BUG/API Series/Index integer floordiv/mod inconsistency

[sinhrks]

Series looks to be coerced to float if the result contains NaN or inf. Otherwise Index doesn't. Make it to be consistent in either way.

Code Sample, a copy-pastable example if possible

pd.Series([1]) // pd.Series([0])
#0    inf
# dtype: float64

pd.Index([1]) // pd.Index([0])
# Int64Index([0], dtype='int64')

np.array([1]) // np.array([0])
# array([0])

pd.Series([1]) % pd.Series([0])
#0   NaN
# dtype: float64

pd.Index([1]) % pd.Index([0])
# Int64Index([0], dtype='int64')

np.array([1]) % np.array([0])
# array([0])

output of pd.show_versions()

on current master.

[jreback]

this was set this way a while ago - Index should be the same

numpy is wrong as it doesn't coerce types

[sinhrks]

OK. Found inconsistencies between int/float Series also. Both should be inf as the same as truediv?

np.__version__
# '1.11.1'

# OK
pd.Series([4]) // pd.Series([0])
# 0    inf
# dtype: float64

# NG
pd.Series([4.]) // pd.Series([0.])
# 0   NaN
# dtype: float64

# truediv, OK
# pd.Series([4.]) / pd.Series([0.])
# 0    inf
# dtype: float64

[jbrockmendel]

Added to the TODO list in #18824. FWIW all of the examples above except for pd.Series([4.]) // pd.Series([0.]) have been fixed.