Add a counter and a deprecation message when navigating in an unload handler.

third_party/WebKit/Source/core/loader/FrameLoader.cpp

 /*
  * Copyright (C) 2006, 2007, 2008, 2009, 2010, 2011 Apple Inc. All rights reserved.
  * Copyright (C) 2008 Nokia Corporation and/or its subsidiary(-ies)
  * Copyright (C) 2008, 2009 Torch Mobile Inc. All rights reserved. (http://www.torchmobile.com/)
  * Copyright (C) 2008 Alp Toker <alp@atoker.com>
  * Copyright (C) Research In Motion Limited 2009. All rights reserved.
  * Copyright (C) 2011 Kris Jordan <krisjordan@gmail.com>
  * Copyright (C) 2011 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ skipping 1069 matching lines @@
         float ratio = static_cast<float>(nodeCount) / totalNodeCount;
         ThreadState::current()->schedulePageNavigationGCIfNeeded(ratio);
     }
 
     // Don't allow any new child frames to load in this frame: attaching a new
     // child frame during or after detaching children results in an attached
     // frame on a detached DOM tree, which is bad.
     SubframeLoadingDisabler disabler(m_frame->document());
     if (m_documentLoader) {
         client()->dispatchWillClose();
+
+        NavigationCounterForUnload counter;

[Nate Chapin, 2016/05/13 21:56:10]

Is there any reason not to put this in dispatchUnloadEvent()?

[lfg, 2016/05/14 01:23:28]

The only reason is because I didn't want to fire the deprecation warning in
cases where we already block the navigations (scheduled navigations get
cancelled here:
https://code.google.com/p/chromium/codesearch#chromium/src/third_party/WebKit...).

[Nate Chapin, 2016/05/16 21:00:31]

Right, only same-document navigations can run and complete here. But
dispatchUnloadEvent() is also called in other places for frame detach, page
close, and swap out, and the unload event could certainly do a same-document
navigation from there. Is it intentional to exclude those cases from the
counter?

[lfg, 2016/05/16 23:46:54]

Ah, good point. Yes, we should count those cases as well, I've moved the code to
dispatchUnloadEvent().

         dispatchUnloadEvent();
     }
     m_frame->detachChildren();
     // The previous calls to dispatchUnloadEvent() and detachChildren() can
     // execute arbitrary script via things like unload events. If the executed
     // script intiates a new load or causes the current frame to be detached,
     // we need to abandon the current load.
     if (pdl != m_provisionalDocumentLoader)
         return false;
     // detachFromFrame() will abort XHRs that haven't completed, which can
@@ skipping 464 matching lines @@
     tracedValue->setString("documentLoaderURL", m_documentLoader ? m_documentLoader->url() : String());
     return tracedValue.release();
 }
 
 inline void FrameLoader::takeObjectSnapshot() const
 {
     TRACE_EVENT_OBJECT_SNAPSHOT_WITH_ID("loading", "FrameLoader", this, toTracedValue());
 }
 
 } // namespace blink