Behavior of negative exponentiation inconsistent between integer types

[jakevdp]

>>> np.int64(2) ** -2
0.25
>>> np.int32(2) ** -2
0

Tested on OSX 10.8.5 w/ numpy 1.10 on both Python 2.7 and 3.5.

[navneethc]

I second this.

Following tested using:
Python 3.5.1 (Python 3 environment created using Anaconda 4 64-bit)
Numpy 1.11.0

In [34]: np.int32(2)**-2
Out[34]: 0.25

In [35]: np.uint32(2)**-2
Out[35]: 0

In [36]: np.int64(2)**-2
Out[36]: 0.25

In [37]: np.uint64(2)**-2
Out[37]: 0.25

In [38]: np.int16(2)**-2
Out[38]: 0

In [39]: np.uint16(2)**-2
Out[39]: 0

In [40]: np.int8(2)**-2
Out[40]: 0

In [41]: np.uint8(2)**-2
Out[41]: 0

[ewmoore]

See issue #7449 and PR #7459. The PR fixes this, but could use some review.

[mhvk]

Just in case (have to run): Note that the array case should also keep working. In numpy 1.11:

np.ones((2,), dtype=int) ** -2
# array([1, 1])

In current master:

np.ones((2,), dtype=int) ** -2
# array([0, 0])

[ewmoore]

Gah. That's mine.

[ewmoore]

See #7651.