gh-139103: Use borrowed references for positional args in _PyStack_UnpackDict (gh-144407)

The positional arguments passed to _PyStack_UnpackDict are already
kept alive by the caller, so we can avoid the extra reference count
operations by using borrowed references instead of creating new ones.

This reduces reference count contention in the free-threaded build
when calling functions with keyword arguments. In particular, this
avoids contention on the type argument to `__new__` when instantiating
namedtuples with keyword arguments.

Author: colesbury

Objects/call.c

@@ -935,6 +935,10 @@ _PyStack_AsDict(PyObject *const *values, PyObject *kwnames)
 
    The newly allocated argument vector supports PY_VECTORCALL_ARGUMENTS_OFFSET.
 
+   The positional arguments are borrowed references from the input array
+   (which must be kept alive by the caller). The keyword argument values
+   are new references.
+
    When done, you must call _PyStack_UnpackDict_Free(stack, nargs, kwnames) */
 PyObject *const *
 _PyStack_UnpackDict(PyThreadState *tstate,
@@ -970,9 +974,9 @@ _PyStack_UnpackDict(PyThreadState *tstate,
 
     stack++;  /* For PY_VECTORCALL_ARGUMENTS_OFFSET */
 
-    /* Copy positional arguments */
+    /* Copy positional arguments (borrowed references) */
     for (Py_ssize_t i = 0; i < nargs; i++) {
-        stack[i] = Py_NewRef(args[i]);
+        stack[i] = args[i];
     }
 
     PyObject **kwstack = stack + nargs;
@@ -1009,9 +1013,10 @@ void
 _PyStack_UnpackDict_Free(PyObject *const *stack, Py_ssize_t nargs,
                          PyObject *kwnames)
 {
-    Py_ssize_t n = PyTuple_GET_SIZE(kwnames) + nargs;
-    for (Py_ssize_t i = 0; i < n; i++) {
-        Py_DECREF(stack[i]);
+    /* Only decref kwargs values, positional args are borrowed */
+    Py_ssize_t nkwargs = PyTuple_GET_SIZE(kwnames);
+    for (Py_ssize_t i = 0; i < nkwargs; i++) {
+        Py_DECREF(stack[nargs + i]);
     }
     _PyStack_UnpackDict_FreeNoDecRef(stack, kwnames);
 }

Python/ceval.c

@@ -2000,11 +2000,16 @@ _PyEvalFramePushAndInit_Ex(PyThreadState *tstate, _PyStackRef func,
             PyStackRef_CLOSE(func);
             goto error;
         }
-        size_t total_args = nargs + PyDict_GET_SIZE(kwargs);
+        size_t nkwargs = PyDict_GET_SIZE(kwargs);
         assert(sizeof(PyObject *) == sizeof(_PyStackRef));
         newargs = (_PyStackRef *)object_array;
-        for (size_t i = 0; i < total_args; i++) {
-            newargs[i] = PyStackRef_FromPyObjectSteal(object_array[i]);
+        /* Positional args are borrowed from callargs tuple, need new reference */
+        for (Py_ssize_t i = 0; i < nargs; i++) {
+            newargs[i] = PyStackRef_FromPyObjectNew(object_array[i]);
+        }
+        /* Keyword args are owned by _PyStack_UnpackDict, steal them */
+        for (size_t i = 0; i < nkwargs; i++) {
+            newargs[nargs + i] = PyStackRef_FromPyObjectSteal(object_array[nargs + i]);
         }
     }
     else {