2011-03-14 Geoffrey Garen <ggaren@apple.com>
Reviewed by Oliver Hunt.
Made JSWrapperObject and subclasses moving-GC-safe
https://bugs.webkit.org/show_bug.cgi?id=56346
SunSpider reports no change.
* runtime/BooleanObject.cpp:
(JSC::BooleanObject::BooleanObject):
* runtime/DateInstance.cpp:
(JSC::DateInstance::DateInstance): No more need for JSGlobalData, since
we don't initialize the wrapped value in our constructor.
* runtime/DateInstance.h: Don't set the OverridesMarkChildren flag because
we do not in fact override markChildren.
* runtime/DatePrototype.h: Declare an anonymous slot, since wrapper object
no longer does so for us. Also added an ASSERT to catch a latent bug,
where DatePrototype stomped on its base class's anonymous slot. Hard-coded
anonymous slots are a plague on our code. This doesn't cause any problems
in our existing code since the base class never reads the anonymous slot
it declares, but it caused crashes when I tried to start using the slot
in an initial version of this patch.
* runtime/JSWrapperObject.h:
(JSC::JSWrapperObject::JSWrapperObject):
(JSC::JSWrapperObject::internalValue):
(JSC::JSWrapperObject::setInternalValue): Resolved a problem where
our internal value was stored in two places: an anonymous slot, and a
data member which was not always visited during GC. Now, we only use the
data member, and we always visit it. (Instead of relying on certain
subclasses to set the OverridesMarkChildren bit, we set it ourselves.)
* runtime/NumberObject.cpp:
(JSC::NumberObject::NumberObject): No more need for JSGlobalData, since
we don't initialize the wrapped value in our constructor.
* runtime/NumberObject.h: Removed meaningless declaration.
* runtime/StringObject.cpp:
(JSC::StringObject::StringObject): No more need for JSGlobalData, since
we don't initialize the wrapped value in our constructor.
* runtime/StringObject.h: Don't set the OverridesMarkChildren flag because
we do not in fact override markChildren.
* runtime/StringPrototype.h: Declare an anonymous slot, since wrapper object
no longer does so for us. Also added an ASSERT to catch a latent bug,
where DatePrototype stomped on its base class's anonymous slot. Hard-coded
anonymous slots are a plague on our code.
git-svn-id: svn://svn.chromium.org/blink/trunk@81086 bbb929c8-8fbe-4397-9dbb-9b2b20218538
Showing
Please register or sign in to comment