Problem pickling length-1 boolean arrays (Trac #1802)

[numpy-gitbot]

Original ticket http://projects.scipy.org/numpy/ticket/1802 on 2011-04-18 by trac user eewallace, assigned to unknown.

If I pickle a dictionary containing two length one numpy arrays with dtype 'bool' and the same value (e.g. dict(a1=numpy.array([False]),a2=numpy.array([False]), when the data is unpickled, the resulting arrays contain references to the same boolean objects, so that changing the value of one of them changes both. The attached text file contains a transcript of a short IPython session illustrating the problem.
This seems to come up specifically with the standard pickle module, not with cPickle. It also occurs with arrays that are data members of separate class instances, which is where I first encountered it.
I've produced this on two Linux installations, one Arch Linux with python 2.7.1, and numpy 1.5.1, both built under gcc 4.5, the other with Red Hat 4.1, python 2.6.6, numpy 1.5.1 and gcc 4.1.

[numpy-gitbot]

Attachment added by trac user eewallace on 2011-04-18: numpy_pickle_bug.txt

[charris]

Still broken in 1.9-devel. The two unpickled arrays are different, but they have the same data pointer. This doesn't happen with integer arrays.

array "a1" data: <read-write buffer for 0x21fccb0, size 1, offset 0 at 0x2200a30>
array "a2" data: <read-write buffer for 0x21fcda0, size 1, offset 0 at 0x2200a30>

[mattip]

Fixed on 3.6, still happens on 2.7 Closing, if someone wants to submit a PR for python2 against the maintenance/1.16.x branch they should reopen. Here is the test

import pickle, numpy as np
>>> c = pickle.loads(pickle.dumps(dict(a1=np.array([False]),a2=np.array([False])))
>>> c['a1'][0] = True
>>> c #  On python 3.6.7
{'a1': array([ True]), 'a2': array([False])}