SkyCoord initialized with two CartesianRepresentations each with different units generates nonsense units

[thevorpalblade]

Here's a MWE:

from astropy.time import Time
from astropy import coordinates as coord

now = Time.now()
# create a tuple of CartesianRepresentations with different units
earth = coord.get_body_barycentric_posvel("earth", now)
print(earth)
(<CartesianRepresentation (x, y, z) in AU
    (0.86689474, -0.47190455, -0.2046662)>,
 <CartesianRepresentation (x, y, z) in AU / d
    (0.00857658, 0.01346705, 0.00583879)>)

bad_units = coord.SkyCoord(earth)
print(bad_units)
<SkyCoord (ICRS): (ra, dec, distance) in (deg, deg, AU / d)
    [(331.43780716, -11.7147513 , 1.00801222),
     ( 57.50870768,  20.08731918, 0.01700031)]>

Notably, the resulting SkyCoord splits the difference in terms of units, giving the angular components units of degrees and the radial component units of velocity. I'd suggest that initialization of the SkyCoord should fail rather than silently produce a inconsistent result if given inconsistent units.

I'm using astropy version 3.0.4 and numpy version 1.15.0 with python 3.7 under archlinux.

[pllim]

cc @eteq and @adrn

[adrn]

Ah, here this specific issue is because the second object returned by earth is a velocity. But it's strange that it's returned as a separate object instead of as a differential attached to the position...

[mhvk]

Yes, returning a tuple predates us having the ability of storing velocities as part of a representation. I don't know that we can trivially change that, but at least we should give the option of including it as a derivative (either via an argument or another function; e.g., it could use get_body, which already returns a SkyCoord of the position).

Separately, though, the example should really error since (I think) the result is interpreted as a stack of two coordinates, but those have inconsistent units. So, it is definitely a bug from that perspective.

[adrn]

Agreed!

[thevorpalblade]

Indeed, thinking that it should return a differential attached to a position is how I discovered this bug!
You can see the discussion and my subsequent enlightenment here: #7743