Assigning from other NDFrame broken on multiple MultiIndex columns

[toobaz]

In [2]: df = pd.DataFrame(columns=pd.MultiIndex.from_product([range(2), range(2)]))

In [3]: df.loc[0] = range(4)

In [4]: df.loc[1] = pd.Series(range(4))

In [5]: df
Out[5]: 
     0         1     
     0    1    0    1
0  0.0  1.0  2.0  3.0
1  NaN  NaN  NaN  NaN

That is: the first loc is doing what we expect, but the second is only creating a row of NaNs. By the way, giving an explicit columns list does not help, and

In [6]: df.loc[2] = pd.Series(range(40))

does not even raise an error.

[jreback]

this is correct. you are reindexing in both cases and you don't have any matches.
hHe reason [3] works is that it does have any indices.

In [17]: s = pd.Series(range(4))

In [18]: s
Out[18]: 
0    0
1    1
2    2
3    3
dtype: int64

In [19]: s.reindex(df.columns)
Out[19]: 
0  0   NaN
   1   NaN
1  0   NaN
   1   NaN
dtype: float64

[toobaz]

Ouch!

[toobaz]

While my example was remarkably stupid, it came from an oversimplification of

In [2]: df = pd.DataFrame(columns=pd.MultiIndex.from_product([range(2), range(2)]))

In [3]: df.loc[2,0] = pd.Series(range(2), index=range(2))

In [4]: df
Out[4]: 
     0         1     
     0    1    0    1
2  NaN  NaN  NaN  NaN

... which I assumed should work, since

In [5]: df = pd.DataFrame(columns=range(2))

In [6]: df.loc[2] = pd.Series(range(2), index=range(2))

In [7]: df
Out[7]: 
     0    1
2  0.0  1.0

does. What do you think?

[jreback]

In [13]: df = pd.DataFrame(columns=pd.MultiIndex.from_product([range(2), range(2)]))

In [14]: df.loc[(2,0)] = pd.Series(range(2), index=range(2)).values

In [15]: df
Out[15]: 
   0       1     
   0  1    0    1
2  0  1  NaN  NaN

In [29]: df.loc[2] = pd.Series(range(2), index=MultiIndex.from_tuples([(0,0),(0,1)]))

In [30]: df
Out[30]: 
     0        1    
     0    1   0   1
2  0.0  1.0 NaN NaN

I suppose you could make an argument that

In [10]: idx = pd.IndexSlice

In [24]: df = pd.DataFrame(columns=pd.MultiIndex.from_product([range(2), range(2)]))

In [25]: df.loc[2,idx[0,:]] = pd.Series(range(2), index=range(2))

In [26]: df
Out[26]: 
     0         1     
     0    1    0    1
2  NaN  NaN  NaN  NaN

this should work. I don't think there is ANY support for this and would make an already very complicated interface even worse.

If you want to see what you can do great. But would need to make things simpler and not more complicated.

[jreback]

this is also a dupe of #6699