BUG/API: Series arithmetic ops inconsistently hold names

[sinhrks]

Based on Index behavior, Series arithmetic should reset name when names are different?

import pandas as pd
s1 = pd.Series([1, 2, 3], name='a')
s2 = pd.Series([3, 4, 5], name='b')

(s1 + s2).name
# None

s1.add(s2).name
# a

[jreback]

This looks like just need to pass name to the constructor https://github.com/pydata/pandas/blob/master/pandas/core/series.py#L1510

[vikramls]

The fix is in ops.py - https://github.com/pydata/pandas/blob/master/pandas/core/ops.py#L509 - I added a check for if name is None and used the name from left if this is True. This however breaks an explicit test - https://github.com/pydata/pandas/blob/master/pandas/tests/test_series.py#L266. If the names don't match, neither is selected. Should I update this test as part of the fix as well?

[shoyer]

@vikramls I don't think we want to change that behavior (if names are different, the resulting name should be None). So I guess this needs a different fix.

[vikramls]

@shoyer - The test-case checks the same scenario that this bug is addressed to. Does this mean that the bug is not valid?

from test_series.py:

    result = self.ts + cp # self.ts and cp are both Series objects
    self.assertIsNone(result.name)

Series.add is executed via the _binop method wheres Series + Series is executed via the added __add__ method.

[shoyer]

@vikramls To clarify, I think this behavior is correct:

(s1 + s2).name
# None

and this behavior is the bug:

s1.add(s2).name
# a

[vikramls]

@shoyer, thanks for the clarification. In this case the bug is triggered by the call to __finalize__ which copies attributes from other to the newly created object. One fix is to remove the call to __finalize__ in _binop but it doesn't seem right to special-case _binop in this manner.