Implement NavigationThrottle::BLOCK_REQUEST_AND_COLLAPSE.

content/common/frame_messages.h

Index: content/common/frame_messages.h
diff --git a/content/common/frame_messages.h b/content/common/frame_messages.h
index cc85b72bc625c42a62289dde958744a21505d589..e99ba76f3516da440b3556bfd4429906504da368 100644
--- a/content/common/frame_messages.h
+++ b/content/common/frame_messages.h
@@ -741,6 +741,10 @@ IPC_MESSAGE_ROUTED1(FrameMsg_SetAccessibilityMode,
 // Dispatch a load event in the iframe element containing this frame.
 IPC_MESSAGE_ROUTED0(FrameMsg_DispatchLoad)
 
+// Sent to a child frame to control whether its frame owner element in the
+// parent frame should be collapsed, i.e. removed from the layout.
+IPC_MESSAGE_ROUTED1(FrameMsg_CollapseFrameOwner, bool /* collapsed */)

[clamy, 2017/01/18 15:10:55]

Wouldn't is be easier to send the IPC to the parent instead, telling it to
collapse a particular child? (Note sure if doable though).

[engedy, 2017/01/18 23:37:14]

Theoretically, we could use |unique_name|. This would also mean that we never
have to send IPCs to proxies. @Nasko, are unique names reliable enough for this
purpose?

[engedy, 2017/02/17 18:09:25]

@Nasko, what do think about this?  To me, sending the IPC to the parent and then
routing it to the right WebFrame child based on unique name sounds a bit like
duplicating the main routing logic, but I am not strongly against this.

Also, now that we settled on storing the state on the child FrameTreeNode, would
be a bit awkward to pipe it through the parent RFManager.

+
 // Notifies the frame that its parent has changed the frame's sandbox flags.
 IPC_MESSAGE_ROUTED1(FrameMsg_DidUpdateSandboxFlags, blink::WebSandboxFlags)