.ptp() can return negative numbers

This is worked around in https://github.com/numpy/numpy/pull/14381 and mentioned in https://github.com/numpy/numpy/issues/3292#issuecomment-543626173 but still seems to behave like this. I don’t see why anyone would ever want it to return a negative number, so I think it should be considered a bug.

Reproducing code example:

import numpy as np
np.array([32767, -1], np.int16).ptp()

It returns np.int16(-32768), but it should logically return np.uint16(32768) or int(32768).

Numpy/Python version information:

1.17.4 3.7.6 (default, Jan 8 2020, 20:23:39) [MSC v.1916 64 bit (AMD64)]

[WarrenWeckesser]

At a minimum, I guess we should add a warning to the docstring

See https://github.com/numpy/numpy/pull/16240

[rgommers]

If there is more interest now in getting peaktopeak–or something like it–into numpy, I can start working on that.

I’d rather not see that, too trivial/niche for numpy at this point.