gh-143189: fix insertdict() for non-Unicode key

When iserting non unicode key into split table, matching Unicode key
may be in the shared split table without its value.

Author: methane

Lib/test/test_dict.py

@@ -1602,6 +1602,7 @@ def __hash__(self):
             d.get(key2)
 
     def test_clear_at_lookup(self):
+        # gh-140551 dict crash if clear is called at lookup stage
         class X:
             def __hash__(self):
                 return 1
@@ -1622,13 +1623,31 @@ def __eq__(self, other):
         self.assertEqual(len(d), 1)
 
     def test_split_table_update_with_str_subclass(self):
+        # gh-142218: inserting into a split table dictionary with a non str
+        # key that matches an existing key.
         class MyStr(str): pass
         class MyClass: pass
         obj = MyClass()
         obj.attr = 1
         obj.__dict__[MyStr('attr')] = 2
         self.assertEqual(obj.attr, 2)
 
+    def test_split_table_insert_with_str_subclass(self):
+        # gh-143189: inserting into split table dictionary with a non str
+        # key that matches an existing key in the shared table but not in
+        # the dict yet.
+
+        class MyStr(str): pass
+        class MyClass: pass
+
+        obj = MyClass()
+        obj.attr1 = 1
+
+        obj2 = MyClass()
+        d = obj2.__dict__
+        d[MyStr("attr1")] = 2
+        assert isinstance(list(d)[0], MyStr)
+
 
 class CAPITest(unittest.TestCase):
 

Objects/dictobject.c

@@ -1877,7 +1877,7 @@ static int
 insertdict(PyDictObject *mp,
            PyObject *key, Py_hash_t hash, PyObject *value)
 {
-    PyObject *old_value;
+    PyObject *old_value = NULL;
     Py_ssize_t ix;
 
     ASSERT_DICT_LOCKED(mp);
@@ -1898,11 +1898,14 @@ insertdict(PyDictObject *mp,
             goto Fail;
     }
 
-    if (ix == DKIX_EMPTY) {
+    if (old_value == NULL) {
         // insert_combined_dict() will convert from non DICT_KEYS_GENERAL table
         // into DICT_KEYS_GENERAL table if key is not Unicode.
         // We don't convert it before _Py_dict_lookup because non-Unicode key
         // may change generic table into Unicode table.
+        //
+        // NOTE: ix may not be DKIX_EMPTY because split table may have key
+        // without value.
         if (insert_combined_dict(mp, hash, key, value) < 0) {
             goto Fail;
         }