BUG: ptp fails on datetime types with out parameter

[eric-wieser]

>>> d = np.array([np.datetime64('now'), np.datetime64('now') + 10])
>>> d.ptp()
numpy.timedelta64(10,'s')
>>> d.ptp(out=np.empty((), np.timedelta64))
Traceback (most recent call last):
  File "<pyshell#63>", line 1, in <module>
    d.ptp(out=np.empty((), np.timedelta64))
TypeError: ufunc subtract cannot use operands with types dtype('<m8') and dtype('<M8[s]')

This is because it assumes that d.max() and d.ptp() have the same dtype, which they do not. Related to #10277.

Perhaps we need an optional_out argument, which is only used if the dtypes match?

[WarrenWeckesser]

FYI, I just updated the gufunc peaktopeak in ufunclab to handle this case:

In [46]: d = np.array([np.datetime64('now'), np.datetime64('now') + 10, np.datetime64('now') - 15])

In [47]: peaktopeak(d, out=np.empty((), np.timedelta64))
Out[47]: array(25, dtype=timedelta64)

In [48]: peaktopeak(d, out=np.empty((), dtype='timedelta64[s]'))
Out[48]: array(25, dtype='timedelta64[s]')

In [49]: peaktopeak(d, out=np.empty((), dtype='timedelta64[ms]'))
Out[49]: array(25000, dtype='timedelta64[ms]')