TSan CI is flaky

[ngoldbaum]

I've noticed that the TSan CI is failing regularly. See for example, this CI run on a recent PR:

https://github.com/python-cffi/cffi/actions/runs/24686671067/job/72197736127?pr=241

Unfortunately there's an issue with the docker images we're using in our tests that is preventing the tracebacks in the race report from being useful:

nascheme/cpython_sanity#12

I'm going to try to reproduce this manually without the docker image to see if I can get more detail.

EDIT: full race report with tracebacks here: https://gist.github.com/ngoldbaum/566cc486a9e16ebb36a05c435f7944cd

[ngoldbaum]

I'm unable to reproduce this race on my Mac with a 3.14 interpreter, so I'm still not able to track down exactly what the issue is.

That said, I tried updating the image used in CI to run on a 3.15 interpreter and that also seems to fix it. See #243.

[ngoldbaum]

I was able to reproduce this on a Linux x86_64 machine using the docker container just like the CI job. After installing llvm on the container, I get useful tracebacks in the TSan report: https://gist.github.com/ngoldbaum/566cc486a9e16ebb36a05c435f7944cd.

I think this is hitting python/cpython#113956. I'll comment on the upstream issue and add a new TSan suppression for this.

[ngoldbaum]

I closed #243 in favor of #244 which just adds a suppression.