BUG: ensure to always return new objects in Index set operations (avoid metadata mutation) (#63174)

Author: deniskrds

doc/source/whatsnew/v3.0.0.rst

@@ -1227,6 +1227,7 @@ Indexing
 - Bug in :meth:`DataFrame.loc.__getitem__` and :meth:`DataFrame.iloc.__getitem__` with a :class:`CategoricalDtype` column with integer categories raising when trying to index a row containing a ``NaN`` entry (:issue:`58954`)
 - Bug in :meth:`Index.__getitem__` incorrectly raising with a 0-dim ``np.ndarray`` key (:issue:`55601`)
 - Bug in :meth:`Index.get_indexer` not casting missing values correctly for new string datatype (:issue:`55833`)
+- Bug in :meth:`Index.intersection`, :meth:`Index.union`, :meth:`MultiIndex.intersection`, and :meth:`MultiIndex.union` returning a reference to the original Index instead of a new instance when operating on identical indexes, which could cause metadata corruption when modifying the result (:issue:`63169`)
 - Bug in adding new rows with :meth:`DataFrame.loc.__setitem__` or :class:`Series.loc.__setitem__` which failed to retain dtype on the object's index in some cases (:issue:`41626`)
 - Bug in indexing on a :class:`DatetimeIndex` with a ``timestamp[pyarrow]`` dtype or on a :class:`TimedeltaIndex` with a ``duration[pyarrow]`` dtype (:issue:`62277`)
 

pandas/core/indexes/base.py

@@ -2984,13 +2984,12 @@ def __bool__(self) -> NoReturn:
     def _get_reconciled_name_object(self, other):
         """
         If the result of a set operation will be self,
-        return self, unless the name changes, in which
-        case make a shallow copy of self.
+        return a shallow copy of self.
         """
         name = get_op_result_name(self, other)
         if self.name is not name:
             return self.rename(name)
-        return self
+        return self.copy(deep=False)
 
     @final
     def _validate_sort_keyword(self, sort) -> None:

pandas/core/indexes/multi.py

@@ -4099,13 +4099,12 @@ def _is_comparable_dtype(self, dtype: DtypeObj) -> bool:
     def _get_reconciled_name_object(self, other) -> MultiIndex:
         """
         If the result of a set operation will be self,
-        return self, unless the names change, in which
-        case make a shallow copy of self.
+        return a shallow copy of self.
         """
         names = self._maybe_match_names(other)
         if self.names != names:
             return self.rename(names)
-        return self
+        return self.copy(deep=False)
 
     def _maybe_match_names(self, other):
         """

pandas/tests/indexes/test_setops.py

@@ -727,7 +727,7 @@ def test_intersection(self, index, sort):
 
         # Corner cases
         inter = first.intersection(first, sort=sort)
-        assert inter is first
+        assert inter is not first
 
     @pytest.mark.parametrize(
         "index2_name,keeps_name",
@@ -812,16 +812,16 @@ def test_union_identity(self, index, sort):
         first = index[5:20]
 
         union = first.union(first, sort=sort)
-        # i.e. identity is not preserved when sort is True
-        assert (union is first) is (not sort)
+        # GH#63169 - identity is not preserved to prevent shared mutable state
+        assert union is not first
 
         # This should no longer be the same object, since [] is not consistent,
         # both objects will be recast to dtype('O')
         union = first.union(Index([], dtype=first.dtype), sort=sort)
-        assert (union is first) is (not sort)
+        assert union is not first
 
         union = Index([], dtype=first.dtype).union(first, sort=sort)
-        assert (union is first) is (not sort)
+        assert union is not first
 
     @pytest.mark.parametrize("index", ["string"], indirect=True)
     @pytest.mark.parametrize("second_name,expected", [(None, None), ("name", "name")])
@@ -984,3 +984,83 @@ def test_union_pyarrow_timestamp(self):
         res = left.union(right)
         expected = Index(["2020-01-01", "2020-01-02"], dtype=left.dtype)
         tm.assert_index_equal(res, expected)
+
+
+def test_intersection_mutation_safety():
+    # GH#63169
+    index1 = Index([0, 1], name="original")
+    index2 = Index([0, 1], name="original")
+
+    result = index1.intersection(index2)
+
+    assert result is not index1
+    assert result is not index2
+
+    tm.assert_index_equal(result, index1)
+    assert result.name == "original"
+
+    index1.name = "changed"
+
+    assert result.name == "original"
+    assert index1.name == "changed"
+
+
+def test_union_mutation_safety():
+    # GH#63169
+    index1 = Index([0, 1], name="original")
+    index2 = Index([0, 1], name="original")
+
+    result = index1.union(index2)
+
+    assert result is not index1
+    assert result is not index2
+
+    tm.assert_index_equal(result, index1)
+    assert result.name == "original"
+
+    index1.name = "changed"
+
+    assert result.name == "original"
+    assert index1.name == "changed"
+
+
+def test_union_mutation_safety_other():
+    # GH#63169
+    index1 = Index([0, 1], name="original")
+    index2 = Index([0, 1], name="original")
+
+    result = index1.union(index2)
+
+    assert result is not index2
+
+    tm.assert_index_equal(result, index2)
+    assert result.name == "original"
+
+    index2.name = "changed"
+
+    assert result.name == "original"
+    assert index2.name == "changed"
+
+
+def test_multiindex_intersection_mutation_safety():
+    # GH#63169
+    mi1 = MultiIndex.from_tuples([("a", 1), ("b", 2)], names=["x", "y"])
+    mi2 = MultiIndex.from_tuples([("a", 1), ("b", 2)], names=["x", "y"])
+
+    result = mi1.intersection(mi2)
+    assert result is not mi1
+
+    mi1.names = ["changed1", "changed2"]
+    assert result.names == ["x", "y"]
+
+
+def test_multiindex_union_mutation_safety():
+    # GH#63169
+    mi1 = MultiIndex.from_tuples([("a", 1), ("b", 2)], names=["x", "y"])
+    mi2 = MultiIndex.from_tuples([("a", 1), ("b", 2)], names=["x", "y"])
+
+    result = mi1.union(mi2)
+    assert result is not mi1
+
+    mi1.names = ["changed1", "changed2"]
+    assert result.names == ["x", "y"]

pandas/tests/indexes/timedeltas/test_setops.py

@@ -114,7 +114,7 @@ def test_intersection_bug_1708(self):
 
     def test_intersection_equal(self, sort):
         # GH 24471 Test intersection outcome given the sort keyword
-        # for equal indices intersection should return the original index
+        # GH#63169 intersection returns a copy to prevent shared mutable state
         first = timedelta_range("1 day", periods=4, freq="h")
         second = timedelta_range("1 day", periods=4, freq="h")
         intersect = first.intersection(second, sort=sort)
@@ -124,7 +124,8 @@ def test_intersection_equal(self, sort):
 
         # Corner cases
         inter = first.intersection(first, sort=sort)
-        assert inter is first
+        assert inter is not first
+        tm.assert_index_equal(inter, first)
 
     @pytest.mark.parametrize("period_1, period_2", [(0, 4), (4, 0)])
     def test_intersection_zero_length(self, period_1, period_2, sort):