This is a cleaned up fix of Christian's original patch in ...

This is a cleaned up fix of Christian's original patch in 
http://codereview.chromium.org/13176

I cleaned it a bit so it does not leak memory. There is a corner case that
can crash a test, so I have to make a workaround.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=6472

[Mads Ager (chromium), 2008-12-06 19:03:36]

LGTM, thanks!

http://codereview.chromium.org/13224/diff/202/205
File webkit/port/bindings/v8/v8_proxy.cpp (right):

http://codereview.chromium.org/13224/diff/202/205#newcode2478
Line 2478: function = GetTemplate(desc_type)->GetFunction();
When do we get in the situation that we cannot use GetConstructor?  I guess this
means that you can get to the resulting DOM wrapper's constructor, which will be
a DOM wrapper constructor function which will actually have Function.prototype
as it's prototype.  It seems unlikely that this will matter since accessing the
constructor functions directly by name will always set it up correctly, but it
would still be nice if we could set the prototype to Object.prototype in this
case as well.

[Feng Qian, 2008-12-08 05:49:40]

http://codereview.chromium.org/13224/diff/202/205
File webkit/port/bindings/v8/v8_proxy.cpp (right):

http://codereview.chromium.org/13224/diff/202/205#newcode2478
Line 2478: function = GetTemplate(desc_type)->GetFunction();
The problem is that it may not get proxy because the Frame is in teardown mode,
and JS is disabled.

We should stick m_dom_constructor_cache and m_object_prototype to the context
instead of on V8Proxy, then we can avoid this problem.

This is a quick workaround one layout test crash. I need to file a bug for this.

On 2008/12/06 19:03:36, Mads Ager wrote:
> When do we get in the situation that we cannot use GetConstructor?  I guess
this
> means that you can get to the resulting DOM wrapper's constructor, which will
be
> a DOM wrapper constructor function which will actually have Function.prototype
> as it's prototype.  It seems unlikely that this will matter since accessing
the
> constructor functions directly by name will always set it up correctly, but it
> would still be nice if we could set the prototype to Object.prototype in this
> case as well.