Pass atol parameter to FITSDiff

[svank]

I noticed that the atol parameter doesn't get passed to astropy.io.fits.diff.FITSDiff when the reference file is FITS, and so setting atol has no effect. (e.g. @pytest.mark.array_compare(file_format='fits', atol=1) still used an atol of 0.) This PR adds a test for atol, ensures atol is passed through, and mentions the atol parameter in the readme alongside rtol (which was already mentioned).

As I was adding the test, I noticed that the existing test for the rtol parameter was buggy, in that the generated and reference files both contained only values of 1.6, so rtol could be zero or any other value and the test would still pass. (Maybe the reference files were regenerated at some point, forgetting that in this particular case, the reference file must be different from the generated file to test the tolerance values.) I've modified that test by scaling up the generated values so that the test fails if rtol isn't set to a large value, to ensure that the value of rtol is being used.

[pllim]

I don't grok the reference_dir setup. Is there no way to generate the truth in memory at run time? Seems weird to have to copy the same file over for a new test.

[svank]

Not sure if this answers your question, but since these tests are testing the machinery that compares generated data to a reference file, I think there has to be a reference file to compare to, even if only trivial data is being generated in this case. And since the reference file name is autogenerated from the test name, I think a new test needs a new copy of the file. I suppose the reference files could be programmatically generated in a temp dir as the tests run, but I'm not sure how to easily pass that temp dir to the pytest-arraydiff wrappers.

I could easily believe I'm missing a better way to do this (or misunderstanding the tests), in this weird context where the testing framework is being used to test the testing framework itself.

[pllim]

You are probably doing it right. I am not familiar with this package. 🤷

[pllim]

I had to re-enable CI in the settings. It was disabled due to inactivity. So I have to close/re-open to kick it off. Please do not be alarmed.

[astrofrog]

Looks good, thanks! Could you add an entry in the CHANGES.rst under 0.6?

[svank]

Done!

[pllim]

Thanks, all!