Change representation arithmetic tests to be n-d - #5441

OK, now I see the problem: the broadcasting really is going wrong because the tests assume the input is 1-dimensional. E.g., one has to replace [::-1] with [:, ::-1] at https://github.com/eteq/astropy/blob/deff32f13f803b30ad4ba1f6848cdd16506d9a6d/astropy/coordinates/tests/test_representation_arithmetic.py#L281 and add an additional np.newaxis in https://github.com/eteq/astropy/blob/deff32f13f803b30ad4ba1f6848cdd16506d9a6d/astropy/coordinates/tests/test_representation_arithmetic.py#L326.

I haven't checked everything (too late here in the UK), but my hope now is that this is a red herring.

Would it make sense to parametrize the tests to make sure they run in say 1, 2, and 3D?

@astrofrog - in principle, one could add higher-D, but even with my 1-D example there are already 2-D tests, and I don't do anything fancy, so my feeling is that this was really a red herring due to @eteq not having looked quite carefully enough where in the tests assumptions were made about dimensionality, and that thus we'd just be checking numpy broadcasting...

@mhvk - I think we do want to have actual tests that are n-D, because even if the current functionality is fine, future additions may not be. I know that's where almost all the problems that have sneaked into releases have come from (i.e., it works in 1d but not n-d).

I'll try re-working this branch to parameterize the tests to work in different dimensions.

@eteq - it be better off checking that things work at a higher level, e.g., using the example mentioned in #5364. (Indeed, this was something I had meant to do for this whole series -- first at least make it possible to do multi-d properly, and then shake out any remaining bugs by adding multi-D examples to the transformation chain.) But obviously it is not like it hurts to do it here too! It just will need a bit of thought (I would actually recommend just adding a few explicit multi-d tests rather than parametrize, but it is obviously up to you).

I'd like to revisit the test organization and coverage in astropy coordinates, and with that I think it does make sense to have fixtures that test scalar, 1d, and nd arrays in some contexts. So, I'd suggest closing this PR and deferring to #6577

Based on the seeming agreement on @adrn's suggestion above, I'm closing this. Please reopen if I misunderstood.