COMPAT: comparisons with as_recarray breaking on older numpies

[jreback]

https://travis-ci.org/pydata/pandas/jobs/135897420

cc @gfyoung

[jreback]

xref #13373

[gfyoung]

Nitpick: can we change the title, as AFAICT that machine is using 1.7.1

[gfyoung]

Arghh...there's a bug in np.array_equal from that version (1.7.1) in which it cannot compare recarray, so I can't check the correctness of as_recarray (which returns a recarray) in that test using tm.assert_numpy_array_equal.

@jreback : How to proceed?

[jreback]

oh, you can put in a path if its a recarry, just maybe assert that both sides are recarray and listify (and assert_almost_equal)? (e.g. that will minimize impact on other code)

[gfyoung]

Hmm...well it isn't actually numpy.recarray that's the issue. Rather, it's numpy arrays like this:

>>> m
array([(1, 2.0), (3, 4.0)], 
      dtype=[('a', '<i8'), ('b', '<f8')])
>>> np.array_equal(m, m)  # should return True
NotImplemented

Might need to do try-except with specific checking of the error message.