Upgrade to Cython 3+ in building

[djhoese]

This PR upgrades the build dependencies to use Cython 3+. It also adds extra annotations required by Cython 3 to avoid extra Python-related checks and stay in C-land longer and with no GIL. As far as I can tell nothing of the generated code should be different to what it was.

Note I had concerns because the modis interpolation has some low-level functions with GIL-related when memory views are returned. I made a stackoverflow question about it and a Cython maintainer responded:

https://stackoverflow.com/questions/76972950/cython-returned-memoryview-is-always-considered-uninitialized

Basically, even though it is checking the memory view for validity/initialization, it is not actually acquiring the GIL except in the failure case which should never happen for us. We could restructure the code to not create a copy of the memory views used but instead adjust the indexing appropriately. The code would be harder to read but would not have the extra memoryview checks. I doubt it would have any performance difference.

I have not run a comparison of the performance, but given the logical changes appearing in the Cython I'm not too worried. I'll have to track down my old benchmarking from the last time the Cython was touched.

• Closes Upgrade to Cython 3.0 and check annotations #56
• Tests added
• Tests passed
• Passes git diff origin/main **/*py | flake8 --diff
• Fully documented

[codecov]

Codecov Report

Merging #57 (f763c7d) into main (285b6b4) will increase coverage by 0.26%.
Report is 2 commits behind head on main.
The diff coverage is 87.50%.

@@            Coverage Diff             @@
##             main      #57      +/-   ##
==========================================
+ Coverage   87.81%   88.07%   +0.26%     
==========================================
  Files          21       21              
  Lines        1436     1434       -2     
==========================================
+ Hits         1261     1263       +2     
+ Misses        175      171       -4     

Flag	Coverage Δ
unittests	88.07% <87.50%> (+0.26%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files Changed	Coverage Δ
geotiepoints/interpolator.py	84.17% <ø> (ø)
geotiepoints/modisinterpolator.py	100.00% <ø> (ø)
geotiepoints/tests/__init__.py	100.00% <ø> (ø)
geotiepoints/multilinear_cython.pyx	35.71% <75.00%> (+3.17%)	⬆️
geotiepoints/_modis_interpolator.pyx	99.23% <100.00%> (-0.02%)	⬇️
geotiepoints/_modis_utils.pyx	92.30% <100.00%> (+0.06%)	⬆️
geotiepoints/_simple_modis_interpolator.pyx	100.00% <100.00%> (ø)
geotiepoints/multilinear.py	90.62% <100.00%> (ø)

[coveralls]

coverage: 87.219% (+0.2%) from 86.971% when pulling ac8909c on djhoese:feature-cython-3 into 285b6b4 on pytroll:main.

[djhoese]

So this should be the results of the same MODIS operation I did in my #38 PR:

I note that the memory profile looks about the same but increases more at the end than previously...but that is from other parts of Satpy, not the interpolation which is only the blue/green blocks near the beginning. I don't like the dips in CPU usage at the beginning even though execution time is about the same.

[mraspaud]

LGTM, even if I haven't checked the new features of Cython 3. Do you want to wait for an answer about the cython 3 issue you found before we merge this?

[djhoese]

Yeah let's wait a bit. I'm more OK with trollimage being released than with this being merged given the issues.

[djhoese]

FYI the bug encountered in cython seems to have been fixed, but I don't think it is released yet.

[djhoese]

FYI I did a test compile with Cython's main branch and CPU usage seems to be much better. Not perfect, but I think it's as good as we're going to get. I'll try to skim through the C code and see if I can find any GIL acquires that I don't agree with.

[djhoese]

Nevermind, this is another major one: cython/cython#5688

[djhoese]

Nevermind on that one, but this one is still a thing: cython/cython#5685