BUG: Fixed concat with reindex and extension types

doc/source/whatsnew/v1.1.0.rst

@@ -1071,6 +1071,7 @@ ExtensionArray
 ^^^^^^^^^^^^^^
 
 - Fixed bug where :meth:`Series.value_counts` would raise on empty input of ``Int64`` dtype (:issue:`33317`)
+- Fixed bug in :func:`concat` when concatenating DataFrames with non-overlaping columns resulting in object-dtype columns rather than preserving the extension dtype (:issue:`27692`, :issue:`33027`)
 - Fixed bug where :meth:`StringArray.isna` would return ``False`` for NA values when ``pandas.options.mode.use_inf_as_na`` was set to ``True`` (:issue:`33655`)
 - Fixed bug in :class:`Series` construction with EA dtype and index but no data or scalar data fails (:issue:`26469`)
 - Fixed bug that caused :meth:`Series.__repr__()` to crash for extension types whose elements are multidimensional arrays (:issue:`33770`).

pandas/core/internals/concat.py

@@ -29,7 +29,7 @@
 
 
 def concatenate_block_managers(
-    mgrs_indexers, axes, concat_axis: int, copy: bool
+    mgrs_indexers, axes, concat_axis: int, copy: bool,
 ) -> BlockManager:
     """
     Concatenate block managers into one.
@@ -76,7 +76,7 @@ def concatenate_block_managers(
             b = make_block(values, placement=placement, ndim=blk.ndim)
         else:
             b = make_block(
-                _concatenate_join_units(join_units, concat_axis, copy=copy),
+                _concatenate_join_units(join_units, concat_axis, copy=copy,),
                 placement=placement,
             )
         blocks.append(b)
@@ -260,6 +260,16 @@ def get_reindexed_values(self, empty_dtype, upcasted_na):
                     pass
                 elif getattr(self.block, "is_extension", False):
                     pass
+                elif is_extension_array_dtype(empty_dtype):
+                    missing_arr = empty_dtype.construct_array_type()._from_sequence(
+                        [], dtype=empty_dtype
+                    )
+                    ncols, nrows = self.shape
+                    assert ncols == 1, ncols
+                    empty_arr = -1 * np.ones((nrows,), dtype="int8")
+                    return missing_arr.take(
+                        empty_arr, allow_fill=True, fill_value=fill_value
+                    )
                 else:
                     missing_arr = np.empty(self.shape, dtype=empty_dtype)
                     missing_arr.fill(fill_value)
@@ -329,7 +339,7 @@ def _concatenate_join_units(join_units, concat_axis, copy):
             # 2D to put it a non-EA Block
             concat_values = np.atleast_2d(concat_values)
     else:
-        concat_values = concat_compat(to_concat, axis=concat_axis)
+        concat_values = concat_compat(to_concat, axis=concat_axis,)
 
     return concat_values
 
@@ -374,6 +384,10 @@ def _get_empty_dtype_and_na(join_units):
             upcast_cls = "category"
         elif is_datetime64tz_dtype(dtype):
             upcast_cls = "datetimetz"
+
+        elif is_extension_array_dtype(dtype):
+            upcast_cls = "extension"
+
         elif issubclass(dtype.type, np.bool_):
             upcast_cls = "bool"
         elif issubclass(dtype.type, np.object_):
@@ -384,8 +398,6 @@ def _get_empty_dtype_and_na(join_units):
             upcast_cls = "timedelta"
         elif is_sparse(dtype):
             upcast_cls = dtype.subtype.name
-        elif is_extension_array_dtype(dtype):
-            upcast_cls = "object"
         elif is_float_dtype(dtype) or is_numeric_dtype(dtype):
             upcast_cls = dtype.name
         else:
@@ -401,6 +413,12 @@ def _get_empty_dtype_and_na(join_units):
 
     if not upcast_classes:
         upcast_classes = null_upcast_classes
+    if "extension" in upcast_classes:
+        if len(upcast_classes) == 1:
+            cls = upcast_classes["extension"][0]
+            return cls, cls.na_value
+        else:
+            return np.dtype("object"), np.nan
 
     # TODO: de-duplicate with maybe_promote?
     # create the result

pandas/core/reshape/concat.py

@@ -497,7 +497,7 @@ def get_result(self):
                 mgrs_indexers.append((obj._mgr, indexers))
 
             new_data = concatenate_block_managers(
-                mgrs_indexers, self.new_axes, concat_axis=self.bm_axis, copy=self.copy
+                mgrs_indexers, self.new_axes, concat_axis=self.bm_axis, copy=self.copy,
             )
             if not self.copy:
                 new_data._consolidate_inplace()

pandas/tests/extension/base/reshaping.py

@@ -107,6 +107,19 @@ def test_concat_extension_arrays_copy_false(self, data, na_value):
         result = pd.concat([df1, df2], axis=1, copy=False)
         self.assert_frame_equal(result, expected)
 
+    def test_concat_with_reindex(self, data):
+        # GH-33027
+        a = pd.DataFrame({"a": data[:5]})
+        b = pd.DataFrame({"b": data[:5]})
+        result = pd.concat([a, b], ignore_index=True)
+        expected = pd.DataFrame(
+            {
+                "a": data.take(list(range(5)) + ([-1] * 5), allow_fill=True),
+                "b": data.take(([-1] * 5) + list(range(5)), allow_fill=True),
+            }
+        )
+        self.assert_frame_equal(result, expected)
+
     def test_align(self, data, na_value):
         a = data[:3]
         b = data[2:5]

pandas/tests/extension/test_categorical.py

@@ -93,7 +93,8 @@ class TestConstructors(base.BaseConstructorsTests):
 
 
 class TestReshaping(base.BaseReshapingTests):
-    pass
+    def test_concat_with_reindex(self, data):
+        pytest.xfail(reason="Deliberately upcast to object?")
 
 
 class TestGetitem(base.BaseGetitemTests):