[Numpy-discussion] Splitting MaskedArray into a separate package

[Marten van Kerkwijk]

Hi All,

I agree with comments above that deprecating/removing MaskedArray is
premature; we certainly depend on it in astropy (which is indeed what
got me started to contribute to numpy -- it was quite buggy!).

 I also think that, unlike Matrix, it is far from a neglected part of
numpy. Eric Wieser in particular has been cleaning it up quite a bit.
Also like Allan, I'm excited about making a new version based on
`__array_ufunc__`.

Beyond this, I think it is actually very useful for numpy to contain
at least one ndarray subclass, so that we have an internal check that
any changes we make to the base ndarray actually work.

So, my own sense would be that we should instead write a NEP with a
roadmap of what we want MaskedArray 2.0 to be like (e.g., no more
`nomask`...)

All the best,

Marten

p.s. And, of course, deprecation of Matrix is actually starting to
happen: with my PRs, it is now in a state where one could remove
`matrixlib` and all tests would still pass, and there is a pending PR
to start giving out PendingDeprectationWarning.