API: should reindex on a level introduce NaNs for missing entries per label of other levels?

[jorisvandenbossche]

Suppose the following dataframe and reindex operation:

In [65]: df = pd.DataFrame(np.arange(12).reshape(4,3), columns=pd.MultiIndex.from_tuples([('A', 'a'), ('A', 'b'), ('B', 'a')]))

In [66]: df
Out[66]:
   A       B
   a   b   a
0  0   1   2
1  3   4   5
2  6   7   8
3  9  10  11

In [67]: df.reindex(columns=['a', 'b'], level=1)
Out[67]:
   A       B
   a   b   a
0  0   1   2
1  3   4   5
2  6   7   8
3  9  10  11

Should this give the following?

In [67]: df.reindex(columns=['a', 'b'], level=1)
Out[67]:
   A       B
   a   b   a    b
0  0   1   2  NaN
1  3   4   5  NaN
2  6   7   8  NaN
3  9  10  11  NaN

I am not sure what the exact behaviour of the level keyword should be, but eg in the following example it does the selection of columns for each of label of the other level:

In [69]: df2 = pd.DataFrame(np.arange(18).reshape(3,6), columns=pd.MultiIndex.from_product([('A', 'B'), ('a', 'b', 'c')]))
In [70]:

In [70]: df2
Out[70]:
    A           B
    a   b   c   a   b   c
0   0   1   2   3   4   5
1   6   7   8   9  10  11
2  12  13  14  15  16  17

In [71]: df2.reindex(columns=['a', 'c'], level=1)
Out[71]:
    A       B
    a   c   a   c
0   0   2   3   5
1   6   8   9  11
2  12  14  15  17

[jreback]

xref #7895 (a lot of dupe as well), maybe close the other?
dupe of #11226, but I like your example better
xref #10549

[jreback]

If this is a leaf level then its easy to do this, but what if its not, then do you do a cartesian product of everything below that level? we could, but would want a broadcast argument I think (though this just maybe making it even more complicated)

[jreback]

xref #7895