GH-46179: [Python] Bump index level once if pandas df already contains __index_level_i__ column

python/pyarrow/pandas_compat.py

@@ -378,7 +378,10 @@ def _index_level_name(index, i, column_names):
     if index.name is not None and index.name not in column_names:
         return _column_name_to_strings(index.name)
     else:
-        return f'__index_level_{i:d}__'
+        j = i
+        while f'__index_level_{j:d}__' in column_names:
+            j += 1
+        return f'__index_level_{j:d}__'
 
 
 def _get_columns_to_convert(df, schema, preserve_index, columns):
@@ -419,7 +422,9 @@ def _get_columns_to_convert(df, schema, preserve_index, columns):
     index_descriptors = []
     index_column_names = []
     for i, index_level in enumerate(index_levels):
-        name = _index_level_name(index_level, i, column_names)
+        name = _index_level_name(
+            index_level, i, column_names + index_column_names
+        )
         if (isinstance(index_level, _pandas_api.pd.RangeIndex) and
                 preserve_index is None):
             descr = _get_range_index_descriptor(index_level)

python/pyarrow/tests/test_pandas.py

@@ -404,14 +404,50 @@ def test_index_metadata_field_name(self):
         assert col3['name'] == col3['field_name']
 
         idx0_descr, foo_descr = js['index_columns']
-        assert idx0_descr == '__index_level_0__'
+        # __index_level_0__ exists, unnamed bumped to __index_level_1__
+        assert idx0_descr == '__index_level_1__'
         assert idx0['field_name'] == idx0_descr
         assert idx0['name'] is None
 
         assert foo_descr == 'foo'
         assert foo['field_name'] == foo_descr
         assert foo['name'] == foo_descr
 
+    def test_index_level_name_bump(self):
+        # GH-46179
+        df = pd.DataFrame(
+            {"col": [1, 2, 3], "__index_level_0__": [4, 5, 6]},
+            index=[10, 20, 30],
+        )
+        _check_pandas_roundtrip(df, preserve_index=True)
+
+        # Explicit test
+        t = pa.table(df)
+        expected_schema = pa.schema([
+            ("col", pa.int64()),
+            ("__index_level_0__", pa.int64()),
+            ("__index_level_1__", pa.int64())
+        ])
+        assert t.schema.equals(expected_schema)
+
+        df2 = t.to_pandas()
+        assert df2.index.equals(pd.Index([10, 20, 30]))
+        assert df2.ndim == df.ndim == 2
+
+    def test_index_level_name_bump_multiindex(self):
+        # GH-46179
+        df = pd.DataFrame(
+            {"col": [1, 2], "__index_level_0__": [3, 4]},
+            index=pd.MultiIndex.from_arrays(
+                [[10, 20], [100, 200]], names=[None, None]
+            ),
+        )
+        _check_pandas_roundtrip(df, preserve_index=True)
+
+        t = pa.Table.from_pandas(df, preserve_index=True)
+        assert t.schema.names == ['col', '__index_level_0__',
+                                  '__index_level_1__', '__index_level_2__']
+
     def test_categorical_column_index(self):
         df = pd.DataFrame(
             [(1, 'a', 2.0), (2, 'b', 3.0), (3, 'c', 4.0)],