LeakDetector: Terminate all WebEmbeddedWorkerImpl before counting

Source/web/WebLeakDetector.cpp

 /*
  * Copyright (C) 2014 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are
  * met:
  *
  *     * Redistributions of source code must retain the above copyright
  * notice, this list of conditions and the following disclaimer.
  *     * Redistributions in binary form must reproduce the above
@@ skipping 23 matching lines @@
 
 #include "bindings/core/v8/V8Binding.h"
 #include "bindings/core/v8/V8GCController.h"
 #include "core/dom/Document.h"
 #include "core/fetch/MemoryCache.h"
 #include "core/fetch/ResourceFetcher.h"
 #include "core/inspector/InspectorCounters.h"
 #include "platform/Timer.h"
 #include "public/web/WebDocument.h"
 #include "public/web/WebLocalFrame.h"
+#include "web/WebEmbeddedWorkerImpl.h"
 
 #include <v8.h>
 
 using namespace blink;
 
 namespace blink {
 
 namespace {
 
 // FIXME: Oilpan: It may take multiple GC to collect on-heap objects referenced from off-heap objects.
@@ skipping 21 matching lines @@
     void delayedReport(Timer<WebLeakDetectorImpl>*);
 
     WebLeakDetectorClient* m_client;
     Timer<WebLeakDetectorImpl> m_delayedGCAndReportTimer;
     Timer<WebLeakDetectorImpl> m_delayedReportTimer;
     int m_numberOfGCNeeded;
 };
 
 void WebLeakDetectorImpl::collectGarbageAndGetDOMCounts(WebLocalFrame* frame)
 {
+    WebEmbeddedWorkerImpl::terminateAll();

[michaeln, 2014/07/25 00:07:09]

i think it will take some time to unwind the background threads, do you need a
continuation (completioncallback) of some kind?

[kouhei (in TOK), 2014/07/25 01:33:09]

LeakDetector already flushes the task queue 2 times for oilpan GC, so I think
this is practically fine. Can I try landing as this first?

[michaeln, 2014/07/25 02:49:52]

i doubt if the collection of running instances is empty in two turns of the
message loop, are you able to test prior to landing?

[kouhei (in TOK), 2014/07/25 05:00:39]

Yes I confirmed that there are no false positive leaks after this patch is
applied.

     memoryCache()->evictResources();
 
     {
         RefPtrWillBeRawPtr<Document> document = PassRefPtrWillBeRawPtr<Document>(frame->document());
         if (ResourceFetcher* fetcher = document->fetcher())
             fetcher->garbageCollectDocumentResources();
     }
 
     // FIXME: HTML5 Notification should be closed because notification affects the result of number of DOM objects.
 
@@ skipping 42 matching lines @@
 }
 
 } // namespace
 
 WebLeakDetector* WebLeakDetector::create(WebLeakDetectorClient* client)
 {
     return new WebLeakDetectorImpl(client);
 }
 
 } // namespace blink