Fix regression with back-button not working on prerendered and instant pages.

content/browser/tab_contents/tab_contents.cc

 // Copyright (c) 2011 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/tab_contents/tab_contents.h"
 
 #include <cmath>
 
 #include "base/command_line.h"
 #include "base/metrics/histogram.h"
@@ skipping 1783 matching lines @@
   upload_position_ = upload_position;
   upload_size_ = upload_size;
   load_state_host_ = net::IDNToUnicode(url.host(),
       content::GetContentClient()->browser()->GetAcceptLangs(this));
   if (load_state_ == net::LOAD_STATE_READING_RESPONSE)
     SetNotWaitingForResponse();
   if (IsLoading())
     NotifyNavigationStateChanged(INVALIDATE_LOAD | INVALIDATE_TAB);
 }
 
+void TabContents::OffsetAndPruneHistory(int offset) {
+  RenderViewHost* rvh = render_view_host();
+  if (!rvh)
+    return;
+  rvh->Send(new ViewMsg_OffsetAndPruneHistory(rvh->routing_id(), offset));

[brettw, 2011/08/09 16:29:47]

Don't we need to do something similar to this for the swapping RVH case for
normal cross-site transitions? How does that work? I'm surprised we needed new
machinery for doing this.

[cbentzel, 2011/08/09 16:47:22]

cross-site transitions work correctly with the existing code - this was
something I looked at early on when debugging why the prerender case doesn't
work.

from the bug:

I was concerned about the general cross-domain navigation case as well when
debugging this last night.

The reason it works is the initial OnNavigate call has non-zero
current_history_list_offset and current_history_list_length.

Let's say the navigation controller has a sequence like the following

chrome://newtab
www.a.com
www.b.com
www.b.com/foo

We'll create a new RenderView when transitioning from www.a.com to www.b.com.
However, the ViewMsg_Navigate_Params has a current_history_list_offset of 2 and
length of 3 [I think, will need to see the log statements again]. The first two
entries are filled in with -1, since they are controlled by other render views.

When you do the Go Back at this point, a Navigate is sent with
pending_history_list_offset to 2, and that's treated like a valid page
correctly.

[brettw, 2011/08/09 18:14:40]

Okay, I think that answers my question: in the cross-site case the offsets are
sent with OnNavigate. Since we don't necessarily know the right ones at the time
of preloading, they may be wrong.

This brings up an interesting point that history.length will be wrong in a
prerendered page. I think this probably isn't worth fixing right now, but it's
worth keeping in mind (maybe we should file a bug or put a note somewhere in the
code about this?)

[cbentzel, 2011/08/09 19:12:14]

Thanks for thinking of that case. I entered a bug:
http://code.google.com/p/chromium/issues/detail?id=92202

+}
+
 void TabContents::WorkerCrashed() {
   if (delegate())
     delegate()->WorkerCrashed(this);
 }
 
 void TabContents::BeforeUnloadFiredFromRenderManager(
     bool proceed,
     bool* proceed_to_fire_unload) {
   if (delegate())
     delegate()->BeforeUnloadFired(this, proceed, proceed_to_fire_unload);
@@ skipping 90 matching lines @@
 }
 
 void TabContents::SwapInRenderViewHost(RenderViewHost* rvh) {
   render_manager_.SwapInRenderViewHost(rvh);
 }
 
 void TabContents::CreateViewAndSetSizeForRVH(RenderViewHost* rvh) {
   RenderWidgetHostView* rwh_view = view()->CreateViewForWidget(rvh);
   rwh_view->SetSize(view()->GetContainerSize());
 }