Sort index in pandas

gistfile1.txt

In [297]: d = {0: {'foo': {'a': 12.68, 'b': 54.44, 'c': 83.98},
     ...:          'bar': {'a': 11.73, 'b': 53.34, 'c': 82.93}},
     ...:      2: {'foo': {'a': 11.12, 'b': 57.99, 'c': 81.05},
     ...:          'bar': {'a': 10.05, 'b': 56.12, 'c': 80.01}},
     ...:      1: {'foo': {'a': 13.41, 'b': 54.32, 'c': 82.74},
     ...:          'bar': {'a': 12.77, 'b': 53.15, 'c': 82.01}}}

In [298]: ds = (pd.DataFrame.from_dict(d, orient='index')
     ...:       .stack().apply(pd.Series)
     ...:       .rename_axis(['experiment', 'setup']))

In [299]: ds.sort_index(level = 0, inplace = True)

In [300]: ds
Out[300]:
                      a      b      c
experiment setup
0          foo    12.68  54.44  83.98
           bar    11.73  53.34  82.93
1          foo    13.41  54.32  82.74
           bar    12.77  53.15  82.01
2          foo    11.12  57.99  81.05
           bar    10.05  56.12  80.01

Now I start to see the weiredness ... it looks like 'foo' and 'bar' is sorted like 'foo' < 'bar'. Why should that be the case? I tried to set ascending = False, here it becomes even more obvious ...

Sorting 'foo' and 'bar' seems to be a special case. Try 'f' instead of 'foo' and 'b' instead of 'bar'.