BUG: Fix concat dtype preservation through concat

pandas/core/dtypes/concat.py

@@ -69,6 +69,7 @@ def concat_compat(
     -------
     a single array, preserving the combined dtypes
     """
+
     if len(to_concat) and lib.dtypes_all_equal([obj.dtype for obj in to_concat]):
         # fastpath!
         obj = to_concat[0]
@@ -92,6 +93,27 @@ def concat_compat(
                 to_concat_eas,
                 axis=axis,  # type: ignore[call-arg]
             )
+    # Special handling for categorical arrays solves #51362
+    if (
+        len(to_concat)
+        and all(isinstance(arr.dtype, CategoricalDtype) for arr in to_concat)
+        and axis == 0
+    ):
+        # Filter out empty arrays before union, similar to non_empties logic
+        non_empty_categoricals = [x for x in to_concat if _is_nonempty(x, axis)]
+
+        if len(non_empty_categoricals) == 0:
+            # All arrays are empty, return the first one (they're all categorical)
+            return to_concat[0]
+        elif len(non_empty_categoricals) == 1:
+            # Only one non-empty array, return it directly
+            return non_empty_categoricals[0]
+        else:
+            # Multiple non-empty arrays, use union_categoricals
+            return union_categoricals(
+                non_empty_categoricals, sort_categories=True
+            )  # Performance cost, but necessary to keep tests passing.
+            # see pandas/tests/reshape/concat/test_append_common.py:498
 
     # If all arrays are empty, there's nothing to convert, just short-cut to
     # the concatenation, #3121.

pandas/tests/dtypes/test_concat.py

@@ -3,7 +3,10 @@
 import pandas.core.dtypes.concat as _concat
 
 import pandas as pd
-from pandas import Series
+from pandas import (
+    DataFrame,
+    Series,
+)
 import pandas._testing as tm
 
 
@@ -14,12 +17,12 @@ def test_concat_mismatched_categoricals_with_empty():
 
     result = _concat.concat_compat([ser1._values, ser2._values])
     expected = pd.concat([ser1, ser2])._values
-    tm.assert_numpy_array_equal(result, expected)
+    tm.assert_categorical_equal(result, expected)
 
 
 def test_concat_single_dataframe_tz_aware():
     # https://github.com/pandas-dev/pandas/issues/25257
-    df = pd.DataFrame(
+    df = DataFrame(
         {"timestamp": [pd.Timestamp("2020-04-08 09:00:00.709949+0000", tz="UTC")]}
     )
     expected = df.copy()
@@ -53,7 +56,7 @@ def test_concat_series_between_empty_and_tzaware_series(using_infer_string):
     ser2 = Series(dtype=float)
 
     result = pd.concat([ser1, ser2], axis=1)
-    expected = pd.DataFrame(
+    expected = DataFrame(
         data=[
             (0.0, None),
         ],
@@ -64,3 +67,21 @@ def test_concat_series_between_empty_and_tzaware_series(using_infer_string):
         dtype=float,
     )
     tm.assert_frame_equal(result, expected)
+
+
+def test_concat_categorical_dataframes():
+    df = DataFrame({"a": [0, 1]}, dtype="category")
+    df2 = DataFrame({"a": [2, 3]}, dtype="category")
+
+    result = pd.concat([df, df2], axis=0)
+
+    assert result["a"].dtype.name == "category"
+
+
+def test_concat_categorical_series():
+    ser = Series([0, 1], dtype="category")
+    ser2 = Series([2, 3], dtype="category")
+
+    result = pd.concat([ser, ser2], axis=0)
+
+    assert result.dtype.name == "category"