change after review

Author: makbigc

pandas/core/reshape/merge.py

@@ -1607,7 +1607,9 @@ def _factorize_keys(lk, rk, sort=True):
 
         lk = ensure_int64(lk.codes)
         rk = ensure_int64(rk)
-    elif is_extension_array_dtype(lk) and is_extension_array_dtype(rk):
+    elif (is_extension_array_dtype(lk) and
+          is_extension_array_dtype(rk) and
+          lk.dtype == rk.dtype):
         klass = libhashtable.Factorizer
         lk = ensure_object(lk)
         rk = ensure_object(rk)
@@ -1620,9 +1622,8 @@ def _factorize_keys(lk, rk, sort=True):
     elif (issubclass(lk.dtype.type, (np.timedelta64, np.datetime64)) and
           issubclass(rk.dtype.type, (np.timedelta64, np.datetime64))):
         # GH#23917 TODO: Needs tests for non-matching dtypes
-        klass = libhashtable.Int64Factorizer
-        lk = ensure_int64(com.values_from_object(lk))
-        rk = ensure_int64(com.values_from_object(rk))
+        lk, _ = lk._values_for_factorize()
+        rk, _ = rk._values_for_factorize()
     else:
         klass = libhashtable.Factorizer
         lk = ensure_object(lk)

pandas/tests/extension/base/reshaping.py

@@ -237,3 +237,11 @@ def test_unstack(self, data, index, obj):
             result = result.astype(object)
 
             self.assert_frame_equal(result, expected)
+
+    def test_merge_on_int_array(self, df_merge_on_int_array):
+        # GH  23020
+        result = pd.merge(df_merge_on_int_array, df_merge_on_int_array, on='A')
+        expected = pd.DataFrame({'A': pd.Series([1, 2, np.nan], dtype='Int64'),
+                                 'B_x': 1,
+                                 'B_y': 1})
+        self.assert_frame_equal(result, expected, check_dtype=True)

pandas/tests/extension/conftest.py

@@ -1,3 +1,5 @@
+import pandas as pd
+import numpy as np
 import operator
 
 import pytest
@@ -108,3 +110,9 @@ def data_for_grouping():
 def box_in_series(request):
     """Whether to box the data in a Series"""
     return request.param
+
+
+@pytest.fixture
+def df_merge_on_int_array():
+    return pd.DataFrame({'A': pd.Series([1, 2, np.nan], dtype='Int64'),
+                         'B': 1})

pandas/tests/reshape/merge/test_merge.py

@@ -1326,16 +1326,6 @@ def test_merging_with_bool_or_int_cateorical_column(self, category_column,
             CDT(categories, ordered=ordered))
         assert_frame_equal(expected, result)
 
-    def test_merge_on_int_array(self):
-        # GH  23020
-        df = pd.DataFrame({'A': pd.Series([1, 2, np.nan], dtype='Int64'),
-                           'B': 1})
-        result = pd.merge(df, df, on='A')
-        expected = pd.DataFrame({'A': pd.Series([1, 2, np.nan], dtype='Int64'),
-                                 'B_x': 1,
-                                 'B_y': 1})
-        assert_frame_equal(result, expected, check_dtype=True)
-
 
 @pytest.fixture
 def left_df():