BUG: Ensure dataframe preserves categorical indices with categorial series

pandas/core/indexes/base.py

@@ -212,6 +212,7 @@
         PeriodArray,
     )
 
+
 __all__ = ["Index"]
 
 _unsortable_types = frozenset(("mixed", "mixed-integer"))
@@ -2912,7 +2913,22 @@ def union(self, other, sort=None):
         self._assert_can_do_setop(other)
         other, result_name = self._convert_can_do_setop(other)
 
-        if self.dtype != other.dtype:
+        if isinstance(self.dtype, CategoricalDtype) and isinstance(
+            other.dtype, CategoricalDtype
+        ):
+            both_categories = self.categories  # type: ignore[attr-defined]
+            if len(self.categories) != len(other.categories) or any(  # type: ignore[attr-defined]
+                self.categories != other.categories  # type: ignore[attr-defined]
+            ):
+                # Unite both categories
+                both_categories = np.union1d(self.categories, other.categories)  # type: ignore[attr-defined]
+            # if ordered and unordered, we set categories to be unordered
+            ordered = False if self.ordered != other.ordered else None  # type: ignore[attr-defined]
+            # Convert both indexes to have the same categories
+            self = self.set_categories(both_categories, ordered=ordered)  # type: ignore[attr-defined]
+            other = other.set_categories(both_categories, ordered=ordered)
+
+        elif self.dtype != other.dtype:
             if (
                 isinstance(self, ABCMultiIndex)
                 and not is_object_dtype(_unpack_nested_dtype(other))
@@ -3006,7 +3022,7 @@ def _union(self, other: Index, sort: bool | None):
         else:
             missing = algos.unique1d(self.get_indexer_non_unique(other)[1])
 
-        result: Index | MultiIndex | ArrayLike
+        result: Index | MultiIndex | CategoricalIndex | ArrayLike
         if self._is_multi:
             # Preserve MultiIndex to avoid losing dtypes
             result = self.append(other.take(missing))

pandas/tests/frame/test_constructors.py

@@ -2349,7 +2349,7 @@ def test_construct_with_two_categoricalindex_series(self):
         result = DataFrame([s1, s2])
         expected = DataFrame(
             np.array([[39, 6, 4, np.nan, np.nan], [152.0, 242.0, 150.0, 2.0, 2.0]]),
-            columns=["female", "male", "unknown", "f", "m"],
+            columns=CategoricalIndex(["female", "male", "unknown", "f", "m"]),
         )
         tm.assert_frame_equal(result, expected)
 

pandas/tests/indexes/test_setops.py

@@ -573,7 +573,7 @@ def test_union_duplicate_index_subsets_of_each_other(
 
     expected = Index([1, 2, 2, 3, 3, 4], dtype=dtype)
     if isinstance(a, CategoricalIndex):
-        expected = Index([1, 2, 2, 3, 3, 4])
+        expected = CategoricalIndex([1, 2, 2, 3, 3, 4])
     result = a.union(b)
     tm.assert_index_equal(result, expected)
     result = a.union(b, sort=False)
@@ -670,7 +670,7 @@ def test_union_with_duplicate_index_not_subset_and_non_monotonic(
     b = Index([0, 0, 1], dtype=dtype)
     expected = Index([0, 0, 1, 2], dtype=dtype)
     if isinstance(a, CategoricalIndex):
-        expected = Index([0, 0, 1, 2])
+        expected = CategoricalIndex([0, 0, 1, 2])
 
     result = a.union(b)
     tm.assert_index_equal(result, expected)

pandas/tests/reshape/concat/test_append.py

@@ -234,6 +234,8 @@ def test_append_different_columns_types(self, df_columns, series_index):
         result = df._append(ser)
         idx_diff = ser.index.difference(df_columns)
         combined_columns = Index(df_columns.tolist()).append(idx_diff)
+        if isinstance(result.columns, pd.CategoricalIndex):
+            combined_columns = pd.CategoricalIndex(combined_columns)
         expected = DataFrame(
             [
                 [1.0, 2.0, 3.0, np.nan, np.nan, np.nan],