squeeze on memmap returns numpy.ndarray instead of memmap

When performing squeeze operation on a memmap, it is converted to a normal in-memory array. Other array operation functions, such as reshape or expand_dims preserve the memmap type. Therefore the conversion upon squeeze is somewhat unexpected behavior in my point of view.

import numpy as np
from tempfile import mkdtemp
import os.path as path
filename = path.join(mkdtemp(), 'newfile.dat')

fp = np.memmap(filename, dtype='float32', mode='w+', shape=(3,4,1))

print('type before squeeze')
print(type(fp))
print(fp.shape)

print('type after squeeze')
squeezed = fp.squeeze()
print(type(squeezed))
print(squeezed.shape)

my output with numpy version 1.14.3

type before squeeze
<class 'numpy.core.memmap.memmap'>
(3, 4, 1)
type after squeeze
<class 'numpy.ndarray'>
(3, 4)

Start by running the provided memmap and squeeze reproduction with the current NumPy test suite, then compare its behavior with reshape and expand_dims. Trace the squeeze entry point and existing memmap coverage; done means the intended behavior is established by a regression test and is consistent with the project’s expectations.