This PR adds a reproducibility unit test that I was using to debug non-reproducibility at a certain computing center on Monday (i.e. it was failing), but subsequently something else changed and now I can't get the tests to fail anymore. Either way, it seems useful to add these tests in case the problem crops up again.
While I'm at it, I changed astropy.io.fits.write(clobber...) to overwrite due to a deprecation warning (deprecated since astropy 2 and we're moving on to astropy 4 by now...)
I'm also updating the Travis testing to astropy 3, with the intension of letting Travis test backwards compatibility with astropy 3, while letting nightly tests at NERSC test astropy 4 (I don't want to bog down travis testing with multiple combinations of astropy, but we could consider it).
This PR adds a reproducibility unit test that I was using to debug non-reproducibility at a certain computing center on Monday (i.e. it was failing), but subsequently something else changed and now I can't get the tests to fail anymore. Either way, it seems useful to add these tests in case the problem crops up again.
While I'm at it, I changed
astropy.io.fits.write(clobber...)
tooverwrite
due to a deprecation warning (deprecated since astropy 2 and we're moving on to astropy 4 by now...)I'm also updating the Travis testing to astropy 3, with the intension of letting Travis test backwards compatibility with astropy 3, while letting nightly tests at NERSC test astropy 4 (I don't want to bog down travis testing with multiple combinations of astropy, but we could consider it).