Fix leaks in exensions_unittests after enabling Oilpan

extensions/renderer/module_system_test.cc

Index: extensions/renderer/module_system_test.cc
diff --git a/extensions/renderer/module_system_test.cc b/extensions/renderer/module_system_test.cc
index 68c4bd1fedbea1ba0756956cf5a14423eca2e402..50de79fe954fb4adb62c484aac81fc2e32634df9 100644
--- a/extensions/renderer/module_system_test.cc
+++ b/extensions/renderer/module_system_test.cc
@@ -22,6 +22,7 @@
 #include "extensions/renderer/object_backed_native_handler.h"
 #include "extensions/renderer/safe_builtins.h"
 #include "extensions/renderer/utils_native_handler.h"
+#include "third_party/WebKit/public/web/WebHeap.h"
 #include "ui/base/resource/resource_bundle.h"
 
 namespace extensions {
@@ -240,6 +241,7 @@ void ModuleSystemTest::TearDown() {
     old_heap_size = stats.used_heap_size();
     isolate_->RequestGarbageCollectionForTesting(
         v8::Isolate::kFullGarbageCollection);
+    blink::WebHeap::collectAllGarbageForTesting();

[sof, 2016/01/26 06:09:43]

The v8 major GC will trigger Oilpan GCs also, so unclear if this extra round
will make a difference.

[sof, 2016/01/26 08:05:44]

https://build.chromium.org/p/chromium.memory.fyi/builders/Linux%20Tests%20(va...
shows that it didn't.

Should we force a GC with collectGarbageForTesting() instead? (I cannot
reproduce these leaks locally with memcheck, btw.)

[haraken, 2016/01/26 08:26:40]

Doesn't collectAllGarbage always collect more objects than collectGarbage?

     isolate_->GetHeapStatistics(&stats);
   }
 }