Fix _assert_shared (#2136)

* Fix _assert_shared

Author: djkirkham

lib/iris/pandas.py

@@ -1,4 +1,4 @@
-# (C) British Crown Copyright 2013 - 2015, Met Office
+# (C) British Crown Copyright 2013 - 2016, Met Office
 #
 # This file is part of Iris.
 #
@@ -129,6 +129,15 @@ def _assert_shared(np_obj, pandas_obj):
         base = pandas_obj.base
     else:
         base = pandas_obj[0].base
+
+    # Prior to Pandas 0.17, when pandas_obj is a Series, pandas_obj.values
+    # returns a view of the underlying array, and pandas_obj.base, which calls
+    # pandas_obj.values.base, returns the underlying array. In 0.17 and 0.18
+    # pandas_obj.values returns the underlying array, so base may be None even
+    # if the array is shared.
+    if base is None:
+        base = pandas_obj.values
+
     # Chase the stack of NumPy `base` references back to see if any of
     # them are our original array.
     while base is not None: