Fix PandasBlocks implementation for missmatching categories

partd/pandas.py

@@ -211,6 +211,17 @@ def join(dfs):
     if not dfs:
         return pd.DataFrame()
     else:
-        return pd.concat(dfs)
+        result = pd.concat(dfs)
+        dtypes = {
+            col: "category"
+            for col in result.columns
+            if (
+                isinstance(dfs[0][col].dtype, pd.CategoricalDtype)
+                and not isinstance(result[col].dtype, pd.CategoricalDtype)
+            )
+        }
+        if dtypes:
+            result = result.astype(dtypes)
+        return result
 
 PandasBlocks = partial(Encode, serialize, deserialize, join)

partd/tests/test_pandas.py

@@ -146,3 +146,17 @@ def test_index_non_numeric_extension_types(dtype):
     df.index = df.index.astype(dtype)
     df2 = deserialize(serialize(df))
     tm.assert_frame_equal(df, df2)
+
+
+def test_categorical_concat():
+    pytest.importorskip("pandas", minversion="2")
+
+    df1 = pd.DataFrame({"a": ["x", "y"]}, dtype="category")
+    df2 = pd.DataFrame({"a": ["y", "z"]}, dtype="category")
+
+    with PandasBlocks() as p:
+        p.append({'x': df1})
+        p.append({'x': df2})
+
+        result = p.get(["x"])
+        pd.testing.assert_frame_equal(result[0], pd.concat([df1, df2]).astype("category"))