Finding a test_importlib refleak on aarch64 RHEL8 #101394

[sobolevn]

This reverts commit 39017e0.

I am experimenting to detect the source of the leak here: #101766

This PR was the first one that has a recorded history of this failure. However, it does not look directly related.

Refs #101394
Refs #101766

• Issue: int and set subclass __sizeof__ under-reports the instance dictionary pointer #101266

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @sobolevn for commit 357c542 🤖

If you want to schedule another build, you need to add the :hammer: test-with-refleak-buildbots label again.

[sobolevn]

I am 95% certain, that this is not the case. Looks like Refleak was not run for a long period of time and it just happened to run on this PR. But, I have to start somewhere!

[corona10]

I would like to recommend you fix the issue rather than reverting right away.

[corona10]

Ah you are trying to check the refleak was caused by this PR.

[sobolevn]

@corona10 yes :)

[sobolevn]

@corona10 @mdickinson wow, I amused to say that the failing job has passed: https://buildbot.python.org/all/#/builders/802/builds/630

[mdickinson]

Hmm. Very interesting. Is there any way we can get access to the machine to debug directly? Or can we at least find out which test in test_import is leaking?

[mdickinson]

It looks as though at least some recent runs (e.g., https://buildbot.python.org/all/#/builders/802/builds/629) that include the changes reverted in this PR do pass. Is it possible that the result is dependent on the order in which tests are run, or the ordering of files on the filesystem, or some such?

[corona10]

@sobolevn It does not explain anything
https://buildbot.python.org/all/#/builders/802/builds/585 was already passed at #101394

[gvanrossum]

Please, please, please. When you submit a PR to revert another PR, please always put in a comment explaining why. "Refs #XXX" is not sufficient. Please spend the extra few minutes to communicate.

[sobolevn]

@gvanrossum Got it! I've updated the description.

[sobolevn]

@corona10 yes, seems like it is a flaky case. Here's another build for the same PR: https://buildbot.python.org/all/#/builders/802/builds/582 And it fails :(

[corona10]

I reproduce the issue on my docker environment

beginning 6 repetitions
123456
......
test_importlib leaked [134, 134, 134] references, sum=402
test_importlib leaked [56, 56, 56] memory blocks, sum=168
test_importlib failed (reference leak)

== Tests result: FAILURE ==

1 test failed:
    test_importlib

Total duration: 598 ms
Tests result: FAILURE
ran 1 tests/2
exit 2
Tests failed: continuing with this subtest

Tests (1):
* test.test_importlib.test_metadata_api.APITests.test_version_egg_info_file

Bisection completed in 16 iterations and 0:00:21
[root@65cf87cd990d cpython]#

[corona10]

Closing the PR since the root PR was found