Try to fix PyPy39 tests

[gpotter2]

• try to understand why tests fail on pypy39

[codecov]

Codecov Report

Merging #3924 (46744bb) into master (378be02) will decrease coverage by 5.22%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #3924      +/-   ##
==========================================
- Coverage   86.96%   81.74%   -5.22%     
==========================================
  Files         310      327      +17     
  Lines       70619    74712    +4093     
==========================================
- Hits        61417    61077     -340     
- Misses       9202    13635    +4433     

see 45 files with indirect coverage changes

[evverx]

Judging by #3927 669506b seems to be the culprit. With that commit reverted the test suite no longer times out with PyPy. I haven't figured out what exactly caused the regression yet though.

[gpotter2]

Nice catch @evverx. I had some suspicions but that lead me on the right track: it silently broke the config of coverage.py which actually ran coverage on.. all of Python, leading to huge perf losses & some automotive tests to crash.
see 46fde3e

[evverx]

it silently broke the config of coverage.py

Given how hard it is to spot I think it would be great if coverage complained about that. Looking at https://coverage.readthedocs.io/en/stable/config.html it seems all options always go into their own subsections.

see 46fde3e

I have to admit it didn't even occur to me to look at that part of pyproject.toml. My guess was that wheels broke something somehow.