REGR: setting column with setitem should not modify existing array inplace

Author: jorisvandenbossche

pandas/core/indexing.py

@@ -1684,15 +1684,21 @@ def isetter(loc, v):
                     com.is_null_slice(idx) or com.is_full_slice(idx, len(self.obj))
                     for idx in pi
                 ):
-                    ser = v
+                    if ser._mgr.any_extension_types:
+                        # avoid `iset` for extension arrays, as this doesn't
+                        # change the underlying values inplace (GH33457)
+                        ser._mgr = ser._mgr.setitem(indexer=pi, value=v)
+                    else:
+                        ser = v
+                        self.obj._iset_item(loc, ser)
                 else:
                     # set the item, possibly having a dtype change
                     ser = ser.copy()
                     ser._mgr = ser._mgr.setitem(indexer=pi, value=v)
                     ser._maybe_update_cacher(clear=True)
 
-                # reset the sliced object if unique
-                self.obj._iset_item(loc, ser)
+                    # reset the sliced object if unique
+                    self.obj._iset_item(loc, ser)
 
             # we need an iterable, with a ndim of at least 1
             # eg. don't pass through np.array(0)

pandas/core/internals/blocks.py

@@ -1589,7 +1589,7 @@ def should_store(self, value: ArrayLike) -> bool:
 
     def set(self, locs, values):
         assert locs.tolist() == [0]
-        self.values[:] = values
+        self.values = values
 
     def putmask(
         self, mask, new, inplace: bool = False, axis: int = 0, transpose: bool = False,

pandas/tests/indexing/test_iloc.py

@@ -705,6 +705,15 @@ def test_iloc_setitem_categorical_updates_inplace(self):
         expected = pd.Categorical(["C", "B", "A"])
         tm.assert_categorical_equal(cat, expected)
 
+        # __setitem__ under the other hand does not work in-place
+        cat = pd.Categorical(["A", "B", "C"])
+        df = pd.DataFrame({1: cat, 2: [1, 2, 3]})
+
+        df["cat"] = cat[::-1]
+
+        expected = pd.Categorical(["A", "B", "C"])
+        tm.assert_categorical_equal(cat, expected)
+
     def test_iloc_with_boolean_operation(self):
         # GH 20627
         result = DataFrame([[0, 1], [2, 3], [4, 5], [6, np.nan]])

pandas/tests/indexing/test_indexing.py

@@ -1100,3 +1100,22 @@ def test_long_text_missing_labels_inside_loc_error_message_limited():
     error_message_regex = "long_missing_label_text_0.*\\\\n.*long_missing_label_text_1"
     with pytest.raises(KeyError, match=error_message_regex):
         s.loc[["a", "c"] + missing_labels]
+
+
+def test_setitem_EA_column_update():
+    # https://github.com/pandas-dev/pandas/issues/33457
+
+    df = pd.DataFrame(
+        {
+            "int": [1, 2, 3],
+            "int2": [3, 4, 5],
+            "float": [0.1, 0.2, 0.3],
+            "EA": pd.array([1, 2, None], dtype="Int64"),
+        }
+    )
+    original_arr = df.EA.array
+
+    # overwrite column with new array
+    df["EA"] = pd.array([1, 2, 3], dtype="Int64")
+    assert original_arr is not df.EA.array
+    tm.assert_extension_array_equal(original_arr, pd.array([1, 2, None], dtype="Int64"))