DEP,BUG: Coercion/cast of array to a subarray dtype will be fixed

[seberg]

This currently appends the subarray dtype dimensions first and then
tries to assign to the result array which uses incorrect broadcasting
(broadcasting against the subarray dimensions instead of repeating
each element according to the subarray dimensions).

This also fixes the python scalar pathway np.array(2, dtype="(2)f4,")
which previously only filled the first value. I consider that a clear
bug fix.

Closes gh-17511

[seberg]

To be clear, the first case here previously returned [3, uninitialized, uninitialized], while the second case actually did the right thing. So I considered that a simple bugfix here.

[mhvk]

@seberg - this looks quite good. With this change, will np.array([], '2f4') still raise a FutureWarning? I think it doesn't have to, since the result will not change in the future, but I didn't see a test to check that (and didn't quite follow the code close enough to see if this special case was caught).

[seberg]

Thanks for taking a look! I am scanning the full input for array(-like) objects in the code, it is a bit annoying, but feel this is fringe enough that it is OK. Since I look for array likes, it will not trigger for empty sequences.

There are cases where a warning will be given even though nothing will change, e.g. np.array(3).astype("(4)f4") has the same future behaviour as with the current broadcast behaviour.

[mhvk]

OK, that sounds good! And I can see that you cannot do everything... But perhaps good to add a test for the empty list case, just in case?

[seberg]

Oh, seems I accidentally deleted that sentence snippet: I had just added the test.

[mattip]

Thanks @seberg