[DRAFT] Create RenderFrameHostManager.

content/browser/loader/resource_loader.cc

Index: content/browser/loader/resource_loader.cc
diff --git a/content/browser/loader/resource_loader.cc b/content/browser/loader/resource_loader.cc
index 94561ba3497ccbf1a709028b1240185b7124db47..0757b3ce281e7b6a64f255b69a6fef40e365769a 100644
--- a/content/browser/loader/resource_loader.cc
+++ b/content/browser/loader/resource_loader.cc
@@ -160,8 +160,10 @@ void ResourceLoader::ReportUploadProgress() {
 }
 
 void ResourceLoader::MarkAsTransferring(const GURL& target_url) {
-  CHECK_EQ(GetRequestInfo()->GetResourceType(), ResourceType::MAIN_FRAME)
-      << "Cannot transfer non-main frame navigations";
+  // TODO(creis): Why was this check here?  Is it a problem for cookies?

[davidben, 2013/11/18 22:20:50]

I actually came across that line some time ago and looked into this.

It's from http://crrev.com/217113. For sub-frames, set_first_party_for_cookies
probably shouldn't get called because an frame's first-party for cookies should
still be the top-level frame or something like that? Hence the MAIN_FRAME check.

I think neither that target_url parameter nor the set_first_party_for_cookies
call are needed anymore anyway. There is another set_first_party_for_cookies
call in AsyncResourceHandler that comes from the renderer, and the transfer
doesn't happen until after redirects are followed anyway. I never checked at
what point it stopped being necessary, but I would guess some combination of
when the handler chain stopped being rebuilt on transfers and when transfers
stopped happening at every redirect.

+  CHECK(GetRequestInfo()->GetResourceType() == ResourceType::MAIN_FRAME ||
+        GetRequestInfo()->GetResourceType() == ResourceType::SUB_FRAME)
+      << "Can only transfer navigations";
   is_transferring_ = true;
 
   // When transferring a request to another process, the renderer doesn't get