_global
0:00:00 load avg: 0.80 Run tests sequentially
0:00:00 load avg: 0.80 [1/1] test__xxsubinterpreters
beginning 6 repetitions
123456
......
test__xxsubinterpreters leaked [1, 1, 1] references, sum=3
test__xxsubinterpreters leaked [1, 1, 1] memory blocks, sum=3
test__xxsubinterpreters failed

== Tests result: FAILURE ==

1 test failed:
    test__xxsubinterpreters

Total duration: 819 ms
Tests result: FAILURE


It started to leak since:

commit 7dd549eb08939e1927fba818116f5202e76f8d73
Author: Paulo Henrique Silva <[email protected]>
Date:   Tue Mar 24 23:19:58 2020 -0300

    bpo-1635741: Port _functools module to multiphase initialization (PEP 489) (GH-19151)

The folowing test is enough to reproduce the leak:

    def test_ids_global(self):
        interp1 = interpreters.create()
        script, rpipe = _captured_script("pass")
        interpreters.run_string(interp1, script)
        rpipe.close()

        interp2 = interpreters.create()
        script, rpipe = _captured_script("pass")
        interpreters.run_string(interp2, script)
        rpipe.close()

I've got it, will investigate asap.

The module still uses static state. Fixed the leak and will convert it to use per-module state in a separate issue.

New changeset b09ae3ff43111a336c0b706ea32fa07f88c992d9 by Paulo Henrique Silva in branch 'master':
bpo-40071: Fix refleak in _functools module (GH19172)
https://github.com/python/cpython/commit/b09ae3ff43111a336c0b706ea32fa07f88c992d9

> (...) will convert it to use per-module state in a separate issue.

Great!

As discussed on PR19172, this module uses a global state in functions that do not receive a PyModule* and right now converting such cases to per-module state is not trivial. I will wait for PEP-573 implementation that will hopefully make this easier.

Pushed PR19273 to avoid the potential crash induced by the original change.

New changeset eacc07439591c97f69ab4a3d17391b009cd78ae2 by Paulo Henrique Silva in branch 'master':
bpo-40071: Fix potential crash in _functoolsmodule.c (GH-19273)
https://github.com/python/cpython/commit/eacc07439591c97f69ab4a3d17391b009cd78ae2

I closed the issue, the leak is now fixed and _functools has been fixed.

I created bpo-40137: TODO list when PEP 573 "Module State Access from C Extension Methods" will be implemented.