WCS: exception if MJD-OBS missing when some other headers present

Description

If a FITS header has a spectral axis and OBSGEO-* keywords but no MJD-DATE, then _get_components_and_classes crashes because it tries to construct a Time from a NaN MJD. I’m not aware of any requirement in the FITS specification to provide this header, but I didn’t comb the entire spec.

Expected behavior

I’m not familiar with the details of SpectralCoord at all so I don’t know what the appropriate fix is, but I would expect the resulting WCS to be basically usable.

Actual behavior

When running the test code below, received an exception (see first comment - I forgot to paste it when filing the bug).

Steps to Reproduce

#!/usr/bin/env python3

import astropy.wcs
import astropy.io.fits

header = astropy.io.fits.Header()
header['OBSGEO-X'] = 5109360.133321234
header['OBSGEO-Y'] = 2006852.586042906
header['OBSGEO-Z'] = -3238948.127478876
header['DATE'] = '2020-05-14T08:03:43.930'
header['SPECSYS'] = 'TOPOCENT'
header['TIMESYS'] = 'UTC'
header['EQUINOX'] = 2000.0
header['RADESYS'] = 'FK5'
header['NAXIS'] = 3
header['NAXIS1'] = 100
header['NAXIS2'] = 100
header['NAXIS3'] = 1
header['BUNIT'] = 'Jy/beam'
header['CDELT1'] = 1.0
header['CDELT2'] = 1.0
header['CDELT4'] = 1.0
header['CUNIT1'] = 'deg'
header['CUNIT2'] = 'deg'
header['CUNIT3'] = 'Hz'
header['CTYPE1'] = 'RA---SIN'
header['CTYPE2'] = 'DEC--SIN'
header['CTYPE3'] = 'FREQ'
header['CRVAL1'] = 0.0
header['CRVAL2'] = 0.0
header['CRVAL3'] = 1e9
header['CRPIX1'] = 50.0
header['CRPIX2'] = 50.0
header['CRPIX3'] = 1.0
wcs = astropy.wcs.WCS(header)
wcs.world_axis_object_classes

System Details

Linux-5.4.0-60-generic-x86_64-with-glibc2.29 Python 3.8.5 (default, Jul 28 2020, 12:59:40) [GCC 9.3.0] Numpy 1.19.1 astropy 4.3.dev450+gaa71c2f45 Scipy 1.5.2 Matplotlib 3.3.0

[bmerry]

I think one could just proceed with observer = None as already in the case for missing wcs.wcs.obsgeo.

That is one solution I had in mind. An alternative might be to create the observer without an obstime, but I don’t know if that’s just going to cause an exception down the road when you try to change frames (possibly it makes sense for some values of SPECSYS?)

However DATE-AVG/MJD-AVG, and with lower priority -BEG/-END, would certainly acceptable replacements

In fact the spec says “In order to compute the velocities it is necessary to have the date and time of the observation; the keywords DATE-AVG and MJD-AVG are reserved for this purpose.” That sounds like they should be considered before DATE-OBS/MJD-OBS.

[bmerry]

Whoops, I pasted the sample code twice instead of pasting the traceback. I get the same exception though. I’ll just edit the report to remove the duplicated sample code.