[Python-checkins] cpython (merge 3.3 -> default): Issue #16602: When a weakref's target was part of a long deallocation chain,

[antoine.pitrou]

http://hg.python.org/cpython/rev/17e5acad302e
changeset:   80761:17e5acad302e
parent:      80757:9e4b003a4d7a
parent:      80760:259c1636c884
user:        Antoine Pitrou <solipsis at pitrou.net>
date:        Sat Dec 08 21:18:50 2012 +0100
summary:
  Issue #16602: When a weakref's target was part of a long deallocation chain, the object could remain reachable through its weakref even though its refcount had dropped to zero.

Thanks to Eugene Toder for diagnosing and reporting the issue.

files:
  Include/weakrefobject.h  |  12 +++++++++++-
  Lib/test/test_weakref.py |  21 +++++++++++++++++++++
  Misc/NEWS                |   4 ++++
  Objects/weakrefobject.c  |   5 ++---
  4 files changed, 38 insertions(+), 4 deletions(-)


diff --git a/Include/weakrefobject.h b/Include/weakrefobject.h
--- a/Include/weakrefobject.h
+++ b/Include/weakrefobject.h
@@ -70,7 +70,17 @@
 PyAPI_FUNC(void) _PyWeakref_ClearRef(PyWeakReference *self);
 #endif
 
-#define PyWeakref_GET_OBJECT(ref) (((PyWeakReference *)(ref))->wr_object)
+/* Explanation for the Py_REFCNT() check: when a weakref's target is part
+   of a long chain of deallocations which triggers the trashcan mechanism,
+   clearing the weakrefs can be delayed long after the target's refcount
+   has dropped to zero.  In the meantime, code accessing the weakref will
+   be able to "see" the target object even though it is supposed to be
+   unreachable.  See issue #16602. */
+
+#define PyWeakref_GET_OBJECT(ref)                           \
+    (Py_REFCNT(((PyWeakReference *)(ref))->wr_object) > 0   \
+     ? ((PyWeakReference *)(ref))->wr_object                \
+     : Py_None)
 
 
 #ifdef __cplusplus
diff --git a/Lib/test/test_weakref.py b/Lib/test/test_weakref.py
--- a/Lib/test/test_weakref.py
+++ b/Lib/test/test_weakref.py
@@ -781,6 +781,27 @@
         self.assertEqual(hash(a), hash(42))
         self.assertRaises(TypeError, hash, b)
 
+    def test_trashcan_16602(self):
+        # Issue #16602: when a weakref's target was part of a long
+        # deallocation chain, the trashcan mechanism could delay clearing
+        # of the weakref and make the target object visible from outside
+        # code even though its refcount had dropped to 0.  A crash ensued.
+        class C:
+            def __init__(self, parent):
+                if not parent:
+                    return
+                wself = weakref.ref(self)
+                def cb(wparent):
+                    o = wself()
+                self.wparent = weakref.ref(parent, cb)
+
+        d = weakref.WeakKeyDictionary()
+        root = c = C(None)
+        for n in range(100):
+            d[c] = c = C(c)
+        del root
+        gc.collect()
+
 
 class SubclassableWeakrefTestCase(TestBase):
 
diff --git a/Misc/NEWS b/Misc/NEWS
--- a/Misc/NEWS
+++ b/Misc/NEWS
@@ -10,6 +10,10 @@
 Core and Builtins
 -----------------
 
+- Issue #16602: When a weakref's target was part of a long deallocation
+  chain, the object could remain reachable through its weakref even though
+  its refcount had dropped to zero.
+
 - Issue #16495: Remove extraneous NULL encoding check from bytes_decode().
 
 - Issue #16619: Create NameConstant AST class to represent None, True, and False
diff --git a/Objects/weakrefobject.c b/Objects/weakrefobject.c
--- a/Objects/weakrefobject.c
+++ b/Objects/weakrefobject.c
@@ -52,9 +52,8 @@
 {
     PyObject *callback = self->wr_callback;
 
-    if (PyWeakref_GET_OBJECT(self) != Py_None) {
-        PyWeakReference **list = GET_WEAKREFS_LISTPTR(
-            PyWeakref_GET_OBJECT(self));
+    if (self->wr_object != Py_None) {
+        PyWeakReference **list = GET_WEAKREFS_LISTPTR(self->wr_object);
 
         if (*list == self)
             /* If 'self' is the end of the list (and thus self->wr_next == NULL)

-- 
Repository URL: http://hg.python.org/cpython