combine_first behavior change

[craustin]

from pandas import DataFrame
from datetime import datetime
df = DataFrame({'US': [1]}, index=[datetime(2012,1,1)])
df2 = DataFrame({}, columns=['EU'])
df.combine_first(df2)

In 0.9.0, this returns:

            US
2012-01-01   1

In 0.10.0b1, this returns:

            EU  US
2012-01-01 NaN   1

[ghost]

this was 9edd478 due to #2307,
it's mentioned in RELEASE.rst (a bit opaquely).

Note that as far back as 0.9.0 the docstring read:

Combine two DataFrame objects and default to non-null values in frame
calling the method. Result index will be the union of the two indexes

so this was a long-standing bug fixed.

[ghost]

the columns behaviour is not clearly documented, though.
@wesm, what's the right thing here?

[wesm]

Yes, it should be the union of the columns. I added a unit test-- sorry for the disruption, but I think doing otherwise is a bug