Fix resource loads dropped when iframe is transferred to a new window and original window closes.

Fix resource loads dropped when iframe is transferred to a new window and original window closes.
BUG=55200
TEST=MagicIframeBrowserTest.TransferIframeCloseWindow

[Dmitry Titov, 2011-08-09 00:04:31]

The issue here is when iframe gets reparented to another page and the original
page closes, the route_id associated with the old page is not valid anymore, and
also all the pending requests for this route_id are forcefully closed.

The proposed fix is to catch the WebCore notification that tells us about every
request that is ongoing inside the reparented iframe, and send a message to
browser to update the route_id of the pending request with the route_id of the
new RenderView associated with the new page.

The WebKit side of this fix: https://bugs.webkit.org/show_bug.cgi?id=65890

[jam, 2011-08-09 02:49:04]

With this change, the chrome code would have 3 ids for a resource
(routing/request/identifier), which might be a bit confusing.

Can the webkit code update this through WebKit::WebURLLoader?

[Dmitry Titov, 2011-08-09 05:45:38]

On 2011/08/09 02:49:04, John Abd-El-Malek wrote:
> With this change, the chrome code would have 3 ids for a resource
> (routing/request/identifier), which might be a bit confusing.
> 
> Can the webkit code update this through WebKit::WebURLLoader?

I'm not sure I understand. Could you please explain it more?

[jam, 2011-08-09 15:53:45]

On 2011/08/09 05:45:38, Dmitry Titov wrote:
> On 2011/08/09 02:49:04, John Abd-El-Malek wrote:
> > With this change, the chrome code would have 3 ids for a resource
> > (routing/request/identifier), which might be a bit confusing.
> > 
> > Can the webkit code update this through WebKit::WebURLLoader?
> 
> I'm not sure I understand. Could you please explain it more?

What I mean is that ResourceDispatcher now knows about routing_id and
request_id. It would be preferable to not make it know about yet another ID,
which looks to me like it's an internal webkit id.

If the request is transferred through WebFrameClient, as it is now, then the
only way to do that is to pass webkit's identifier as well and store it
initially, so it can find the request. My question is if we can avoid this by
having the webkit code call a method on WebURLLoader instead. That way, we
wouldn't need to store the webkit identifier of each request in chrome code.

[jam, 2011-08-09 19:07:15]

another option, which Darin recommended, is to avoid doing this per request.
that would avoid an IPC per request, and instead do it once per magic iframe
move. if an iframe has dozens of pending XHR, this would be much more efficient.
RDH can then move all the requests over, and this would cover plugins/html5
video etc.

[Dmitry Titov, 2011-08-10 17:54:29]

Thanks for feedback! I will close this patch, there is a different idea on how
to do it (from Darin), I'll give it a try.