GH-117760: Streamline the trashcan mechanism (GH-117763)

Include/cpython/object.h

@@ -448,8 +448,8 @@ without deallocating anything (and so unbounded call-stack depth is avoided).
 When the call stack finishes unwinding again, code generated by the END macro
 notices this, and calls another routine to deallocate all the objects that
 may have been added to the list of deferred deallocations.  In effect, a
-chain of N deallocations is broken into (N-1)/(_PyTrash_UNWIND_LEVEL-1) pieces,
+chain of N deallocations is broken into (N-1)/(Py_TRASHCAN_HEADROOM-1) pieces,
-with the call stack never exceeding a depth of _PyTrash_UNWIND_LEVEL.
+with the call stack never exceeding a depth of Py_TRASHCAN_HEADROOM.
 
 Since the tp_dealloc of a subclass typically calls the tp_dealloc of the base
 class, we need to ensure that the trashcan is only triggered on the tp_dealloc
@@ -461,30 +461,33 @@ passed as second argument to Py_TRASHCAN_BEGIN().
 /* Python 3.9 private API, invoked by the macros below. */
 PyAPI_FUNC(int) _PyTrash_begin(PyThreadState *tstate, PyObject *op);
 PyAPI_FUNC(void) _PyTrash_end(PyThreadState *tstate);
-/* Python 3.10 private API, invoked by the Py_TRASHCAN_BEGIN(). */
-PyAPI_FUNC(int) _PyTrash_cond(PyObject *op, destructor dealloc);
 
-#define Py_TRASHCAN_BEGIN_CONDITION(op, cond) \
+PyAPI_FUNC(void) _PyTrash_thread_deposit_object(PyThreadState *tstate, PyObject *op);
-    do { \
+PyAPI_FUNC(void) _PyTrash_thread_destroy_chain(PyThreadState *tstate);
-        PyThreadState *_tstate = NULL; \
+
-        /* If "cond" is false, then _tstate remains NULL and the deallocator \
+
-         * is run normally without involving the trashcan */ \
+/* Python 3.10 private API, invoked by the Py_TRASHCAN_BEGIN(). */
-        if (cond) { \
+
-            _tstate = PyThreadState_GetUnchecked(); \
+/* To avoid raising recursion errors during dealloc trigger trashcan before we reach
-            if (_PyTrash_begin(_tstate, _PyObject_CAST(op))) { \
+ * recursion limit. To avoid trashing, we don't attempt to empty the trashcan until
-                break; \
+ * we have headroom above the trigger limit */
-            } \
+#define Py_TRASHCAN_HEADROOM 50
-        }
-        /* The body of the deallocator is here. */
-#define Py_TRASHCAN_END \
-        if (_tstate) { \
-            _PyTrash_end(_tstate); \
-        } \
-    } while (0);
 
 #define Py_TRASHCAN_BEGIN(op, dealloc) \
-    Py_TRASHCAN_BEGIN_CONDITION((op), \
+do { \
-        _PyTrash_cond(_PyObject_CAST(op), (destructor)(dealloc)))
+    PyThreadState *tstate = PyThreadState_Get(); \
+    if (tstate->c_recursion_remaining <= Py_TRASHCAN_HEADROOM && Py_TYPE(op)->tp_dealloc == (destructor)dealloc) { \
+        _PyTrash_thread_deposit_object(tstate, (PyObject *)op); \
+        break; \
+    } \
+    tstate->c_recursion_remaining--;
+    /* The body of the deallocator is here. */
+#define Py_TRASHCAN_END \
+    tstate->c_recursion_remaining++; \
+    if (tstate->delete_later && tstate->c_recursion_remaining > (Py_TRASHCAN_HEADROOM*2)) { \
+        _PyTrash_thread_destroy_chain(tstate); \
+    } \
+} while (0);
 
 
 PyAPI_FUNC(void *) PyObject_GetItemData(PyObject *obj);

Include/cpython/pystate.h

@@ -56,11 +56,6 @@ typedef struct _stack_chunk {
     PyObject * data[1]; /* Variable sized */
 } _PyStackChunk;
 
-struct _py_trashcan {
-    int delete_nesting;
-    PyObject *delete_later;
-};
-
 struct _ts {
     /* See Python/ceval.c for comments explaining most fields */
 
@@ -152,7 +147,7 @@ struct _ts {
      */
     unsigned long native_thread_id;
 
-    struct _py_trashcan trash;
+    PyObject *delete_later;
 
     /* Tagged pointer to top-most critical section, or zero if there is no
      * active critical section. Critical sections are only used in

Include/internal/pycore_object.h

@@ -615,7 +615,6 @@ _PyObject_GET_WEAKREFS_LISTPTR_FROM_OFFSET(PyObject *op)
     return (PyWeakReference **)((char *)op + offset);
 }
 
-
 // Fast inlined version of PyObject_IS_GC()
 static inline int
 _PyObject_IS_GC(PyObject *obj)

Objects/object.c

@@ -2709,33 +2709,31 @@ finally:
 
 /* Trashcan support. */
 
-#define _PyTrash_UNWIND_LEVEL 50
-
 /* Add op to the gcstate->trash_delete_later list.  Called when the current
  * call-stack depth gets large.  op must be a currently untracked gc'ed
  * object, with refcount 0.  Py_DECREF must already have been called on it.
  */
-static void
+void
-_PyTrash_thread_deposit_object(struct _py_trashcan *trash, PyObject *op)
+_PyTrash_thread_deposit_object(PyThreadState *tstate, PyObject *op)
 {
     _PyObject_ASSERT(op, _PyObject_IS_GC(op));
     _PyObject_ASSERT(op, !_PyObject_GC_IS_TRACKED(op));
     _PyObject_ASSERT(op, Py_REFCNT(op) == 0);
 #ifdef Py_GIL_DISABLED
     _PyObject_ASSERT(op, op->ob_tid == 0);
-    op->ob_tid = (uintptr_t)trash->delete_later;
+    op->ob_tid = (uintptr_t)tstate->delete_later;
 #else
-    _PyGCHead_SET_PREV(_Py_AS_GC(op), (PyGC_Head*)trash->delete_later);
+    _PyGCHead_SET_PREV(_Py_AS_GC(op), (PyGC_Head*)tstate->delete_later);
 #endif
-    trash->delete_later = op;
+    tstate->delete_later = op;
 }
 
 /* Deallocate all the objects in the gcstate->trash_delete_later list.
  * Called when the call-stack unwinds again. */
-static void
+void
-_PyTrash_thread_destroy_chain(struct _py_trashcan *trash)
+_PyTrash_thread_destroy_chain(PyThreadState *tstate)
 {
-    /* We need to increase trash_delete_nesting here, otherwise,
+    /* We need to increase c_recursion_remaining here, otherwise,
        _PyTrash_thread_destroy_chain will be called recursively
        and then possibly crash.  An example that may crash without
        increase:
@@ -2746,17 +2744,17 @@ _PyTrash_thread_destroy_chain(struct _py_trashcan *trash)
                tups = [(tup,) for tup in tups]
            del tups
     */
-    assert(trash->delete_nesting == 0);
+    assert(tstate->c_recursion_remaining > Py_TRASHCAN_HEADROOM);
-    ++trash->delete_nesting;
+    tstate->c_recursion_remaining--;
-    while (trash->delete_later) {
+    while (tstate->delete_later) {
-        PyObject *op = trash->delete_later;
+        PyObject *op = tstate->delete_later;
         destructor dealloc = Py_TYPE(op)->tp_dealloc;
 
 #ifdef Py_GIL_DISABLED
-        trash->delete_later = (PyObject*) op->ob_tid;
+        tstate->delete_later = (PyObject*) op->ob_tid;
         op->ob_tid = 0;
 #else
-        trash->delete_later = (PyObject*) _PyGCHead_PREV(_Py_AS_GC(op));
+        tstate->delete_later = (PyObject*) _PyGCHead_PREV(_Py_AS_GC(op));
 #endif
 
         /* Call the deallocator directly.  This used to try to
@@ -2767,92 +2765,10 @@ _PyTrash_thread_destroy_chain(struct _py_trashcan *trash)
          */
         _PyObject_ASSERT(op, Py_REFCNT(op) == 0);
         (*dealloc)(op);
-        assert(trash->delete_nesting == 1);
     }
-    --trash->delete_nesting;
+    tstate->c_recursion_remaining++;
 }
 
-
-static struct _py_trashcan *
-_PyTrash_get_state(PyThreadState *tstate)
-{
-    if (tstate != NULL) {
-        return &tstate->trash;
-    }
-    // The current thread must be finalizing.
-    // Fall back to using thread-local state.
-    // XXX Use thread-local variable syntax?
-    assert(PyThread_tss_is_created(&_PyRuntime.trashTSSkey));
-    struct _py_trashcan *trash =
-        (struct _py_trashcan *)PyThread_tss_get(&_PyRuntime.trashTSSkey);
-    if (trash == NULL) {
-        trash = PyMem_RawMalloc(sizeof(struct _py_trashcan));
-        if (trash == NULL) {
-            Py_FatalError("Out of memory");
-        }
-        PyThread_tss_set(&_PyRuntime.trashTSSkey, (void *)trash);
-    }
-    return trash;
-}
-
-static void
-_PyTrash_clear_state(PyThreadState *tstate)
-{
-    if (tstate != NULL) {
-        assert(tstate->trash.delete_later == NULL);
-        return;
-    }
-    if (PyThread_tss_is_created(&_PyRuntime.trashTSSkey)) {
-        struct _py_trashcan *trash =
-            (struct _py_trashcan *)PyThread_tss_get(&_PyRuntime.trashTSSkey);
-        if (trash != NULL) {
-            PyThread_tss_set(&_PyRuntime.trashTSSkey, (void *)NULL);
-            PyMem_RawFree(trash);
-        }
-    }
-}
-
-
-int
-_PyTrash_begin(PyThreadState *tstate, PyObject *op)
-{
-    // XXX Make sure the GIL is held.
-    struct _py_trashcan *trash = _PyTrash_get_state(tstate);
-    if (trash->delete_nesting >= _PyTrash_UNWIND_LEVEL) {
-        /* Store the object (to be deallocated later) and jump past
-         * Py_TRASHCAN_END, skipping the body of the deallocator */
-        _PyTrash_thread_deposit_object(trash, op);
-        return 1;
-    }
-    ++trash->delete_nesting;
-    return 0;
-}
-
-
-void
-_PyTrash_end(PyThreadState *tstate)
-{
-    // XXX Make sure the GIL is held.
-    struct _py_trashcan *trash = _PyTrash_get_state(tstate);
-    --trash->delete_nesting;
-    if (trash->delete_nesting <= 0) {
-        if (trash->delete_later != NULL) {
-            _PyTrash_thread_destroy_chain(trash);
-        }
-        _PyTrash_clear_state(tstate);
-    }
-}
-
-
-/* bpo-40170: It's only be used in Py_TRASHCAN_BEGIN macro to hide
-   implementation details. */
-int
-_PyTrash_cond(PyObject *op, destructor dealloc)
-{
-    return Py_TYPE(op)->tp_dealloc == dealloc;
-}
-
-
 void _Py_NO_RETURN
 _PyObject_AssertFailed(PyObject *obj, const char *expr, const char *msg,
                        const char *file, int line, const char *function)

Python/pystate.c

@@ -1485,6 +1485,8 @@ init_threadstate(_PyThreadStateImpl *_tstate,
     tstate->what_event = -1;
     tstate->previous_executor = NULL;
 
+    tstate->delete_later = NULL;
+
     llist_init(&_tstate->mem_free_queue);
 
     if (interp->stoptheworld.requested || _PyRuntime.stoptheworld.requested) {