Move FrameMsg_NewFrame/NewFrameProxy to mojom

content/browser/renderer_host/render_process_host_impl.h

Index: content/browser/renderer_host/render_process_host_impl.h
diff --git a/content/browser/renderer_host/render_process_host_impl.h b/content/browser/renderer_host/render_process_host_impl.h
index e2326e0008c57d097f90f756274366825fb69685..305cda947cb9206025766ce62c1f355a90eda254 100644
--- a/content/browser/renderer_host/render_process_host_impl.h
+++ b/content/browser/renderer_host/render_process_host_impl.h
@@ -177,7 +177,8 @@ class CONTENT_EXPORT RenderProcessHostImpl
   void PurgeAndSuspend() override;
 
   mojom::RouteProvider* GetRemoteRouteProvider();
-  mojom::Renderer* GetRendererInterface();
+
+  static mojom::Renderer* GetRendererInterface(RenderProcessHost* host);

[ncarter (slow), 2016/10/03 22:45:09]

I'm okay with this approach. Putting a static method in RPHI that accepts an
RPH* seems to be an established pattern -- RegisterHost, IsSuitableHost, and a
few others work that way.

Really the existence of this pattern is a sign that something is lacking in the
RenderProcessHost/RenderProcessHostImpl/MockRenderProcessHost design, which I
would characterize as some combination of, "RenderProcessHost is doing
double-duty as both the public interface of RenderProcessHostImpl and as the
base class for a Mock version of RenderProcessHostImpl," "RenderProcessHostImpl
is too big to mock."

Historically, MockRenderProcessHost used to inherit from RenderProcessHostImpl
(this changed as part of the content api extraction:
https://chromium.googlesource.com/chromium/src/+/f3b1a084a01ab82caf998daefcb9...
)

One fix would be for us to extract a content-private class
{RenderProcessHostBase : public RenderProcessHost}, and move methods there from
both the public interface (like RenderProcessHost::Init) and static methods from
RenderProcessHostImpl. RPH -> RPHB would be a safe cast.

A more extreme fix, would be to make RPH->RPHI a safe cast, ideally by moving
the child-process-launcher behaviors into an interface that's mockable, and
making RPHI a final class. MockRPH would really only exist as a
RenderProcessHostFactory that passes a mock process-laucher-delegate or whatever
instance to the RPHI ctor, plus some easy way to get at the IPC::TestSink and
friends. This actually sounds pretty good to me as an end state, though the
reality of what the code needs to get done seems like it could throw any number
of wrenches in the feasibility of this idea.

I've put the above griping into a bug: 652474

 
   // IPC::Sender via RenderProcessHost.
   bool Send(IPC::Message* msg) override;
@@ -575,7 +576,6 @@ class CONTENT_EXPORT RenderProcessHostImpl
   scoped_refptr<ResourceMessageFilter> resource_message_filter_;
 
   mojom::RouteProviderAssociatedPtr remote_route_provider_;
-  mojom::RendererAssociatedPtr renderer_interface_;
 
   // A WeakPtrFactory which is reset every time Cleanup() runs. Used to vend
   // WeakPtrs which are invalidated any time the RPHI is recycled.