When you're done making the requested changes, leave the comment:I have made the requested changes; please review again.

Hmm, looks like this commitb0fcc2c has causedtest.test_gc.IncrementalGCTests.test_incremental_gc_handles_fast_cycle_creation failures for iOS and Android:

• iOS ARM64 Simulator:https://buildbot.python.org/#/builders/1380/builds/1907
• AMD64 Android:https://buildbot.python.org/#/builders/1591/builds/516
• aarch64 Android:https://buildbot.python.org/#/builders/1594/builds/600

Reminder the 3.14 alpha 2 is due tomorrow (2024-11-19):https://buildbot.python.org/#/release_status

cc@freakboy3742@mhsmith

Extracted from the Android failure:

FAIL: test_incremental_gc_handles_fast_cycle_creation (test.test_gc.IncrementalGCTests.test_incremental_gc_handles_fast_cycle_creation)----------------------------------------------------------------------          Traceback (most recent call last):                                                File "/data/user/0/org.python.testbed/files/python/lib/python3.14/contextlib.py", line 85, in inner    return func(*args, **kwds)                                                    File "/data/user/0/org.python.testbed/files/python/lib/python3.14/test/test_gc.py", line 1175, in test_incremental_gc_handles_fast_cycle_creation     self.assertLess(new_objects, 25_000)                                            ~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^                                        AssertionError: 25032 not less than 25000                                                                                                                       ----------------------------------------------------------------------          Ran 53 tests in 3.744s                                                                                                                                          FAILED (failures=1, skipped=8)                                                  test test_gc failed                                                             test_gc failed (1 failure)                                                      1 test failed again:                                                                test_gc

Also some tier 1 and 2 "test_import failed (reference leak)" that include this change (and some others) are now failing:

• AMD64 RHEL8 Refleaks:https://buildbot.python.org/#/builders/259/builds/1712
• aarch64 RHEL8 Refleaks:https://buildbot.python.org/#/builders/551/builds/348

Plus s390x RHEL8 + RHEL9 and AMD64 FreeBSD refleaks buildbots.

I'll open a draft revert PR so we can run the buildbots on it.

Edit:#126983

Hmm, looks like this commitb0fcc2c has caused test.test_gc.IncrementalGCTests.test_incremental_gc_handles_fast_cycle_creation failures for iOS and Android:

I don't have ready access to an iOS or Android device to test on.
The failing test is designed to check that the heap doesn't keep growing. If the initial heap size is significantly smaller than on linux/windows then the heap would grow a bit more before stabilizing. Setting the limit to 27k or 30k should work.
It might be worth double checking that the heap doesn't continue to grow by increasing the number of iterations from 20k to 100k+

I've done some local testing on iOS. The test_gc module passes as-is if you run it by itself. If I increase the loop count in the failing test to 100k, the test fails with exactly the same object count (25,032).

My guess is that this one of those issues that iOS/Android expose because they don't run the test suite in parallel - my guess is that you'd likely be able to reproduce this failure on a desktop machine if you run the test suite with--single-process.

On that basis, following@markshannon's suggestion I'll increase the object count to 26k. PR incoming.

#126984 is a PR increasing the test threshold.

I've abandoned the "increase the threshold" aapproach - that's clearly not enough to avoid the issue. The threshold needed to pass the test is clearlymuch higher than 30k.

The good news (well... good for me, not so good for@markshannon 😄 ) is that I can reproduce this bug on macOS. If you run the test suite sequentially in a single process (./python.exe -m test -W --single-process),test_gc fails with the same error on macOS as is being reported on iOS and Android.

Youdon't see this error if you runtest_gc in isolation - the 25k check passes. This is also consistent with the behavior on iOS and Android. You have to run the full sequence of (alphabetical) tests up totest_gc to see the failure. I'll take a quick pass to see if I can narrow down a smaller (and faster) reproduction case.

I'm able to reliably reproduce the failure on macOS with./python.exe -m test test_email test_fnmatch test_frame test_gc.

I've reverted this because it's release day and we need to keep tier-1 green anyway:#126983.

Let's make sure to summon the buildbots next time :)

I'm able to re-produce on Linux with this set of tests:test_dis test_email test_funcattrs test_functools test_gc.