np.sort of memmap is still memmap

[ZisIsNotZis]

I know this is not a big problem. Just mentioning here.

sort of a memmap (which uses different memory and is not a mmap anymore) is of type memmap.

I "kind of" remember one time that I got a readonly memmap with some non-inplace operaton, but this might not be true since I can't remember it clearly.

Reproducing code example:

import numpy as np
a = np.memmap('xxx.B', mode='r')
b = np.sort(a)
np.shares_memory(a, b), type(b), b.flags.writeable

Output:

(False, numpy.memmap, True)

Numpy/Python version information:

1.16.4 3.6.8 |Anaconda, Inc.| (default, Dec 30 2018, 01:22:34)
[GCC 7.3.0]

[seberg]

I remember having a year old patch that tries to change this a bit. Maybe worth a shot. The idea is really just to not do anything in __array_finalize__ and __array_wrap__. Now that 1.17 is branched off, might be a good time (since I suppose it might affect downstream, even if it should not).