Issue with repr() for differential components on frames after set_representation_cls

[adrn]

Found a bug in BaseCoordinateFrame._data_repr(). I haven't tracked down exactly what the issue is, but here's how to reproduce:

>>> import astropy.coordinates as coord
>>> import astropy.units as u
>>> c = coord.ICRS(ra=150*u.deg, dec=-11*u.deg, 
...                pm_ra_cosdec=100*u.mas/u.yr,
...                pm_dec=199*u.mas/u.yr)
>>> c
<ICRS Coordinate: (ra, dec) in deg
    ( 150., -11.)
 (pm_ra_cosdec, pm_dec) in mas / yr
    ( 100.,  199.)>
>>> c.set_representation_cls(s=coord.CartesianDifferential)
>>> c
<ICRS Coordinate: (ra, dec) in deg
    ( 150., -11.)
 (d_lon_coslat, d_lat) in mas / yr
    ( 100.,  199.)>

After setting the differential class, the repr for the differential components reverts to the UnitSphericalCosLatDifferential component names. This has something to do with the

if isinstance(dif_data, (r.UnitSphericalDifferential,
                                         r.UnitSphericalCosLatDifferential,
                                         r.RadialDifferential)):

in _data_repr.

cc @mhvk

[mhvk]

@adrn - git blames you for putting that code in... So, if I change the if statement to be closer to what happens above for the representation:

                if (issubclass(dif_cls, (r.SphericalDifferential,
                                         r.SphericalCosLatDifferential)) and
                    isinstance(dif_data, (r.UnitSphericalDifferential,
                                          r.UnitSphericalCosLatDifferential,
                                          r.RadialDifferential))):

then I get

<ICRS Coordinate: (ra, dec) in deg
    ( 150., -11.)
 (v_x, v_y, v_z) in mas / (rad yr)
    (-82.88384202, -67.61704534,  195.34380951)>

The units are of course awkward since there is no distance, but arguably reasonable.

[mhvk]

Note: I do wonder whether we should have these if-statements at all - I think this is really to guard against changes from UnitSpherical to Spherical in transformations, but perhaps we should treat such occurrences as bugs rather than paper over them. After all, from the dimension of the distance it is trivial to infer whether something should be unit-spherical or not.