BUG: df fails when columns arg is a list containing dupes

[ghost]

In [1]: DataFrame(data,columns=["a","a"])

...
pandas/pandas/core/internals.pyc in _stack_dict(dct, ref_items, dtype)
1344 stacked = np.empty(shape, dtype=dtype)
1345 for i, item in enumerate(items):
-> 1346 stacked[i] = _asarray_compat(dct[item])
1347
1348 # stacked = np.vstack([_asarray_compat(dct[k]) for k in items])

IndexError: index out of bounds

5e6db32 is a failing test for this.

it looks like _to_sdict threads down to a call to _convert_object_array which builds a dict
keyed on column names, so dupe columns get squashed and you end up with a mismatch
between the length of the columns arg to df.__init__ and the data.
_to_sdict is not used for ndarrays so this doesn't haoppen, I was able to reuse
_init_ndarray for the case of columns being a flat list and have things work as expected.

still, too much code touching this, better left to the core devs to decide how to handle this.

[wesm]

Fixing this is quite an undertaking since there's a lot of existing constructor code that assumes unique column names. I'm on it; probably get it sorted out over next day or so

[ghost]

Should this work?

pd.DataFrame.from_items([('a',['foo']),('a',['bar'])],columns=['a','a'])
Out[6]: 
     a    a
0  bar  bar

[ghost]

Also, forgive the nitpick, but since sdict is now abandoned, it would be good to rename the methods
that reference it, enhance readability...

def _list_to_sdict(data, columns, coerce_float=False):
def _list_of_series_to_sdict(data, columns, coerce_float=False):
def _list_of_dict_to_sdict(data, columns, coerce_float=False):

[wesm]

Yeah that should work.