bpo-44531: Add _PyType_AllocNoTrack() function (GH-26947)

Add an internal _PyType_AllocNoTrack() function to allocate an object
without tracking it in the GC.

Modify dict_new() to use _PyType_AllocNoTrack(): dict subclasses are
now only tracked once all PyDictObject members are initialized.
Calling _PyObject_GC_UNTRACK() is no longer needed for the dict type.

Similar change in tuple_subtype_new() for tuple subclasses.

Replace tuple_gc_track() with _PyObject_GC_TRACK().

Include/internal/pycore_object.h

@@ -178,6 +178,8 @@ extern int _Py_CheckSlotResult(
 // See also the Py_TPFLAGS_READY flag.
 #define _PyType_IsReady(type) ((type)->tp_dict != NULL)
 
+extern PyObject* _PyType_AllocNoTrack(PyTypeObject *type, Py_ssize_t nitems);
+
 #ifdef __cplusplus
 }
 #endif

Objects/dictobject.c

@@ -3324,19 +3324,16 @@ static PyNumberMethods dict_as_number = {
 static PyObject *
 dict_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
 {
-    PyObject *self;
-    PyDictObject *d;
+    assert(type != NULL);
+    assert(type->tp_alloc != NULL);
+    // dict subclasses must implement the GC protocol
+    assert(_PyType_IS_GC(type));
 
-    assert(type != NULL && type->tp_alloc != NULL);
-    self = type->tp_alloc(type, 0);
-    if (self == NULL)
+    PyObject *self = type->tp_alloc(type, 0);
+    if (self == NULL) {
         return NULL;
-    d = (PyDictObject *)self;
-
-    /* The object has been implicitly tracked by tp_alloc */
-    if (type == &PyDict_Type) {
-        _PyObject_GC_UNTRACK(d);
     }
+    PyDictObject *d = (PyDictObject *)self;
 
     d->ma_used = 0;
     d->ma_version_tag = DICT_NEXT_VERSION();
@@ -3344,6 +3341,17 @@ dict_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
     d->ma_keys = Py_EMPTY_KEYS;
     d->ma_values = empty_values;
     ASSERT_CONSISTENT(d);
+
+    if (type != &PyDict_Type) {
+        // Don't track if a subclass tp_alloc is PyType_GenericAlloc()
+        if (!_PyObject_GC_IS_TRACKED(d)) {
+            _PyObject_GC_TRACK(d);
+        }
+    }
+    else {
+        // _PyType_AllocNoTrack() does not track the created object
+        assert(!_PyObject_GC_IS_TRACKED(d));
+    }
     return self;
 }
 
@@ -3441,7 +3449,7 @@ PyTypeObject PyDict_Type = {
     0,                                          /* tp_descr_set */
     0,                                          /* tp_dictoffset */
     dict_init,                                  /* tp_init */
-    PyType_GenericAlloc,                        /* tp_alloc */
+    _PyType_AllocNoTrack,                       /* tp_alloc */
     dict_new,                                   /* tp_new */
     PyObject_GC_Del,                            /* tp_free */
     .tp_vectorcall = dict_vectorcall,

Objects/tupleobject.c

@@ -25,13 +25,6 @@ get_tuple_state(void)
 #endif
 
 
-static inline void
-tuple_gc_track(PyTupleObject *op)
-{
-    _PyObject_GC_TRACK(op);
-}
-
-
 /* Print summary info about the state of the optimized allocator */
 void
 _PyTuple_DebugMallocStats(FILE *out)
@@ -48,10 +41,12 @@ _PyTuple_DebugMallocStats(FILE *out)
 #endif
 }
 
-/* Allocate an uninitialized tuple object. Before making it public following
+/* Allocate an uninitialized tuple object. Before making it public, following
    steps must be done:
-   - initialize its items
-   - call tuple_gc_track() on it
+
+   - Initialize its items.
+   - Call _PyObject_GC_TRACK() on it.
+
    Because the empty tuple is always reused and it's already tracked by GC,
    this function must not be called with size == 0 (unless from PyTuple_New()
    which wraps this function).
@@ -161,7 +156,7 @@ PyTuple_New(Py_ssize_t size)
     for (Py_ssize_t i = 0; i < size; i++) {
         op->ob_item[i] = NULL;
     }
-    tuple_gc_track(op);
+    _PyObject_GC_TRACK(op);
     return (PyObject *) op;
 }
 
@@ -257,7 +252,7 @@ PyTuple_Pack(Py_ssize_t n, ...)
         items[i] = o;
     }
     va_end(vargs);
-    tuple_gc_track(result);
+    _PyObject_GC_TRACK(result);
     return (PyObject *)result;
 }
 
@@ -473,7 +468,7 @@ _PyTuple_FromArray(PyObject *const *src, Py_ssize_t n)
         Py_INCREF(item);
         dst[i] = item;
     }
-    tuple_gc_track(tuple);
+    _PyObject_GC_TRACK(tuple);
     return (PyObject *)tuple;
 }
 
@@ -551,7 +546,7 @@ tupleconcat(PyTupleObject *a, PyObject *bb)
         Py_INCREF(v);
         dest[i] = v;
     }
-    tuple_gc_track(np);
+    _PyObject_GC_TRACK(np);
     return (PyObject *)np;
 }
 
@@ -588,7 +583,7 @@ tuplerepeat(PyTupleObject *a, Py_ssize_t n)
             p++;
         }
     }
-    tuple_gc_track(np);
+    _PyObject_GC_TRACK(np);
     return (PyObject *) np;
 }
 
@@ -783,6 +778,9 @@ tuple_subtype_new(PyTypeObject *type, PyObject *iterable)
     Py_ssize_t i, n;
 
     assert(PyType_IsSubtype(type, &PyTuple_Type));
+    // tuple subclasses must implement the GC protocol
+    assert(_PyType_IS_GC(type));
+
     tmp = tuple_new_impl(&PyTuple_Type, iterable);
     if (tmp == NULL)
         return NULL;
@@ -798,6 +796,11 @@ tuple_subtype_new(PyTypeObject *type, PyObject *iterable)
         PyTuple_SET_ITEM(newobj, i, item);
     }
     Py_DECREF(tmp);
+
+    // Don't track if a subclass tp_alloc is PyType_GenericAlloc()
+    if (!_PyObject_GC_IS_TRACKED(newobj)) {
+        _PyObject_GC_TRACK(newobj);
+    }
     return newobj;
 }
 
@@ -857,7 +860,7 @@ tuplesubscript(PyTupleObject* self, PyObject* item)
                 dest[i] = it;
             }
 
-            tuple_gc_track(result);
+            _PyObject_GC_TRACK(result);
             return (PyObject *)result;
         }
     }

Objects/typeobject.c

@@ -1164,7 +1164,7 @@ type_call(PyTypeObject *type, PyObject *args, PyObject *kwds)
 }
 
 PyObject *
-PyType_GenericAlloc(PyTypeObject *type, Py_ssize_t nitems)
+_PyType_AllocNoTrack(PyTypeObject *type, Py_ssize_t nitems)
 {
     PyObject *obj;
     const size_t size = _PyObject_VAR_SIZE(type, nitems+1);
@@ -1189,6 +1189,16 @@ PyType_GenericAlloc(PyTypeObject *type, Py_ssize_t nitems)
     else {
         _PyObject_InitVar((PyVarObject *)obj, type, nitems);
     }
+    return obj;
+}
+
+PyObject *
+PyType_GenericAlloc(PyTypeObject *type, Py_ssize_t nitems)
+{
+    PyObject *obj = _PyType_AllocNoTrack(type, nitems);
+    if (obj == NULL) {
+        return NULL;
+    }
 
     if (_PyType_IS_GC(type)) {
         _PyObject_GC_TRACK(obj);