BUG: Fixed merge on dtype equal categories

doc/source/whatsnew/v0.23.0.txt

@@ -679,6 +679,9 @@ Categorical
 - Bug in :func:`pandas.api.types.union_categoricals` returning the wrong result
   when all the categoricals had the same categories, but in a different order.
   This affected :func:`pandas.concat` with Categorical data (:issue:`19096`).
+- Bug in :func:`pandas.merge` when joining on a ``Categorical`` column where
+  the categories were unordered and had the same values but in a different
+  order (:issue:`19551`)
 - Bug in ``Categorical.equals`` between two unordered categories with the same categories, but in a different order (:issue:`16603`)
 - Bug in :meth:`Index.astype` with a categorical dtype where the resultant index is not converted to a :class:`CategoricalIndex` for all types of index (:issue:`18630`)
 - Bug in :meth:`Series.astype` and ``Categorical.astype()`` where an existing categorical data does not get updated (:issue:`10696`, :issue:`18593`)

pandas/core/reshape/merge.py

@@ -12,6 +12,7 @@
 
 from pandas import (Categorical, DataFrame,
                     Index, MultiIndex, Timedelta)
+from pandas.core.arrays.categorical import _recode_for_categories
 from pandas.core.frame import _merge_doc
 from pandas.core.dtypes.common import (
     is_datetime64tz_dtype,
@@ -1540,8 +1541,15 @@ def _factorize_keys(lk, rk, sort=True):
             is_categorical_dtype(rk) and
             lk.is_dtype_equal(rk)):
         klass = libhashtable.Int64Factorizer
+
+        if lk.categories.equals(rk.categories):
+            rk = rk.codes
+        else:
+            # Same categories in different orders -> recode
+            rk = _recode_for_categories(rk.codes, rk.categories, lk.categories)
+
         lk = _ensure_int64(lk.codes)
-        rk = _ensure_int64(rk.codes)
+        rk = _ensure_int64(rk)
     elif is_int_or_datetime_dtype(lk) and is_int_or_datetime_dtype(rk):
         klass = libhashtable.Int64Factorizer
         lk = _ensure_int64(com._values_from_object(lk))

pandas/tests/reshape/merge/test_merge.py

@@ -1643,6 +1643,25 @@ def test_merge_categorical(self):
         result = pd.merge(cleft, cright, how='left', left_on='b', right_on='c')
         tm.assert_frame_equal(result, expected)
 
+    def tests_merge_categorical_unordered_equal(self):
+        # GH-19551
+        df1 = DataFrame({
+            'Foo': Categorical(['A', 'B', 'C'], categories=['A', 'B', 'C']),
+            'Left': ['A0', 'B0', 'C0'],
+        })
+
+        df2 = DataFrame({
+            'Foo': Categorical(['C', 'B', 'A'], categories=['C', 'B', 'A']),
+            'Right': ['C1', 'B1', 'A1'],
+        })
+        result = pd.merge(df1, df2, on=['Foo'])
+        expected = DataFrame({
+            'Foo': pd.Categorical(['A', 'B', 'C']),
+            'Left': ['A0', 'B0', 'C0'],
+            'Right': ['A1', 'B1', 'C1'],
+        })
+        assert_frame_equal(result, expected)
+
     def test_other_columns(self, left, right):
         # non-merge columns should preserve if possible
         right = right.assign(Z=right.Z.astype('category'))