DEP: FutureWarning for void.item(): Will return bytes - #10044

I don't like the idea of an unconditional warning, especially when numpy has a habit of calling .item internally.

While it would be ideal to make it configurable, it also seems tricky, and the cost/benefit ratio of coding it up seems high.

Some points in favor of an unconditional warning:

1. Even internal uses of void.item() are probably bugs which we would want to hear about.
2. None of the unit test code ever calls void.item(), even internally.
3. Any casting or other code that tried to do arr.setitem(void.getitem()) already fails since buffers can't be converted to voids.

Running int(np.void(b'10')) results in void.item being called, but maybe that's a bug too.

I'm not sure it's a bug, but I'd also add that it raises an error like in my case 3. (In fact, it gives different errors in Python3 and Python2).

It might not be a bug because once we switch to returning bytes it will "work" (eg, int(np.void(b'10').tobytes() gives 10).

Let's go with the FutureWarning, that will at least get us some feedback.

Yeah I think I should remove the filter_waring_emitted as you said.

As to the question of how to turn off the warning if it is not needed, I think the user should just use warnings.filterwarnings - the warnings module already provides a way of silencing it.

Merged, thanks Allan. We'll see how it goes ...