How long the test takes to run? I am asking because if it is low, we can run the test multiple times so it does not magically pass on the main CI and then fail somewhere else. Or is more or less reliable that will fail almost all the time?

@pablogsal: Using this PR, the test is way more likely when run in a loop using:

./python -m test  -F -m test_threads_join_2  test_threading

The CI only runs the test once and so was just lucky. You might tune my PR to make the crash more reliable, but I didn't bother to go that far ;-)

I used this PR to debug https://bugs.python.org/issue36402

I merged my PR #13948 fix which also includes changes of this PR.