np.array([], dtype=object) triggers error when copied with copy.deepcopy()

[aojeda]

The following code fails with numpy 1.12.0 (this is the only version we have tested).

>>> arr = np.array([], dtype=object)
>>> copy.deepcopy(arr)
Traceback (most recent call last):
  File "/usr/lib/python3/dist-packages/IPython/core/interactiveshell.py", line 2821, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-16-e314fe8a32a6>", line 1, in <module>
    copy.deepcopy(arr)
  File "/usr/lib/python3.4/copy.py", line 166, in deepcopy
    y = copier(memo)
ValueError: Iteration of zero-sized operands is not enabled

Note that this code already works if the dtype was not object but, e.g., int or str.

[seberg]

The error is the new iterator missing the zerosize ok flag, a bit annoying really since you usually want it.... It is gh-8125. CC @pelson in case you have time/want to fix it up.

[seberg]

Probably a regression in 1.12.

[charris]

@seberg Do you mean that we should implement a zerosize flag for nditer?

[seberg]

The flag already exists. Though since you have to test for an empty iterator afterwards then, could just as well test before the iterator here probably, since its a single array there is no complexity there.