Fixed str.split IndexError with Empty String and expand=True

doc/source/whatsnew/v0.23.0.txt

@@ -936,6 +936,7 @@ Indexing
 - Bug in :class:`IntervalIndex` where set operations that returned an empty ``IntervalIndex`` had the wrong dtype (:issue:`19101`)
 - Bug in :meth:`DataFrame.drop_duplicates` where no ``KeyError`` is raised when passing in columns that don't exist on the ``DataFrame`` (issue:`19726`)
 - Bug in ``Index`` subclasses constructors that ignore unexpected keyword arguments (:issue:`19348`)
+- Bug in :meth:`Series.str.split` where splitting an empty string without a `pater` and with ``expand=True`` would raise an ``IndexError`` (:issue:`20002`)
 
 
 MultiIndex

pandas/core/strings.py

@@ -1615,7 +1615,8 @@ def cons_row(x):
             if result:
                 # propagate nan values to match longest sequence (GH 18450)
                 max_len = max(len(x) for x in result)
-                result = [x * max_len if x[0] is np.nan else x for x in result]
+                result = [x * max_len if len(x) and x[0] is np.nan else x for x
+                          in result]
 
         if not isinstance(expand, bool):
             raise ValueError("expand must be True or False")

pandas/tests/test_strings.py

@@ -2144,6 +2144,13 @@ def test_split_nan_expand(self):
         # tm.assert_frame_equal does not differentiate
         assert all(np.isnan(x) for x in result.iloc[1])
 
+    def test_split_no_pat_and_expand(self):
+        # GH 20002
+        df = DataFrame({'foo': ['']})
+        exp = DataFrame([[]])
+        res = df['foo'].str.split(expand=True)
+        tm.assert_frame_equal(res, exp)
+
     def test_split_with_name(self):
         # GH 12617