Repository navigation
Python long raises OverflowError while same value as np.uint64 does not. #5612
Description
Trying to assign a value of 2**64 - 1 to an array of dtype uint64 raises an OverflowError, but only if the value is a Python long rather than a Numpy scalar. This seems like a bug to me because the value obviously doesn't overflow and they compare equal.
Python 3.4.2 (default, Jan 12 2015, 11:38:40)
[GCC 4.9.2 20141224 (prerelease)] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import numpy as np
>>> np.version.version
'1.9.1'
>>> arr = np.empty((1,), dtype=np.uint64)
>>> arr.fill(np.uint64(2**64-1))
>>> arr.fill(2**64-1)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
OverflowError: int too big to convert
>>> np.uint64(2**64-1) == 2**64 - 1
True
>>> arr == 2**64-1
array([ True], dtype=bool)
Activity
jaimefrio commentedon Feb 27, 2015
It only happens when using
fill, not when doing a direct assignment:This points to
PyArray_FillWithScalaras the culprit, and I think it is indeed to blame: when converting integer types, it callsPyLong_AsLongLong, and doesn't try withPyLong_AsUnsignedLongLongif that fails, see here for the numpy code.PRs are welcome...
johntyree commentedon Feb 28, 2015
A real bug and not just pebkac! How exciting! I'll try to put together a PR this weekend.
Update to 1.9.2
Update to 1.9.2