Delete entry in WebFrame->RenderFrameProxy map at WebFrame::close time.

content/renderer/render_frame_impl.cc

Index: content/renderer/render_frame_impl.cc
diff --git a/content/renderer/render_frame_impl.cc b/content/renderer/render_frame_impl.cc
index ad515e04ec88f74915717fe28144c4a654ecfeec..0a8692df74dfe603d4836f24f4f2f826cf60daec 100644
--- a/content/renderer/render_frame_impl.cc
+++ b/content/renderer/render_frame_impl.cc
@@ -730,8 +730,10 @@ RenderFrameImpl::~RenderFrameImpl() {
     render_view_->UnregisterVideoHoleFrame(this);
 #endif
 
-  if (render_frame_proxy_)
-    delete render_frame_proxy_;
+  // RenderFrameProxy is only "owned" by RenderFrameImpl in the case it is the
+  // main frame. Ensure it is deleted along with this object.
+  if (!is_subframe_ && render_frame_proxy_)

[ncarter (slow), 2015/05/07 23:33:36]

I was initially worried that this frameDetached call would trigger a new Detach
IPC, but it seems it's safe: the proxies detached this way are never subframes,
so they'll never have parents, so we won't reach the detach IPC.

So I think we're okay.

[nasko, 2015/05/07 23:46:05]

I did at the subframe check exactly to guard against that. It also shouldn't
have ever been called for subframes.

+    render_frame_proxy_->frameDetached();

[ncarter (slow), 2015/05/07 23:33:36]

Gosh it's kinda gross that the rfph dtor calls back into this class
(render_frame->set_render_frame_proxy(null)) before returning. But, I suppose
hard to avoid when you have circular references.

Might be worth adding a comment like this:

render_frame_proxy_->frameDetached();  // deletes/clears |render_frame_proxy_|

[nasko, 2015/05/07 23:46:05]

I'm hopeful that the CL I'm working on is going to allow us to get rid of this.
I've added a comment in the meantime :(.

 
   render_view_->UnregisterRenderFrame(this);
   g_routing_id_frame_map.Get().erase(routing_id_);