Address review

Author: Fidget-Spinner

Include/internal/pycore_magic_number.h

@@ -294,7 +294,6 @@ Known values:
     Python 3.15a4 3661 (Lazy imports IMPORT_NAME opcode changes)
     Python 3.15a8 3662 (Add counter to RESUME)
     Python 3.15a8 3663 (Merge GET_ITER and GET_YIELD_FROM_ITER. Modify SEND to make it a bit more like FOR_ITER)
-    Python 3.15b1 3664 (All code objects now have 1 extra stack slot space)
 
 
     Python 3.16 will start with 3700
@@ -308,7 +307,7 @@ PC/launcher.c must also be updated.
 
 */
 
-#define PYC_MAGIC_NUMBER 3664
+#define PYC_MAGIC_NUMBER 3663
 /* This is equivalent to converting PYC_MAGIC_NUMBER to 2 bytes
    (little-endian) and then appending b'\r\n'. */
 #define PYC_MAGIC_NUMBER_TOKEN \

Include/internal/pycore_opcode_metadata.h

@@ -2732,6 +2732,7 @@ def testfunc(n):
         uops = get_opnames(ex)
         self.assertIn("_CALL_BUILTIN_FAST", uops)
         self.assertNotIn("_GUARD_CALLABLE_BUILTIN_FAST", uops)
+        self.assertGreaterEqual(count_ops(ex, "_POP_TOP_NOP"), 6)
 
     def test_call_builtin_fast_with_keywords(self):
         def testfunc(n):

Lib/test/test_capi/test_opt.py

@@ -1686,9 +1686,9 @@ def func():
             return sys._getframe()
         x = func()
         if support.Py_GIL_DISABLED:
-            INTERPRETER_FRAME = '9PihcPP'
+            INTERPRETER_FRAME = '9PihcP'
         else:
-            INTERPRETER_FRAME = '9PhcPP'
+            INTERPRETER_FRAME = '9PhcP'
         check(x, size('3PiccPPP' + INTERPRETER_FRAME + 'P'))
         # function
         def func(): pass

Lib/test/test_sys.py

@@ -386,7 +386,6 @@ dummy_func(
             PyStackRef_CLOSE(nos);
         }
 
-        // This is expanded to POP_TOP + POP_TOP + POP_TOP + ... (oparg times) in the JIT trace.
         op(_POP_TOP_OPARG, (args[oparg] -- )) {
             _PyStackRef_CloseStack(args, oparg);
             DEAD(args);
@@ -4635,7 +4634,7 @@ dummy_func(
             EXIT_IF(PyCFunction_GET_FLAGS(callable_o) != METH_FASTCALL);
         }
 
-        op(_CALL_BUILTIN_FAST, (callable, self_or_null, args[oparg] -- res, c, s, a[oparg])) {
+        op(_CALL_BUILTIN_FAST, (callable, self_or_null, args[oparg] -- res, s, a[oparg])) {
             /* Builtin METH_FASTCALL functions, without keywords */
             int total_args = oparg;
             _PyStackRef *arguments = args;
@@ -4649,13 +4648,15 @@ dummy_func(
                 arguments,
                 total_args
             );
-            c = callable;
+            // To maintain a consistent view of the stack in the GC, callable's stack location
+            // must be overwritten before we close it.
+            _PyStackRef *callable_ptr = args - 2;
+            PyStackRef_XSETREF(*callable_ptr, PyStackRef_NULL);
+            DEAD(callable);
             s = self_or_null;
-            // Swap the first argument with the first empty slot after args to allows space for res.
-            // We are guaranteed by the bytecode compiler to have a spare stack slot.
-            args[oparg] = args[0];
             DEAD(self_or_null);
             DEAD(callable);
+            (void)a;
             DEAD(args);
             res = res_o == NULL ? PyStackRef_NULL : PyStackRef_FromPyObjectSteal(res_o);
         }
@@ -4672,7 +4673,6 @@ dummy_func(
             _CALL_BUILTIN_FAST +
             _POP_TOP_OPARG +
             POP_TOP +
-            POP_TOP +
             _ERROR_IF_TOS_NULL +
             _CHECK_PERIODIC_AT_END;
 

Modules/_testinternalcapi/test_cases.c.h

@@ -945,12 +945,6 @@ _PyJit_translate_single_bytecode_to_trace(
                     assert(next->op.code == STORE_FAST);
                     operand = next->op.arg;
                 }
-                else if (uop == _POP_TOP_OPARG) {
-                    for (int i = 0; i < oparg; i++) {
-                        ADD_TO_TRACE(_POP_TOP, 0, operand, target);
-                    }
-                    continue;
-                }
                 else if (_PyUop_Flags[uop] & HAS_RECORDS_VALUE_FLAG) {
                     PyObject *recorded_value = tracer->prev_state.recorded_value;
                     tracer->prev_state.recorded_value = NULL;

Python/bytecodes.c

@@ -410,6 +410,29 @@ lookup_attr(JitOptContext *ctx, _PyBloomFilter *dependencies, _PyUOpInstruction
     return sym_new_not_null(ctx);
 }
 
+static void
+optimize_pop_top(JitOptContext *ctx, _PyUOpInstruction *this_instr, JitOptRef value)
+{
+    PyTypeObject *typ = sym_get_type(value);
+    if (PyJitRef_IsBorrowed(value) ||
+        sym_is_immortal(PyJitRef_Unwrap(value)) ||
+        sym_is_null(value)) {
+        ADD_OP(_POP_TOP_NOP, 0, 0);
+    }
+    else if (typ == &PyLong_Type) {
+        ADD_OP(_POP_TOP_INT, 0, 0);
+    }
+    else if (typ == &PyFloat_Type) {
+        ADD_OP(_POP_TOP_FLOAT, 0, 0);
+    }
+    else if (typ == &PyUnicode_Type) {
+        ADD_OP(_POP_TOP_UNICODE, 0, 0);
+    }
+    else {
+        ADD_OP(_POP_TOP, 0, 0);
+    }
+}
+
 static
 PyCodeObject *
 get_current_code_object(JitOptContext *ctx)