Tentative fix for pandas-dev#4297.

The problem finally seem to be an odd restriction: acceptable sequences
were list, tuple, or numpy.ndarray.

A guess is that strings in Python wanted to be considered scalars,
but the above restriction was implemented rather than take anything
with a __len__ except str (or unicode for Python 2.x) in.

Considerations about the use of the buffer interface are not needed
for now.

note: I cannot run tests on Python 2.7.

```
nosetests pandas
```
ends with:
```
    from . import hashtable, tslib, lib
ImportError: cannot import name hashtable
```

Author: lgautier

pandas/core/frame.py

@@ -5700,9 +5700,20 @@ def extract_index(data):
             elif isinstance(v, dict):
                 have_dicts = True
                 indexes.append(v.keys())
-            elif isinstance(v, (list, tuple, np.ndarray)):
-                have_raw_arrays = True
-                raw_lengths.append(len(v))
+            elif not (isinstance(v, str) or \
+                          isinstance(v, unicode)):
+                # Although strings have a __len__,
+                # they are likely to be considered scalars.
+                try:
+                    l = len(v)
+                except TypeError:
+                    continue
+                # Anything else with a __len__ is considered
+                # a sequence (to be safe, we check
+                # that there is __getitem__)
+                if hasattr(v, '__getitem__'):
+                    have_raw_arrays = True
+                    raw_lengths.append(l)
 
         if not indexes and not raw_lengths:
             raise ValueError('If using all scalar values, you must pass'