Index.__add__ OK but __sub__ not?

[jbrockmendel]

Curious about why Index.__add__ tries does the natural thing but Index.__sub__ raises TypeError.

    def __add__(self, other):
        return Index(np.array(self) + other)

    def __radd__(self, other):
        return Index(other + np.array(self))

    __iadd__ = __add__

    def __sub__(self, other):
        raise TypeError("cannot perform __sub__ with this index type: "
                        "{typ}".format(typ=type(self)))

[jreback]

this is why comments are important!

add is concat

eg
Index(list(‘abc’)) + ‘d’

sub doesn’t make sense

[jbrockmendel]

That seems pretty string-specific. Shouldn't the object-dtype try to broadcast the scalar operation?

pd.Index([1, 2, 3], dtype='o') + 1
pd.Index([1, 2, 3], dtype='o') - 1

The motivation here is in https://github.com/pandas-dev/pandas/blob/master/pandas/core/indexes/datetimes.py#L933

where _add_offset_array can use self.astype('O') + np.array(other) but _sub_offset_array has to use self.__class__(self.astype('O').values - np.array(other))