BUG: Fixed merging on tz-aware (#25033)

Author: TomAugspurger

doc/source/whatsnew/v0.24.1.rst

@@ -23,6 +23,7 @@ Fixed Regressions
 - Bug in :meth:`DataFrame.itertuples` with ``records`` orient raising an ``AttributeError`` when the ``DataFrame`` contained more than 255 columns (:issue:`24939`)
 - Bug in :meth:`DataFrame.itertuples` orient converting integer column names to strings prepended with an underscore (:issue:`24940`)
 - Fixed regression in :class:`Index.intersection` incorrectly sorting the values by default (:issue:`24959`).
+- Fixed regression in :func:`merge` when merging an empty ``DataFrame`` with multiple timezone-aware columns on one of the timezone-aware columns (:issue:`25014`).
 
 .. _whatsnew_0241.enhancements:
 

pandas/core/internals/concat.py

@@ -183,7 +183,7 @@ def get_reindexed_values(self, empty_dtype, upcasted_na):
                         is_datetime64tz_dtype(empty_dtype)):
                     if self.block is None:
                         array = empty_dtype.construct_array_type()
-                        return array(np.full(self.shape[1], fill_value),
+                        return array(np.full(self.shape[1], fill_value.value),
                                      dtype=empty_dtype)
                     pass
                 elif getattr(self.block, 'is_categorical', False):
@@ -335,8 +335,10 @@ def get_empty_dtype_and_na(join_units):
     elif 'category' in upcast_classes:
         return np.dtype(np.object_), np.nan
     elif 'datetimetz' in upcast_classes:
+        # GH-25014. We use NaT instead of iNaT, since this eventually
+        # ends up in DatetimeArray.take, which does not allow iNaT.
         dtype = upcast_classes['datetimetz']
-        return dtype[0], tslibs.iNaT
+        return dtype[0], tslibs.NaT
     elif 'datetime' in upcast_classes:
         return np.dtype('M8[ns]'), tslibs.iNaT
     elif 'timedelta' in upcast_classes:

pandas/tests/reshape/merge/test_merge.py

@@ -616,6 +616,24 @@ def test_merge_on_datetime64tz(self):
         assert result['value_x'].dtype == 'datetime64[ns, US/Eastern]'
         assert result['value_y'].dtype == 'datetime64[ns, US/Eastern]'
 
+    def test_merge_on_datetime64tz_empty(self):
+        # https://github.com/pandas-dev/pandas/issues/25014
+        dtz = pd.DatetimeTZDtype(tz='UTC')
+        right = pd.DataFrame({'date': [pd.Timestamp('2018', tz=dtz.tz)],
+                              'value': [4.0],
+                              'date2': [pd.Timestamp('2019', tz=dtz.tz)]},
+                             columns=['date', 'value', 'date2'])
+        left = right[:0]
+        result = left.merge(right, on='date')
+        expected = pd.DataFrame({
+            'value_x': pd.Series(dtype=float),
+            'date2_x': pd.Series(dtype=dtz),
+            'date': pd.Series(dtype=dtz),
+            'value_y': pd.Series(dtype=float),
+            'date2_y': pd.Series(dtype=dtz),
+        }, columns=['value_x', 'date2_x', 'date', 'value_y', 'date2_y'])
+        tm.assert_frame_equal(result, expected)
+
     def test_merge_datetime64tz_with_dst_transition(self):
         # GH 18885
         df1 = pd.DataFrame(pd.date_range(