additional revisions

Author: veljanin

doc/source/whatsnew/v3.0.0.rst

@@ -544,7 +544,7 @@ Bug fixes
 
 Categorical
 ^^^^^^^^^^^
--
+- Bug in :func:`convert_dtypes` with ``dtype_backend='pyarrow'`` parameter where empty categorical series raise error or get converted to null[pyarrow] (:issue:`59934`)
 -
 
 Datetimelike

pandas/core/dtypes/cast.py

@@ -1112,7 +1112,7 @@ def convert_dtypes(
 
     else:
         inferred_dtype = input_array.dtype
-    
+
     if dtype_backend == "pyarrow":
         from pandas.core.arrays.arrow.array import to_pyarrow_type
         from pandas.core.arrays.string_ import StringDtype
@@ -1143,14 +1143,15 @@ def convert_dtypes(
                 base_dtype.kind == "O"  # type: ignore[union-attr]
                 and input_array.size > 0
                 and isna(input_array).all()
+                and not isinstance(input_array.dtype, CategoricalDtype)
             ):
                 import pyarrow as pa
-                
+
                 pa_type = pa.null()
             else:
                 pa_type = to_pyarrow_type(base_dtype)
             if pa_type is not None:
-                if isna(input_array).all() and hasattr(input_array, 'categories'):
+                if isna(input_array).all() and hasattr(input_array, "categories"):
                     inferred_dtype = input_array.dtype
                 else:
                     inferred_dtype = ArrowDtype(pa_type)

pandas/tests/frame/methods/test_convert_dtypes.py

@@ -34,22 +34,27 @@ def test_convert_empty(self):
         # Empty DataFrame can pass convert_dtypes, see GH#40393
         empty_df = pd.DataFrame()
         tm.assert_frame_equal(empty_df, empty_df.convert_dtypes())
-    
+
     def test_convert_empty_categorical_to_pyarrow(self):
+        # GH#59934
         df = pd.DataFrame(
             {
-                "A": pd.Series(pd.Categorical([None] * 5)),
-                "B": pd.Series(pd.Categorical([None] * 5, categories=["B1", "B2"])),
-             }
+                "A": pd.Categorical([None] * 5),
+                "B": pd.Categorical([None] * 5, categories=["B1", "B2"]),
+            }
         )
         converted = df.convert_dtypes(dtype_backend="pyarrow")
         expected = df
         tm.assert_frame_equal(converted, expected)
-        
-        assert df.A.dtype == "category", "Dtype in column A is not 'category'"
-        assert df.B.dtype == "category", "Dtype in column B is not 'category'"
-        assert df.A.cat.categories.empty, "Categories in column A are not empty"
-        assert (df.B.cat.categories == ["B1", "B2"]).all(), "Categories in column A are not empty"
+
+        assert converted.A.dtype == "category", "Dtype in column A is not 'category'"
+        assert converted.B.dtype == "category", "Dtype in column B is not 'category'"
+        assert converted.A.cat.categories.empty, "Categories in column A are not empty"
+        assert converted.B.cat.categories.__contains__(
+            "B1"
+        ) and converted.B.cat.categories.__contains__(
+            "B2"
+        ), "Categories in column B doesn't contain adequate categories"
 
     def test_convert_dtypes_retain_column_names(self):
         # GH#41435

pandas/tests/series/methods/test_convert_dtypes.py

@@ -297,16 +297,21 @@ def test_convert_dtypes_pyarrow_null(self):
         result = ser.convert_dtypes(dtype_backend="pyarrow")
         expected = pd.Series([None, None], dtype=pd.ArrowDtype(pa.null()))
         tm.assert_series_equal(result, expected)
-    
+
     def test_convert_empty_categorical_to_pyarrow(self):
-        ser = pd.Series(pd.Series(pd.Categorical([None] * 5)))
-        
-        converted = ser.convert_dtypes(dtype_backend="pyarrow")
-        expected = ser
-        tm.assert_series_equal(converted, expected)
-        
-        assert ser.dtype == "category", "Series dtype is not 'category'"
-        assert ser.cat.categories.empty, "Series categories are not empty"
-        
-        ser2 = pd.Series(pd.Series(pd.Categorical([None] * 5, categories=["S1", "S2"])))
-        assert (ser2.cat.categories == ["S1", "S2"]).all(), "Series categories are not empty"
+        # GH#59934
+        ser1 = pd.Series(pd.Categorical([None] * 5))
+        converted1 = ser1.convert_dtypes(dtype_backend="pyarrow")
+        expected = ser1
+
+        tm.assert_series_equal(converted1, expected)
+        assert converted1.dtype == "category", "Series dtype is not 'category'"
+        assert converted1.cat.categories.empty, "Series categories are not empty"
+
+        ser2 = pd.Series(pd.Categorical([None] * 5, categories=["S1", "S2"]))
+        converted2 = ser2.convert_dtypes(dtype_backend="pyarrow")
+        assert converted2.cat.categories.__contains__(
+            "S1"
+        ) and converted2.cat.categories.__contains__(
+            "S2"
+        ), "Categories in ser2 doesn't contain adequate categories"