Fix seviri_l2_grib end_time property bug.

[strandgren]

As described in #2942, the changes in #2717 lead to the end_time property not working for SEVIRI data, since we now use the sub-satellite point to determine the scanning mode and thus the temporal resolution used to determine the end_time from the start_time (prior to #2717 it was always set to 15 minutes, which is incorrect for RSS data). However, the self._ssp_lon attribute is only available after a dataset and its metadata have been loaded and thus an AttributeError was raised during Scene initialization. This was not covered by the unit test and a human error lead to the bug being unnoticed also when testing with real data.

This PR adapts the end_time property to return None if no dataset has been loaded, meaning that _ssp_lon is not available. As soon as a dataset has been loaded, both scn.end_time and scn[dataset].attrs['end_time'] will have valid and correct datetime.datetime objects. The unit test has been adapted to cover this bug.

• Closes 0.52.0 breaks seviri_l2_grib-reader with 'EUML2GribFileHandler' object has no attribute '_ssp_lon' #2942
• Tests adapted to capture this bug

[pnuu]

I have a feeling that either pytroll/trollimage#179 or pytroll/trollimage#181 cause the test failure, I created a new release today with these two included. Although there's no new conda-forge package which I would have thought the CI would be using 🤔

[pnuu]

Trollimage seems to be coming from PyPI: https://github.com/pytroll/satpy/blob/main/continuous_integration/environment.yaml#L64

@gerritholl could you perhaps have a look at the failing tests that seem to be handling paletted images and enhancement history?

• satpy/tests/enhancement_tests/test_enhancements.py::TestEnhancementStretch::test_palettize
• satpy/tests/enhancement_tests/test_viirs.py::TestVIIRSEnhancement::test_viirs

[pnuu]

I think it should be as simple as updating the expected results.

[pnuu]

Looks like the enhancement test function is used in many places other the two failing tests, so needs a bit more involved solution. I'll have a go.

[pnuu]

A fix attempted in #2944

[pnuu]

@strandgren the fix to the failing tests has been merged, so please merge main to this PR.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.09%. Comparing base (b9a2f4d) to head (9c1075c).
Report is 4 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2943   +/-   ##
=======================================
  Coverage   96.09%   96.09%           
=======================================
  Files         377      377           
  Lines       55070    55074    +4     
=======================================
+ Hits        52920    52924    +4     
  Misses       2150     2150           

Flag	Coverage Δ
behaviourtests	3.94% <0.00%> (-0.01%)	⬇️
unittests	96.19% <100.00%> (+<0.01%)	⬆️

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[coveralls]

Pull Request Test Coverage Report for Build 11438867601

Details

• 7 of 7 (100.0%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 96.202%

---

Totals
Change from base Build 11438666819:	0.0%
Covered Lines:	53168
Relevant Lines:	55267

---

💛 - Coveralls

[strandgren]

@pnuu Thanks, now the tests passed :)

[mraspaud]

I think I can live with this solution :)
However, I want to point out that it feels strange to have a property return different results depending on whether the file has already been open or not. Ideally, the call to end_time should trigger reading just what is needed in the file to give a sensible result.
But since I don't know the format, I think we can merge this.

[strandgren]

I think I can live with this solution :) However, I want to point out that it feels strange to have a property return different results depending on whether the file has already been open or not. Ideally, the call to end_time should trigger reading just what is needed in the file to give a sensible result. But since I don't know the format, I think we can merge this.

Yes I kind of agree, but the metadata are provided for each grib message i.e. dataset meaning that at Scene initialization it would be ambiguous from which message/dataset to read the ssp_lon from. Of course it will most likely be the same for all datasets and we could read it from the first message, but I still didn't like that solution, also since it invokes some duplicate reading.

[pnuu]

Could this test case be split into several specific tests?

[strandgren]

hm, I've been trying to refactor this into something better, but so far with no satisfying results. I have previously created an issue to work on removing the mocking #2932, do you think we could move this test optimization to the PR that will deal with the mocking (since it will anyway lead to quite some refactoring)?

[pnuu]

Yeah, sounds good! Feel free to merge.

[strandgren]

Thanks! Nah, you do it, you're the reviewer ;)

[pnuu]

LGTM