sanity check

[95-martin-orion]

DO NOT MERGE

Checking if CIBW changes are actually being processed.

[95-martin-orion]

This was enlightening: apparently cp310* does not match any build configs. This disagrees with cibw docs, which suggests that somehow we don't have access to the latest config.

...and looking back at cibw 2.0.1 docs, it's clear that this is the issue. 2.2.2 seems to be the earliest version with explicit 3.10 configs.

[95-martin-orion]

Good news: we've added Python 3.10!

Bad news: everything breaks with Python 3.10!

[95-martin-orion]

Thsi SO issue seems relevant

[95-martin-orion]

Trying numpy==1.16.5 based on scipy-numpy version matching. The scipy version seems to be our complicating factor, but we don't have much control over it from qsim.

[95-martin-orion]

Trying an even newer numpy version since numpy.org indicates numpy did not support Python 3.10 until sometime in the numpy 1.21 era. (Might need to bump to 1.22 if this still fails)

[95-martin-orion]

Ubuntu failure was only on musllinux, which we haven't previously used. Excluding it should resolve ubuntu.

Windows failure log was inscrutable - rerunning to test for transience.

[95-martin-orion]

Windows issue persists, but matches a similar issue described here. Applying the fix described there.

Also disabling the musllinux build with CIBW_SKIP.

[95-martin-orion]

Ubuntu error is resolved, but Windows continues to fail after adding -DMS_WIN64. Unclear if the flag was actually used in compilation; will continue to push on this solution.

[95-martin-orion]

Okay, it appears this is related to the build somehow selecting the ancient pybind11==2.2.4 (from Sept. 2018). Trying to force a more recent pybind version to resolve things.

[95-martin-orion]

Closing - final state is passing checks. See #549 for the less-cluttered change.