Fix copy of mask when copying a Table - #7486

This doesn't quite go right if data.mask is np.ma.nomask - it makes an unnecessary mask array. Perhaps easiest to solve by rewriting the whole stanza a bit:

if mask is None:
    mask = getattr(data, 'mask', np.ma.nomask)

if mask is not np.ma.nomask and copy:
    mask = np.array(mask)  # copy is default; adding.

(Note: np.ma.nomask is better than False - it is what MaskedArray uses when no mask is passed in.)

So, just to sure, this would also replace the deepcopy case, and the mask = False added by @taldcroft recently (#7422) ? Your version makes sense but this is not the same behavior, as it results in using nomask where currently a full mask is created.
Maybe there is a good reason why a full mask is created ? nomask is often problematic as you need to mask values with the "right way" to be sure that the mask is created when needed.

82 failed tests (for astropy.table) with nomask, because something like column.mask[column.data == col.null] = True no more works.

Ah, well, that proofs that nomask is indeed a bad idea (though it must be possible to trigger this in current master by passing in a masked array with nomask...). So, then,

if mask is None:
    mask = getattr(data, 'mask', False)

if mask is np.ma.nomask:
    mask = False
elif mask is not False and copy:
    mask = np.array(mask)

I was trying to find a similar solution but when mask is not None, it is always copied (the deepcopy line), and there is a test for this. So I end up with something like this:

        if mask is None:
            # Issue #7399 with fix #7422.  Passing mask=None to ma.MaskedArray
            # is extremely slow (~3 seconds for 1e7 elements), while mask=False
            # gets quickly broadcast to the expected bool array of False.
            mask = getattr(data, 'mask', False)

            if mask is not False:
                mask = np.array(data.mask, copy=copy)
        elif mask is np.ma.nomask:
            # Force the creation of a full mask array as nomask is tricky to
            # use and will fail in an unexpected manner when setting a value
            # to the mask.
            mask = False
        else:
            mask = deepcopy(mask)

Looks OK to me - that even gets rid of data.mask = nomask. (But it should be np.array(mask), not np.array(data.mask).

Good catch, thanks.

This test is correct (and reflects the original issue) but somehow is more obfuscated than need be.

c = MaskedColumn([1, 2], mask=[False, True])
c2 = MaskedColumn(c, copy=True)
c2.mask[0] = True
assert np.all(c.mask == [False, True])
assert np.all(c2.mask == [True, True])

Indeed, I simplified with your suggestions, thanks.

Add something like

Also fixed a problem when MaskedColumn was initialized with ``mask=np.ma.nomask``.