DataFrame.stack() raises TypeError on mixed type level under Python 3

[toobaz]

Code Sample, a copy-pastable example if possible

In [2]: df = pd.DataFrame(1, index=range(3), columns=pd.MultiIndex.from_product([['a', 'b'], [1, 2]]))

In [3]: df['c'] = 2

In [4]: df.stack(0)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-4-a437419299b6> in <module>()
----> 1 df.stack(0)

/home/nobackup/repo/pandas/pandas/core/frame.py in stack(self, level, dropna)
   4500             return stack_multiple(self, level, dropna=dropna)
   4501         else:
-> 4502             return stack(self, level, dropna=dropna)
   4503 
   4504     def unstack(self, level=-1, fill_value=None):

/home/nobackup/repo/pandas/pandas/core/reshape/reshape.py in stack(frame, level, dropna)
    513 
    514     if isinstance(frame.columns, MultiIndex):
--> 515         return _stack_multi_columns(frame, level_num=level_num, dropna=dropna)
    516     elif isinstance(frame.index, MultiIndex):
    517         new_levels = list(frame.index.levels)

/home/nobackup/repo/pandas/pandas/core/reshape/reshape.py in _stack_multi_columns(frame, level_num, dropna)
    619         # level
    620         level_to_sort = _convert_level_number(0, this.columns)
--> 621         this = this.sort_index(level=level_to_sort, axis=1)
    622 
    623     # tuple list excluding level for grouping columns

/home/nobackup/repo/pandas/pandas/core/frame.py in sort_index(self, axis, level, ascending, inplace, kind, na_position, sort_remaining, by)
   3673             # make sure that the axis is lexsorted to start
   3674             # if not we need to reconstruct to get the correct indexer
-> 3675             labels = labels._sort_levels_monotonic()
   3676             indexer = lexsort_indexer(labels._get_labels_for_sorting(),
   3677                                       orders=ascending,

/home/nobackup/repo/pandas/pandas/core/indexes/multi.py in _sort_levels_monotonic(self)
   1275 
   1276             # indexer to reorder the levels
-> 1277             indexer = lev.argsort()
   1278             lev = lev.take(indexer)
   1279 

/home/nobackup/repo/pandas/pandas/core/indexes/base.py in argsort(self, *args, **kwargs)
   2146         if result is None:
   2147             result = np.array(self)
-> 2148         return result.argsort(*args, **kwargs)
   2149 
   2150     def __add__(self, other):

TypeError: unorderable types: str() > int()

Problem description

This is even more confusing because df[['a', 'b']].stack(0) (where no mixed dtype is left) results in the same error.

Expected Output

In [13]: df.stack(0)
Out[13]: 
       1    2    
0 a  1.0  1.0  NaN
  b  1.0  1.0  NaN
  c  NaN  NaN  2.0
1 a  1.0  1.0  NaN
  b  1.0  1.0  NaN
  c  NaN  NaN  2.0
2 a  1.0  1.0  NaN
  b  1.0  1.0  NaN
  c  NaN  NaN  2.0

Output of pd.show_versions()

[paste the output of pd.show_versions() here below this line]

[gfyoung]

Very odd indeed. Looking at the stacktrace, what is the value of result when the code breaks? Probably a good starting point to find the bug and patch.