BUG: Index.str.partition gives ValueError while trying to compute index name (expand=False)

[pwaller]

Index.str.partition's behaviour changed in 3ab35b4 in a backwards incompatible way which doesn't seem intentional.

Code Sample, a copy-pastable example if possible

python3 -c 'import pandas; pandas.Index(["a,b", "c,d"], name="hello").str.partition(",")'

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/pwaller/.local/src/pandas/pandas/core/strings.py", line 1432, in partition
    return self._wrap_result(result, expand=expand)
  File "/home/pwaller/.local/src/pandas/pandas/core/strings.py", line 1348, in _wrap_result
    return MultiIndex.from_tuples(result, names=name)
  File "/home/pwaller/.local/src/pandas/pandas/indexes/multi.py", line 889, in from_tuples
    return MultiIndex.from_arrays(arrays, sortorder=sortorder, names=names)
  File "/home/pwaller/.local/src/pandas/pandas/indexes/multi.py", line 844, in from_arrays
    names=names, verify_integrity=False)
  File "/home/pwaller/.local/src/pandas/pandas/indexes/multi.py", line 92, in __new__
    result._set_names(names)
  File "/home/pwaller/.local/src/pandas/pandas/indexes/multi.py", line 446, in _set_names
    raise ValueError('Length of names must match number of levels in '
ValueError: Length of names must match number of levels in MultiIndex.

Expected Output

In [1]: import pandas; pandas.Index(["a,b", "c,d"], name="hello").str.partition(",")
Out[1]: 
MultiIndex(levels=[['a', 'c'], [','], ['b', 'd']],
           labels=[[0, 1], [0, 0], [0, 1]])

Versions

I bisected the problem to 3ab35b4 (cc @sinhrks). The problem is not present in 0.17.1 and is present in 0.18.0rc2.

Note: when I use a name="foo", it happens to work because it has the correct length, the index name ends up being ["f", "o", "o"], so it's possible this is tested somewhere and happens to work even though it's not correct for a large number of use cases.

[sinhrks]

Thanks, and sorry for the inconvenience. I'll check soon.

[jreback]

Tthis is not a regression. The name argument you were using in 0.17.1 didn't do anything (a MultiIndex does not have a name, only names for the levels),

So the result (w/o specifying name) is correct.

The expand=False case has a bug in it.

In [50]: pandas.Index(["a,b", "c,d"], expand=True).str.partition(",")
Out[50]: 
MultiIndex(levels=[[u'a', u'c'], [u','], [u'b', u'd']],
           labels=[[0, 1], [0, 0], [0, 1]])

In [51]: pandas.Index(["a,b", "c,d"], expand=True).str.partition(",").names
Out[51]: FrozenList([None, None, None])

In [52]: pd.__version__
Out[52]: '0.18.0+7.g17a5982'