sergey-dryabzhinsky
debugging windows cp36/38
i've reduced the ci here to just the matrixed workflow, just windows, just cpython, just 3.5 - 3.9. but, i added a 20x repeat on each of those jobs to check consistency of results.
https://github.com/sergey-dryabzhinsky/python-zstd/actions/runs/19155744097/job/54755844937?pr=298
of these, there's complete consistency within a version. each version either passes 20x or fails in the same way 20x. 3.5, 3.7, and 3.9 are passing. 3.6 and 3.8 are failing. specifically, the unittest run is terminating with exit code 3221225477 (0xc0000005). my understanding is that this is a memory access violation.
3.5/3.6 are building with cibw v1 while 3.7-3.9 are v2.
@sergey-dryabzhinsky, any chance you've looked into this before? if not, maybe you have some thoughts or curiosity around this?
this PR is only for debugging. after #297, the windows cp36 and cp38 are the only remaining failures in this workflow. i wanted to run them several times to confirm that the failures are consistent. i removed the other workflows and jobs just to avoid the load on CI.
next tidbit. 3.6 passes the first tests of win32 and fails the amd64 tests. 3.8 also runs win32 first but fails the tests.