BUG: warning on abs with nan on python 2.7, but not on 3.5

[Liam3851]

Both of the below are on numpy 1.11.2:

On python 2.7 on Windows, taking np.abs on an array with nan generates a RuntimeWarning:

In [1]: import numpy as np                                                                                             
                                                                                                                       
In [2]: np.abs(np.array([1, -2, np.nan]))                                                                              
C:\AnacondaBleeding\Scripts\ipython-script.py:1: RuntimeWarning: invalid value encountered in absolute                 
  if __name__ == '__main__':                                                                                           
Out[2]: array([  1.,   2.,  nan])

On python 3.5, doing the same command does not generate a warning.

In [1]: import numpy as np                                                                                             
                                                                                                                       
In [2]: np.abs(np.array([1, -2, np.nan]))                                                                              
Out[2]: array([  1.,   2.,  nan])                                                                                      

I don't have access to a Linux machine but I gather the warning is limited to python 2.7 running on Windows. Presumably the warning behavior should be consistent across platforms/python versions?

xref pandas-dev/pandas#14057

[charris]

I suspect it is compiler/library dependent. Python 3.5 on windows uses a much more recent compiler than Python 2.7.

[Liam3851]

Bug no longer exists in numpy version 1.14.2 on Windows.