Changeset 39672 in webkit for trunk/JavaScriptCore

Timestamp:

Jan 6, 2009, 11:24:18 PM (16 years ago)

Author:

[email protected]

Message:

Reviewed by Darin Adler.

​https://bugs.webkit.org/show_bug.cgi?id=23142
ThreadGlobalData leaks seen on buildbot

• wtf/ThreadSpecific.h: (WTF::ThreadSpecific::destroy): Temporarily reset the thread specific value to make getter work on Mac OS X.

• wtf/Platform.h: Touch this file again to make sure all Windows builds use the most recent version of ThreadSpecific.h.

Location:

trunk/JavaScriptCore

Files:

• ChangeLog (modified) (1 diff)
• wtf/Platform.h (modified) (1 diff)
• wtf/ThreadSpecific.h (modified) (1 diff)

trunk/JavaScriptCore/ChangeLog

@@ +1 @@
+2009-01-06  Alexey Proskuryakov  <[email protected]>
+
+        Reviewed by Darin Adler.
+
+        https://bugs.webkit.org/show_bug.cgi?id=23142
+        ThreadGlobalData leaks seen on buildbot
+
+        * wtf/ThreadSpecific.h: (WTF::ThreadSpecific::destroy): Temporarily reset the thread
+        specific value to make getter work on Mac OS X.
+
+        * wtf/Platform.h: Touch this file again to make sure all Windows builds use the most recent
+        version of ThreadSpecific.h.
+
 2009-01-05  Gavin Barraclough  <[email protected]>
 

trunk/JavaScriptCore/wtf/Platform.h

@@ -1 +1 @@
 /*
- * Copyright (C) 2006 Apple Inc.  All rights reserved.
+ * Copyright (C) 2006, 2007, 2008, 2009 Apple Inc.  All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without

trunk/JavaScriptCore/wtf/ThreadSpecific.h

@@ -98 +98 @@
 {
     Data* data = static_cast<Data*>(ptr);
+
+    // We want get() to keep working while data destructor works, because it can be called indirectly by the destructor.
+    // Some pthreads implementations zero out the pointer before calling destroy(), so we temporarily reset it.
+    pthread_setspecific(data->owner->m_key, ptr);
     data->value->~T();
     fastFree(data->value);
-    // Only reset the pointer after value destructor finishes - otherwise, code in destructor could trigger
-    // re-creation of the object.
     pthread_setspecific(data->owner->m_key, 0);
     delete data;