Draft: gh-139103: Use _PyObject_GetMethodStackRef for slot_tp_new

Misc/NEWS.d/next/Core_and_Builtins/2025-11-16-20-28-48.gh-issue-139103.gZNQMz.rst

@@ -0,0 +1 @@
+Improve performance of concurrent object construction by using `_PyObject_GetMethodStackRef` in `slot_tp_new`.

Objects/typeobject.c

@@ -10852,13 +10852,18 @@ slot_tp_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
     PyThreadState *tstate = _PyThreadState_GET();
     PyObject *func, *result;
 
-    func = PyObject_GetAttr((PyObject *)type, &_Py_ID(__new__));
-    if (func == NULL) {
+    _PyCStackRef c_stackref;
+    _PyThreadState_PushCStackRef(tstate, &c_stackref);
+    _PyObject_GetMethodStackRef(tstate, (PyObject *)type, &_Py_ID(__new__), &c_stackref.ref);
+
+    if (PyStackRef_IsNull(c_stackref.ref)) {
+        _PyThreadState_PopCStackRef(tstate, &c_stackref);
         return NULL;
     }
-
+    func = PyStackRef_AsPyObjectBorrow(c_stackref.ref);
+    assert(func != NULL);
     result = _PyObject_Call_Prepend(tstate, func, (PyObject *)type, args, kwds);
-    Py_DECREF(func);
+    _PyThreadState_PopCStackRef(tstate, &c_stackref);
     return result;
 }
 

Tools/ftscalingbench/ftscalingbench.py

@@ -28,6 +28,8 @@
 import threading
 import time
 from operator import methodcaller
+from typing import NamedTuple
+from collections import namedtuple
 
 # The iterations in individual benchmarks are scaled by this factor.
 WORK_SCALE = 100
@@ -38,11 +40,26 @@
 in_queues = []
 out_queues = []
 
-
 def register_benchmark(func):
     ALL_BENCHMARKS[func.__name__] = func
     return func
 
+class Foo(NamedTuple):
+    x: int
+
+
+Bar = namedtuple('Bar', ['x'])
+
+@register_benchmark
+def typing_namedtuple():
+    for i in range(1000 * WORK_SCALE):
+        _ = Foo(x=1)
+
+@register_benchmark
+def namedtuple():
+    for i in range(1000 * WORK_SCALE):
+        _ = Bar(x=1)
+
 @register_benchmark
 def object_cfunction():
     accu = 0