PlzNavigate: improve tracking of improper subframe swap-out.

content/browser/frame_host/frame_tree_node.cc

Index: content/browser/frame_host/frame_tree_node.cc
diff --git a/content/browser/frame_host/frame_tree_node.cc b/content/browser/frame_host/frame_tree_node.cc
index daf4a71255d6b8a9c6f30493364f25644dd18620..9a29fb86b5814e1156a2a2130d4c0e0259f96d5f 100644
--- a/content/browser/frame_host/frame_tree_node.cc
+++ b/content/browser/frame_host/frame_tree_node.cc
@@ -71,10 +71,6 @@ FrameTreeNode::~FrameTreeNode() {
   frame_tree_->FrameRemoved(this);
 
   g_frame_tree_node_id_map.Get().erase(frame_tree_node_id_);
-  if (base::CommandLine::ForCurrentProcess()->HasSwitch(
-          switches::kEnableBrowserSideNavigation)) {
-    navigator_->CancelNavigation(this);
-  }

[carlosk, 2015/04/08 09:01:48]

N::CancelNavigation -> RFHM::CleanUpNavigation -> RFHM::DiscardUnusedFrame,
which would swap out the RFH if there should be more references to its
SiteInstance.

So in RenderFrameHostManagerTest::DetachPendingChild, when it sends the first
FrameHostMsg_Detach an extra reference to the SiteInstance was kept by the
swapped out RFH when it shouldn't. This caused the last check in this test to
fail (site_instance->HasOneRef()).

[clamy, 2015/04/08 11:29:11]

I think this needs to remain in place otherwise the navigation request for the
subframes does not get cleaned up. We ran into very similar issues when I was
fixing this test in the first place, and Nasko mentioned that creating the
swapped out RFH in the parent there might not make much sense. I suggest you
talk to him to see how to address the problem, because I don't think this is the
best approach.

[carlosk, 2015/04/08 16:05:39]

Acknowledged.

 }
 
 bool FrameTreeNode::IsMainFrame() const {