gh-127065: Make methodcaller thread-safe and re-entrant

Modules/_operator.c

@@ -1602,7 +1602,7 @@ typedef struct {
     vectorcallfunc vectorcall;
 } methodcallerobject;
 
-#ifndef Py_GIL_DISABLED
+
 static int _methodcaller_initialize_vectorcall(methodcallerobject* mc)
 {
     PyObject* args = mc->xargs;
@@ -1659,13 +1659,26 @@ methodcaller_vectorcall(
     }
 
     assert(mc->vectorcall_args != 0);
-    mc->vectorcall_args[0] = args[0];
+
+    Py_ssize_t number_of_arguments = PyTuple_GET_SIZE(mc->xargs) +
+        (mc->vectorcall_kwnames? PyTuple_GET_SIZE(mc->vectorcall_kwnames):0) + 1;
+    size_t buffer_size = sizeof(PyObject *) * (number_of_arguments + 1);
+
+    // for number_of_arguments == 1 we could optimize by setting tmp_args equal to &args
+    PyObject **tmp_args = (PyObject **) PyMem_Malloc(buffer_size);
+    if (tmp_args == NULL) {
+        PyErr_NoMemory();
+        return NULL;
+    }
+    memcpy(tmp_args, mc->vectorcall_args, buffer_size);
+    tmp_args[0] = args[0];
     return PyObject_VectorcallMethod(
-            mc->name, mc->vectorcall_args,
+            mc->name, tmp_args,
             (PyTuple_GET_SIZE(mc->xargs)) | PY_VECTORCALL_ARGUMENTS_OFFSET,
             mc->vectorcall_kwnames);
+
+    PyMem_Free(tmp_args);
 }
-#endif
 
 
 /* AC 3.5: variable number of arguments, not currently support by AC */
@@ -1704,15 +1717,7 @@ methodcaller_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
     mc->kwds = Py_XNewRef(kwds);
     mc->vectorcall_args = 0;
 
-
-#ifdef Py_GIL_DISABLED
-    // gh-127065: The current implementation of methodcaller_vectorcall
-    // is not thread-safe because it modifies the `vectorcall_args` array,
-    // which is shared across calls.
-    mc->vectorcall = NULL;
-#else
     mc->vectorcall = (vectorcallfunc)methodcaller_vectorcall;
-#endif
 
     PyObject_GC_Track(mc);
     return (PyObject *)mc;