DataFrame.combine_first() breaks for DataFrame with indices but no columns

[pikeas]

Title says it all. core/frame.py, combine() method at line 3448:

    if other.empty:
        return self.copy()

If other is a DataFrame which has indices but no columns, a combine_first() should add the indices with every column set to NaN.

Sorry for not submitting a patch myself - I'm very new to the pandas codebase, so I'm not sure whether this should be resolved in combine() for all combination attempts, or only in combine_first().

...Maybe empty should check whether the DataFrame has indices? Though that could silently break a lot of old code...

@pikeas , can you provide a testcase? I tried the following:

import pandas as pd
a=pd.DataFrame([1,2])
a.combine_first(pd.DataFrame())

which looks like it's working here.

and:

import pandas as pd
a=pd.DataFrame([1,2])
b=pd.DataFrame(index=[1,2])
a.combine_first(b)

which also works

[pikeas]

Sure thing!

df = pd.DataFrame(index=[2,3,4])
df['A'] = 5
missing = pd.DataFrame(index=[0,1])
combined = df.combine_first(missing)

Actual output:

In [171]: combined
Out[171]: 
   A
2  5
3  5
4  5

Expected output:

In [171]: combined
Out[171]: 
   A
0  NaN
1  NaN
2  5
3  5
4  5