Drop non-dimensional coordinates in Compositor

[sfinkens]

Attempt to fix #795 by dropping non-dimensional coordinates.

• Closes Composites delayed in the presence of non-dimensional coordinates #795
• Tests added and test suite added to parent suite
• Tests passed
• Passes flake8 satpy

[mraspaud]

What if we have lon and lat as 2d coords? They wont be dimensional, but still should be kept right ?

[BENR0]

I've come across this too a couple of days ago. Just calling resample after loading fixes this. I don't know how resample handles 2d coords but a similar approach of handling coords could maybe be used here.
Alternatively the proposed fix maybe could include a check for 2d coords (and any other non dim coords which should be kept and is the same over all datasets) demote those to variables, drop every other coord and promote them to coords again afterwards.

[sfinkens]

I've implemented @djhoese's suggestion to exclude "neglible" non-dimensional coordinates (defined in a global list). That also preserves 2d lon/lat.

[djhoese]

Wasn't the original problem for this PR the acq_time coordinate? You aren't excluding that here, are you? Also, what does PEP257 say about """ comments?

[sfinkens]

@djhoese As you suggested all non-dimensional coordinates with time in their name are excluded. PEP257 says, attribute docstrings are ok and references PEP258 for the details: https://www.python.org/dev/peps/pep-0258/#id40

[djhoese]

Ah I missed the in part of that operation. Looks good (including the docstring).

[djhoese]

Is this ready for review (still a draft)?

[sfinkens]

Added a test, now ready for review!

[codecov]

Codecov Report

Merging #796 into master will increase coverage by 0.66%.
The diff coverage is 95.23%.

@@            Coverage Diff             @@
##           master     #796      +/-   ##
==========================================
+ Coverage   83.06%   83.72%   +0.66%     
==========================================
  Files         163      165       +2     
  Lines       23592    24136     +544     
==========================================
+ Hits        19596    20209     +613     
+ Misses       3996     3927      -69

Impacted Files	Coverage Δ
satpy/composites/abi.py	100% <ø> (ø)	⬆️
satpy/tests/compositor_tests/__init__.py	99.52% <100%> (+0.11%)	⬆️
satpy/composites/cloud_products.py	89.06% <100%> (ø)	⬆️
satpy/composites/viirs.py	83.71% <100%> (ø)	⬆️
satpy/composites/ahi.py	100% <100%> (ø)	⬆️
satpy/composites/__init__.py	74.39% <91.66%> (+3.83%)	⬆️
satpy/readers/avhrr_l1b_gaclac.py	90% <0%> (-1.43%)	⬇️
satpy/tests/test_utils.py	98.79% <0%> (-1.21%)	⬇️
satpy/readers/electrol_hrit.py	91.33% <0%> (-0.67%)	⬇️
satpy/readers/seviri_l1b_native.py	61.06% <0%> (-0.28%)	⬇️
... and 144 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a5733f3...d663f2f. Read the comment docs.

[coveralls]

Coverage increased (+0.7%) to 83.73% when pulling d663f2f on sfinkens:fix-composites into a5733f3 on pytroll:master.

[mraspaud]

Looks good, I like the simple fix to the problem. Just one change request.

[mraspaud]

I'm happy now, LGTM :)

[djhoese]

I'm not sure match is the best name, but it is definitely not the worst. I'd really like if check_areas got its return values back (see comment below).