sgkit-dev / sgkit

Scalable genetics toolkit
https://sgkit-dev.github.io/sgkit
Apache License 2.0
217 stars 32 forks source link

Turn off numba cache by default #1191

Closed jeromekelleher closed 4 months ago

jeromekelleher commented 5 months ago

Partial fix for #1156

This turns off numba caching by default, and at least lets us use the current git version of sgkit without hitting this issue repeatedly.

To fully address I think we probably need to undocument this variable by removing the mention to it in the docs, and maybe add some comments in the source laying out the issue.

jeromekelleher commented 5 months ago

Hmm, it seems like there's an unfortunate side effect here of turning off caching. With caching turned off, when we run the functions with NUMBA_DISABLE_JIT=1 to get coverage we get a whole bunch of numba errors and deprecations that don't happen when caching is on.

I'm not sure what to do here, there's quite a few deprecations and errors that seem non-trival to fix. For example, we seem to hit this problem https://github.com/numba/numba/issues/8886 which is non-trival to workaround.

Any thoughts @tomwhite, @timothymillar?

jeromekelleher commented 4 months ago

I worked around this here by explicitly turning on caching, so that we can get our builds to pass and at least have our code not segfault by default. Opened #1194 to track the CI issue.

codecov-commenter commented 4 months ago

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Comparison is base (a386541) 100.00% compared to head (5c85a92) 100.00%.

:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files ```diff @@ Coverage Diff @@ ## main #1191 +/- ## ========================================= Coverage 100.00% 100.00% ========================================= Files 52 52 Lines 5210 5210 ========================================= Hits 5210 5210 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.