Pandas Sum of all NaNs comes out zero if bottleneck is installed

x = pandas.DataFrame(np.nan, index=['a', 'b'], columns=['c', 'd']) print x.sum()

I expected to see all NaN's in the resulting Series. However, if i have bottleneck installed i get a Series of all zeros instead of all NaNs.

This is with pandas 0.18.1.

Thanks for any comments or suggestions

Comment From: jreback

dupe of #9422 see long discussion there.