Update precision test of AltAz computation. - #11103

Hi @mhvk

Thanks for this update. I'm wondering if we are doing the right thing by tightening the precision so much. The two codes are both implementations of SOFA algorithms that include various treatments of precession etc. They agree very precisely which is wonderful, but should these tests tie us to these algorithmic choices?

Perhaps I haven't explained myself very well?

Well, my objective here has been to get Astropy up to the full accuracy of the bare SOFA model, which we have now accomplished. There are several improvements that you could make to this model that would invalidate this test, the first of which would be to include the effects of free core nutation provided by the dX and dY values given in the IERS tables. For a good introduction to these effects see Chapter 5 here:

https://www.iers.org/IERS/EN/Publications/TechnicalNotes/tn36.html

@mkbrewer - yes, good point. I started a new issue about possible further improvements: #11110.

@StuartLittlefair - I pushed the PR with the revised note.

Annoying that the remote-data run fails because of a numpy-dev problem...

But to be clear, the test passed locally and (I think...) also passed before here.

Looks like a problem reading the IERS table?

The test failed very quickly, with

E   RuntimeWarning: numpy.ndarray size changed, may indicate binary incompatibility. Expected 80 from C header, got 88 from PyObject

Should these tests be rerun now that #11116 has been merged?

Rebased to see if tests now pass. @StuartLittlefair - let me know if you have any remaining comments.

Thanks @mhvk!