[Autofill] Migrate ContentAutofillDriver<-->AutofillAgent IPCs to mojo.

content/browser/frame_host/render_frame_host_impl.cc

 // Copyright 2013 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "content/browser/frame_host/render_frame_host_impl.h"
 
 #include <utility>
 
 #include "base/bind.h"
 #include "base/command_line.h"
@@ skipping 2376 matching lines @@
   // An error page is expected to commit, hence why is_loading_ is set to true.
   is_loading_ = true;
   frame_tree_node_->ResetNavigationRequest(true);
 }
 
 void RenderFrameHostImpl::SetUpMojoIfNeeded() {
   if (interface_registry_.get())
     return;
 
   interface_registry_.reset(new shell::InterfaceRegistry(nullptr));
-  if (!GetProcess()->GetRemoteInterfaces())
+
+  // The RenderProcessHost may not have remote interfaces in some test
+  // environments. In that case we create an empty InterfaceProvider so tests
+  // can register mock implementations.

[dcheng, 2016/07/12 14:22:13]

Can we fix the test harness (ChromeRenderViewTest, I assume?) to just set this
up correctly? I'd rather limit the test-specific code to tests.

[leonhsl(Using Gerrit), 2016/07/13 03:48:31]

I think even we change ChromeRenderViewTest and some other test harness, we
still can't avoid adding some logic into RenderProcessHostImpl to access the
|remote_interfaces_|. And, seems Ben and Ken are working for some big changes
for content layer {InterfaceProvider,InterfaceRegistry}, so later maybe the
problem will disappear or we have to go a different way here. So I don't think
current solution is unacceptable.

[dcheng, 2016/07/14 04:31:44]

To help me understand better, which tests were failing without this?

[leonhsl(Using Gerrit), 2016/07/14 09:57:31]

I encountered the problem in
https://codereview.chromium.org/2007473004/diff/300001/components/autofill/co...
Line267, where I found the |remote_interfaces_| of main render frame host is
always nullptr. 

[dcheng, 2016/07/19 14:34:11]

I think we just need to set an InterfaceProvider for the MockRenderProcesHost's
interface registry to make this work, right? See
https://cs.chromium.org/chromium/src/content/public/test/mock_render_process_...

[leonhsl(Using Gerrit), 2016/07/20 03:01:07]

Which on do you mean?
  1, Initialize MockRenderProcesHost class with an empty InterfaceProvider;
  2, Set an empty InterfaceProvider for MockRenderProcesHost only for this test
case.

Actually I don't understand why we must avoid initializing an empty
InterfaceProvider for RFHI, this behavior aligns with before(When we were using
ServiceRegistry).

+  if (!GetProcess()->GetRemoteInterfaces()) {
+    remote_interfaces_.reset(new shell::InterfaceProvider);
     return;
+  }
 
   RegisterMojoInterfaces();
   mojom::FrameFactoryPtr frame_factory;
   GetProcess()->GetRemoteInterfaces()->GetInterface(&frame_factory);
   frame_factory->CreateFrame(routing_id_, GetProxy(&frame_),
                              frame_host_binding_.CreateInterfacePtrAndBind());
 
 
   shell::mojom::InterfaceProviderPtr remote_interfaces;
   shell::mojom::InterfaceProviderRequest remote_interfaces_request =
@@ skipping 549 matching lines @@
   // handler after it's destroyed so it can't run after the RFHI is destroyed.
   web_bluetooth_service_->SetClientConnectionErrorHandler(base::Bind(
       &RenderFrameHostImpl::DeleteWebBluetoothService, base::Unretained(this)));
 }
 
 void RenderFrameHostImpl::DeleteWebBluetoothService() {
   web_bluetooth_service_.reset();
 }
 
 }  // namespace content