gh-139757: Add BINARY_OP_SUBSCR_USTR_INT

Lib/test/test_capi/test_opt.py

@@ -1904,6 +1904,26 @@ def testfunc(n):
         self.assertNotIn("_GUARD_TOS_UNICODE", uops)
         self.assertIn("_BINARY_OP_ADD_UNICODE", uops)
 
+    def test_binary_subcsr_ustr_int_narrows_to_str(self):
+        def testfunc(n):
+            x = []
+            s = "바이트코f드_특수화"
+            for _ in range(n):
+                y = s[4]       # _BINARY_OP_SUBSCR_USTR_INT
+                z = "bar" + y  # (_GUARD_TOS_UNICODE) + _BINARY_OP_ADD_UNICODE
+                x.append(z)
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, ["barf"] * TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertIn("_BINARY_OP_SUBSCR_USTR_INT", uops)
+        # _BINARY_OP_SUBSCR_USTR_INT narrows the result to 'str' so
+        # the unicode guard before _BINARY_OP_ADD_UNICODE is removed.
+        self.assertNotIn("_GUARD_TOS_UNICODE", uops)
+        self.assertIn("_BINARY_OP_ADD_UNICODE", uops)
+
     def test_binary_op_subscr_str_int(self):
         def testfunc(n):
             x = 0
@@ -1924,6 +1944,26 @@ def testfunc(n):
         self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
         self.assertLessEqual(count_ops(ex, "_POP_TOP_INT"), 1)
 
+    def test_binary_op_subscr_ustr_int(self):
+        def testfunc(n):
+            x = 0
+            s = "hello바"
+            for _ in range(n):
+                c = s[1]  # _BINARY_OP_SUBSCR_USTR_INT
+                if c == 'e':
+                    x += 1
+            return x
+
+        res, ex = self._run_with_optimizer(testfunc, TIER2_THRESHOLD)
+        self.assertEqual(res, TIER2_THRESHOLD)
+        self.assertIsNotNone(ex)
+        uops = get_opnames(ex)
+        self.assertIn("_BINARY_OP_SUBSCR_USTR_INT", uops)
+        self.assertIn("_COMPARE_OP_STR", uops)
+        self.assertIn("_POP_TOP_NOP", uops)
+        self.assertLessEqual(count_ops(ex, "_POP_TOP"), 2)
+        self.assertLessEqual(count_ops(ex, "_POP_TOP_INT"), 1)
+
     def test_call_type_1_guards_removed(self):
         def testfunc(n):
             x = 0

Lib/test/test_opcache.py

@@ -1792,6 +1792,7 @@ def binary_subscr_str_int_non_compact():
                     self.assertEqual(a[idx], expected)
 
         binary_subscr_str_int_non_compact()
+        self.assert_specialized(binary_subscr_str_int_non_compact, "BINARY_OP_SUBSCR_USTR_INT")
         self.assert_no_opcode(binary_subscr_str_int_non_compact, "BINARY_OP_SUBSCR_STR_INT")
 
         def binary_subscr_getitems():

Misc/NEWS.d/next/Core_and_Builtins/2026-01-04-11-08-20.gh-issue-139757.AR6LG0.rst

@@ -0,0 +1,2 @@
+Add ``BINARY_OP_SUBSCR_USTR_INT`` to specialize reading an ASCII character
+from any string. Patch by Chris Eibl.

Python/bytecodes.c

@@ -594,7 +594,7 @@ dummy_func(
             BINARY_OP_SUBSCR_LIST_SLICE,
             BINARY_OP_SUBSCR_TUPLE_INT,
             BINARY_OP_SUBSCR_STR_INT,
-            BINARY_OP_SUBSCR_NCSTR_INT,
+            BINARY_OP_SUBSCR_USTR_INT,
             BINARY_OP_SUBSCR_DICT,
             BINARY_OP_SUBSCR_GETITEM,
             BINARY_OP_INPLACE_ADD_UNICODE,
@@ -943,15 +943,15 @@ dummy_func(
         }
 
         macro(BINARY_OP_SUBSCR_STR_INT) =
-            _GUARD_TOS_INT + _GUARD_NOS_COMPACT_ASCII + unused/5 + _BINARY_OP_SUBSCR_STR_INT + _POP_TOP_INT + POP_TOP;
+            _GUARD_TOS_INT + _GUARD_NOS_COMPACT_ASCII + unused/5 + _BINARY_OP_SUBSCR_STR_INT + _POP_TOP_INT + _POP_TOP_UNICODE;
 
         op(_BINARY_OP_SUBSCR_STR_INT, (str_st, sub_st -- res, s, i)) {
             PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
             PyObject *str = PyStackRef_AsPyObjectBorrow(str_st);
 
             assert(PyLong_CheckExact(sub));
             assert(PyUnicode_CheckExact(str));
-            DEOPT_IF(!_PyLong_IsNonNegativeCompact((PyLongObject *)sub));
+            assert(_PyLong_IsNonNegativeCompact((PyLongObject *)sub));
             Py_ssize_t index = ((PyLongObject*)sub)->long_value.ob_digit[0];
             DEOPT_IF(PyUnicode_GET_LENGTH(str) <= index);
             uint8_t c = PyUnicode_1BYTE_DATA(str)[index];
@@ -964,16 +964,16 @@ dummy_func(
             res = PyStackRef_FromPyObjectBorrow(res_o);
         }
 
-        macro(BINARY_OP_SUBSCR_NCSTR_INT) =
-            _GUARD_TOS_INT + _GUARD_NOS_UNICODE + unused/5 + _BINARY_OP_SUBSCR_NCSTR_INT + _POP_TOP_INT + POP_TOP;
+        macro(BINARY_OP_SUBSCR_USTR_INT) =
+            _GUARD_TOS_INT + _GUARD_NOS_UNICODE + unused/5 + _BINARY_OP_SUBSCR_USTR_INT + _POP_TOP_INT + _POP_TOP_UNICODE;
 
-        op(_BINARY_OP_SUBSCR_NCSTR_INT, (str_st, sub_st -- res, s, i)) {
+        op(_BINARY_OP_SUBSCR_USTR_INT, (str_st, sub_st -- res, s, i)) {
             PyObject *sub = PyStackRef_AsPyObjectBorrow(sub_st);
             PyObject *str = PyStackRef_AsPyObjectBorrow(str_st);
 
             assert(PyLong_CheckExact(sub));
             assert(PyUnicode_CheckExact(str));
-            DEOPT_IF(!_PyLong_IsNonNegativeCompact((PyLongObject *)sub));
+            assert(_PyLong_IsNonNegativeCompact((PyLongObject *)sub));
             Py_ssize_t index = ((PyLongObject*)sub)->long_value.ob_digit[0];
             DEOPT_IF(PyUnicode_GET_LENGTH(str) <= index);
             // Specialize for reading an ASCII character from any string: