Skip stack trace formatting in case the global object is already dead.

Skip stack trace formatting in case the global object is already dead.

This fixes assertion failures when running webkit layout tests.

Also fixes issue 2500.

R=mstarzinger@chromium.org
BUG=v8:2500

[Michael Starzinger, 2013-01-16 13:35:53]

https://chromiumcodereview.appspot.com/11971015/diff/1/src/heap.cc
File src/heap.cc (right):

https://chromiumcodereview.appspot.com/11971015/diff/1/src/heap.cc#newcode7327
src/heap.cc:7327: if
(!isolate->global_proxy()->GetPrototype()->IsJSGlobalProxy()) return;
That is quite a hack that is going on here. I think a better approach would be
for %GetDataProperty to actually check  whether the receiver is a JSGlobalProxy,
skip to the JSGlobalObject and if that is not present return undefined (as it's
implementing safe semantics that should gracefully fail).

[Yang, 2013-01-17 13:57:32]

On 2013/01/16 13:35:53, Michael Starzinger wrote:
> https://chromiumcodereview.appspot.com/11971015/diff/1/src/heap.cc
> File src/heap.cc (right):
> 
> https://chromiumcodereview.appspot.com/11971015/diff/1/src/heap.cc#newcode7327
> src/heap.cc:7327: if
> (!isolate->global_proxy()->GetPrototype()->IsJSGlobalProxy()) return;
> That is quite a hack that is going on here. I think a better approach would be
> for %GetDataProperty to actually check  whether the receiver is a
JSGlobalProxy,
> skip to the JSGlobalObject and if that is not present return undefined (as
it's
> implementing safe semantics that should gracefully fail).

Please take another look.

This also includes the fix for v8:2500.

Patch set 3 disables deferred stack trace formatting for now to not risking
polluting the next roll.

[Yang, 2013-01-17 14:09:48]

On 2013/01/17 13:57:32, Yang wrote:
> On 2013/01/16 13:35:53, Michael Starzinger wrote:
> > https://chromiumcodereview.appspot.com/11971015/diff/1/src/heap.cc
> > File src/heap.cc (right):
> > 
> >
https://chromiumcodereview.appspot.com/11971015/diff/1/src/heap.cc#newcode7327
> > src/heap.cc:7327: if
> > (!isolate->global_proxy()->GetPrototype()->IsJSGlobalProxy()) return;
> > That is quite a hack that is going on here. I think a better approach would
be
> > for %GetDataProperty to actually check  whether the receiver is a
> JSGlobalProxy,
> > skip to the JSGlobalObject and if that is not present return undefined (as
> it's
> > implementing safe semantics that should gracefully fail).
> 
> Please take another look.
> 
> This also includes the fix for v8:2500.
> 
> Patch set 3 disables deferred stack trace formatting for now to not risking
> polluting the next roll.

Patch set 5 actually also changes the test expectations necessary due to the
temporary disable...