Set the ui::PAGE_TRANSITION_CLIENT_REDIRECT flag.

content/renderer/render_frame_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/renderer/render_frame_impl.h"
 
 #include <map>
 #include <string>
 
 #include "base/auto_reset.h"
@@ skipping 2618 matching lines @@
   WebDataSource* data_source =
       provisional_data_source ? provisional_data_source : top_data_source;
 
   ui::PageTransition transition_type = ui::PAGE_TRANSITION_LINK;
   DocumentState* document_state = DocumentState::FromDataSource(data_source);
   DCHECK(document_state);
   InternalDocumentStateData* internal_data =
       InternalDocumentStateData::FromDocumentState(document_state);
   NavigationState* navigation_state = document_state->navigation_state();
   transition_type = navigation_state->transition_type();
+  if (data_source->isClientRedirect()) {

[nasko, 2014/11/07 16:50:05]

This doesn't seem right, as it picks up the top-level frame's data source, which
should be irrelevant for subframes. Is there a test case that demonstrates what
behavior this CL is changing?

[Fredrik Öhrn, 2014/11/10 12:51:57]

Fixed.
I'm not aware of a test case for this.

+    transition_type = static_cast<ui::PageTransition>(
+        transition_type | ui::PAGE_TRANSITION_CLIENT_REDIRECT);

[nasko, 2014/11/07 16:50:05]

Why not use safer pattern that already exists in this file at line 3369?

[Fredrik Öhrn, 2014/11/10 12:51:57]

Because I didn't find PageTransitionFromInt to add anything of value. If it was
a straight up DCHECK it would be useful but the current implementation adds
bloat the compiler can't optimize away and if transition_type actually does
contain a bad value it'll silently discard the PAGE_TRANSITION_CLIENT_REDIRECT
bit.

I'd suggest putting in a DCHECK(PageTransitionIsValidType(transition_type));
instead. However if you really want me to use PageTransitionFromInt, sure I'll
add it.

[nasko, 2014/11/10 15:01:44]

Why not improve the code then? I would go even a step further, why use DCHECK
instead of CHECK? If we ever have code that is using invalid value, this is bad
enough to kill the renderer IMHO.

+  }
 
   GURL request_url(request.url());
   GURL new_url;
   if (GetContentClient()->renderer()->WillSendRequest(
           frame,
           transition_type,
           request_url,
           request.firstPartyForCookies(),
           &new_url)) {
     request.setURL(WebURL(new_url));
@@ skipping 1259 matching lines @@
 
 #if defined(ENABLE_BROWSER_CDMS)
 RendererCdmManager* RenderFrameImpl::GetCdmManager() {
   if (!cdm_manager_)
     cdm_manager_ = new RendererCdmManager(this);
   return cdm_manager_;
 }
 #endif  // defined(ENABLE_BROWSER_CDMS)
 
 }  // namespace content