feat(ci): Run tests for more python versions

[hassec]

I wanted to disentangle the fixing of builds for specific python versions from #293.

So this PR keeps the setup.py way of building OMAS and does not yet fix anything in regard to the Cython warnings.

All that this PR does is update the CI to also run the tests for python versions 3.6-3.11,
and apply some small fixes or workarounds to make the OMAS tests pass for these python versions.

We only use OMAS with python >=3.10 so would love to see these tests as part of the CI to ensure everything works as expected on these versions too.

[orso82]

Thank you @hassec !! Fantastic! @smithsp @kalling can you please take a look as well? Then I think we can merge!

[kalling]

Looks good as far as I can tell.

[kalling]

@orso82 Any idea if this has been fixed on the OMFIT side? I assume we should just cast it?

[orso82]

Probably this needs to be fixed on the OMFIT side. Could you please take care of it @kalling ?

[kalling]

@orso82 Alright.
@hassec Would you mind sending me (or pointing me to) a trace of the error that occurs if you don't skip this test?

[hassec]

@kalling no problem. It's in omfit_gapy.py which has the following lines:

        cplot = max([np.floor(np.sqrt(nplot)), 1.0])
        rplot = np.ceil(nplot / cplot)

which should probably we wrapped in int(...)

Corresponding ci run with the error is here

[kalling]

cplot is used in other calculations later that look like they may need it to be a float, so I submitted a change to OMFIT auto_merge that stores the cast versions passed in to subplot as plot_cols and plot_rows instead of doing the cast at assignment, just to be safe. Re-ran the tests locally and it took care of the issue.

[hassec]

@kalling are you sure? The latest version on pypi is from Nov 2023

[hassec]

@kalling @orso82 How would you feel about merging as is, and I can follow up as soon as the fix on omfit_classes is released on PyPi? Or if it's coming in the next days I'm also happy to wait.

There isn't really a strong urgency, but I'd like to avoid that the PR sits too long, goes stale and gets forgotten.

[kalling]

@hassec Interesting, I wonder if the 2 factor auth introduced messed up the automation for uploading to pypi. It should currently be at 3.2024.09
I'll take a look.

[kalling]

Ok, I fixed the pypi uploading issue for omfit-classes and 3.2024.9.2 was uploaded. Hopefully now it'll actually be weekly again :) (Assuming tests pass)
Thanks for pointing that out.

[hassec]

Thanks @kalling I'll have an update PR for OMAS ready asap to remove the workaround :)

[smithsp]

Please double check that all of the desired tests run for all of the python versions requested.

[hassec]

@smithsp was that comment for me? What would you like to see checked aside from what's shown in the CI?

[kalling]

I verified the tests work locally with those OMFIT changes. Testing here would probably need a new omfit_classes release as @hassec mentioned as a comment in my review.

[smithsp]

@hassec regarding #296 (comment) , yes, I am just asking that you double check the CI outputs to be sure the 3.7 tests run the same tests as the other versions.

[hassec]

@smithsp yes, on all versions we collect 141 tests.
3.6 skips 13
3.7 skips 14 ( this is the omfit_classes bug we will wait for and as soon as that is on pypi we can remove the skipping of that additional test)
For python >= 3.8 it skips 15 tests. This is higher due to omfit + xarray problem.