BUG: fixed merging with empty frame containing an Int64 column (#25183)

doc/source/whatsnew/v0.24.2.rst

@@ -91,7 +91,7 @@ Bug Fixes
 **Other**
 
 - Bug in :meth:`Series.is_unique` where single occurrences of ``NaN`` were not considered unique (:issue:`25180`)
--
+- Bug in :func:`merge` when merging an empty ``DataFrame`` with an ``Int64`` column or a non-empty ``DataFrame`` with an ``Int64`` column that is all ``NaN`` (:issue:`25183`)
 -
 
 .. _whatsnew_0.242.contributors:

pandas/core/internals/concat.py

@@ -190,6 +190,8 @@ def get_reindexed_values(self, empty_dtype, upcasted_na):
                     pass
                 elif getattr(self.block, 'is_sparse', False):
                     pass
+                elif getattr(self.block, 'is_extension', False):
+                    pass
                 else:
                     missing_arr = np.empty(self.shape, dtype=empty_dtype)
                     missing_arr.fill(fill_value)

pandas/tests/reshape/merge/test_merge.py

@@ -2,6 +2,7 @@
 
 from collections import OrderedDict
 from datetime import date, datetime
+import itertools
 import random
 import re
 
@@ -428,6 +429,64 @@ def check2(exp, kwarg):
                 check1(exp_in, kwarg)
                 check2(exp_out, kwarg)
 
+    @pytest.mark.parametrize(
+        'join_col, val_col', list(itertools.product([
+            pd.Series([1], dtype='int64'),
+            pd.Series([1], dtype='Int64'),
+            pd.Series([1.23]),
+            pd.Series(['foo']),
+            pd.Series([True]),
+            pd.Series([pd.Timestamp('2018-01-01')]),
+            pd.Series([pd.Timestamp('2018-01-01', tz='US/Eastern')]),
+        ], repeat=2)),
+        ids=lambda x: x.dtype.name
+    )
+    def test_merge_empty_frame(self, join_col, val_col):
+        # GH 25183
+        df = pd.DataFrame({'a': join_col, 'b': val_col}, columns=['a', 'b'])
+        df_empty = df[:0]
+        exp = pd.DataFrame({
+            'b_x': pd.Series(dtype=df.dtypes['b']),
+            'a': pd.Series(dtype=df.dtypes['a']),
+            'b_y': pd.Series(dtype=df.dtypes['b']),
+        }, columns=['b_x', 'a', 'b_y'])
+        act = df_empty.merge(df, on='a')
+        assert_frame_equal(act, exp)
+
+    @pytest.mark.parametrize(
+        'join_col, val_col', list(itertools.product(
+            [
+                pd.Series([1], dtype='int64'),
+                pd.Series([1], dtype='Int64'),
+                pd.Series([1.23]),
+                pd.Series(['foo']),
+                pd.Series([True]),
+                pd.Series([pd.Timestamp('2018-01-01')]),
+                pd.Series([pd.Timestamp('2018-01-01', tz='US/Eastern')]),
+            ],
+            [
+                pd.Series([np.nan], dtype='Int64'),
+                pd.Series([np.nan], dtype='float'),
+                pd.Series([np.nan], dtype='object'),
+                pd.Series([pd.NaT]),
+            ]
+        )),
+        ids=lambda x: x.dtype.name
+    )
+    def test_merge_all_na_column(self, join_col, val_col):
+        # GH 25183
+        df_left = pd.DataFrame(
+            {'a': join_col, 'b': val_col}, columns=['a', 'b'])
+        df_right = pd.DataFrame(
+            {'a': join_col, 'b': val_col}, columns=['a', 'b'])
+        exp = pd.DataFrame({
+            'a': join_col,
+            'b_x': val_col,
+            'b_y': val_col,
+        }, columns=['a', 'b_x', 'b_y'])
+        act = df_left.merge(df_right, on='a')
+        assert_frame_equal(act, exp)
+
     def test_merge_nosort(self):
         # #2098, anything to do?