Wired behavior: 0 / dataframe = inf while 0.0 / dataframe = 0.0

[skinnymonkey]

In [3]: s = pd.Series([1.0, 2.0])

In [4]: 0 / s
Out[4]:
0    inf
1    inf
dtype: float64

In [5]: 0.0 / s
Out[5]:
0    0
1    0
dtype: float64

This is seems quite wired. Should not 0/s be 0 as well?

[jreback]

that is a bug. Their is a whole set of logic to properly handle division by 0 property (as numpy doesn't handle integer division correctly). And fill appropriately with +-inf/NaN as appropriate.

These both should yield the same as [5]

This op is handled as __rdiv__ IOW, a reversed division op. And I think the logic is a bit off for this case.

Welcome to have you dive in if you'd like.

[Garrett-R]

Good catch @skinnymonkey! I think there's actually two problems related problems. The other was that:

Series([0, 1]) / 0

turned out to evaluate to Series([np.inf, np.inf]) instead of Series([np.nan, np.inf]).