Creating table from list of rows fails with numpy-dev (OK for 1.17)

With astropy-3.2.1 and numpy-1.16.4 I can create one table from the rows of another, as follows:

>>> from astropy.table import Table
>>> t = Table([[1, 2, 3], [4, 5, 6]], names=('a', 'b'))
>>> print(t)
 a   b
--- ---
  1   4
  2   5
  3   6
>>> print(type(t)(rows=[row for row in t], names=t.dtype.names))
 a   b
--- ---
  1   4
  2   5
  3   6

However, with the same astropy, and numpy-1.17.0rc1 I cannot (well, I can, it just doesn’t do what I expect it to):

>>> from astropy.table import Table
>>> t = Table([[1, 2, 3], [4, 5, 6]], names=('a', 'b'))
>>> print(t)
 a   b
--- ---
  1   4
  2   5
  3   6
>>> print(type(t)(rows=[row for row in t], names=t.dtype.names))
a [2]  b [2]
------ ------
1 .. 4 1 .. 4
2 .. 5 2 .. 5
3 .. 6 3 .. 6

If I manually cast each row to list, then things do work again:

>>> print(type(t)(rows=[list(row) for row in t], names=t.dtype.names))
 a   b
--- ---
  1   4
  2   5
  3   6

Is this just a consequence of an upstream change in the new pre-release of numpy, or are tables not really supposed to be created this way in the first place?

[mhvk]

Just noticed that this will get fixed by #9165.

[mhvk]

Note that while the problem is caused upstream, I think it might still be good to fix recarray_fromrecords regardless, at it seems more than a little flaky.