__repr__ method of BaseRepresentation can fail

[StuartLittlefair]

The __repr__ method of the BaseRepresentation class uses a numpy view to access the values. It is possible to create objects for which this view cannot be taken, i.e (using v1.3.1):

from astropy.coordinates import EarthLocation, SkyCoord, get_moon
import astropy.units as u
from astropy.time import Time
import numpy as np

greenwich = EarthLocation.of_site('greenwich')
times = Time("2017-03-20T12:00:00") + np.linspace(-2,2,3)*u.hour
moon = get_moon(times, location=greenwich)
targets = SkyCoord([350.7*u.deg, 260.7*u.deg], [18.4*u.deg, 22.4*u.deg])
targs2d = targets[:, np.newaxis]
targs2d.transform_to(moon)

produces

Traceback (most recent call last):
  File "test_patch2.py", line 11, in <module>
    targs2d.transform_to(moon)
  File "/Users/sl/anaconda3/lib/python3.6/site-packages/astropy/coordinates/sky_coordinate.py", line 454, in transform_to
    return self.__class__(new_coord, **frame_kwargs)
  File "/Users/sl/anaconda3/lib/python3.6/site-packages/astropy/coordinates/sky_coordinate.py", line 212, in __init__
    kwargs = self._parse_inputs(args, kwargs)
  File "/Users/sl/anaconda3/lib/python3.6/site-packages/astropy/coordinates/sky_coordinate.py", line 360, in _parse_inputs
    .format(attr, coord_value, valid_kwargs[attr]))
  File "/Users/sl/anaconda3/lib/python3.6/site-packages/astropy/coordinates/representation.py", line 400, in __repr__
    arrstr = _array2string(self._values, prefix=prefixstr)
  File "/Users/sl/anaconda3/lib/python3.6/site-packages/astropy/coordinates/representation.py", line 229, in _values
    axis=-1).view(dtype).squeeze()
ValueError: new type not compatible with array.

[StuartLittlefair]

Note that I have a fix in place for this by replacing this line with

# for complex shapes and structures, this may require a copy
try:
     return np.concatenate([coo.reshape(sh).value for coo in coordinates],
                           axis=-1).view(dtype).squeeze()
except:
    return np.concatenate([coo.reshape(sh).value for coo in coordinates],
                            axis=-1).astype(dtype).squeeze()

however, I'm struggling to add a regression test, since the example above runs into the exception reported in #5890 , which is thornier issue.

I cannot think of another way to create an N-D representation which cannot be accessed using a view like the one above.

[mhvk]

Hmm, not good, that! Here is a simpler test:

CartesianRepresentation(np.arange(27).reshape(3,3,3), unit='m').T
# ... TypeError

[mhvk]

My own sense would be to rewrite the code to something more robust, along the lines of

result = np.empty(self.shape, dtype)
for c, coo in zip(self.components, coordinates):
    result[str(c)] = coo.value
return result

I'll make a quick PR...

[mhvk]

See #5896 for a fix.