A core dump is a very bad sign of health.

Can you please try to bisect which test is causing the segfault using bisect_cmd? Try the command:

./python -m test.bisect_cmd -o bisect --fail-env-changed test_importlib

The command should takes 10-60 minutes, and you should see the number of tests decreasing. I hope that at the end, the command will identify a single method triggering the crash.

At the end, the create "bisect" file contains the test methods causing the bug.

Then you can re-run tests with:

./python -m test --fail-env-changed test_importlib --matchfile=bisect


> The core dump is caused by SIGTRAP. I need help to learn how to stop the core dump from being cleaned up so I can load it into dbx and hopefully understand/learn with sub-test is actually having issues.

You can re-run the test without libregrtest by running it directly. Example:

./python -m test.test_importlib test_spec.Frozen_ModuleSpecMethodsTests.test_reload -v

It should create a core dump in the current directory, or you run the command in your favorite debugger.


My notes on debugging a Python crash: https://pythondev.readthedocs.io/debug_tools.html#core-dump