Dropna=False is ignored for MultiIndex stack.

[miketkelly]

>>> df = pd.DataFrame({'A': ['a1', 'a2'], 'B': ['b1', 'b2'], 'C': [1, 1]})
>>> df
    A   B  C
0  a1  b1  1
1  a2  b2  1

# Unstack introduces some NaNs
>>> df.set_index(['A', 'B']).unstack()
     C    
B   b1  b2
A         
a1   1 NaN
a2 NaN   1

# Stack with dropna=False should keep the NaNs, but doesn't.
>>> df.set_index(['A', 'B']).unstack().stack(dropna=False)
       C
A  B    
a1 b1  1
a2 b2  1

# This is what I expected to get
>>> df.set_index(['A', 'B']).unstack().stack(dropna=False)
        C
A  B     
a1 b1   1
   b2 NaN
a2 b1 NaN
   b2   1

The problem seems to be with reshape.py, line 495 with forces dropna=True.

if isinstance(frame.columns, MultiIndex):
    return _stack_multi_columns(frame, level=level, dropna=True)