Oilpan: Move synchronous Oilpan GCs out of V8 GC epilogues

third_party/WebKit/Source/bindings/core/v8/V8GCController.cpp

Index: third_party/WebKit/Source/bindings/core/v8/V8GCController.cpp
diff --git a/third_party/WebKit/Source/bindings/core/v8/V8GCController.cpp b/third_party/WebKit/Source/bindings/core/v8/V8GCController.cpp
index 7d08e7c0ce9b7f8bb73a0fa976ddf75ced01b134..e17752b76ee898304fdd088fd54b79873f32ce3a 100644
--- a/third_party/WebKit/Source/bindings/core/v8/V8GCController.cpp
+++ b/third_party/WebKit/Source/bindings/core/v8/V8GCController.cpp
@@ -344,6 +344,8 @@ void V8GCController::gcEpilogue(v8::Isolate* isolate, v8::GCType type, v8::GCCal
         if (isMainThread()) {
             TRACE_EVENT_SET_NONCONST_SAMPLING_STATE(V8PerIsolateData::from(isolate)->previousSamplingState());
         }
+        if (ThreadState::current())

[sof, 2016/02/01 14:41:48]

What does the switch to scheduling via kGCTypeMarkSweepCompact instead of
kGCTypeProcessWeakCallbacks accomplish?

[haraken, 2016/02/01 15:05:57]

V8 is now making a change not to invoke kGCTypeProcessWeakCallbacks if there is
no second-pass callback (in common cases there is no second-pass callback with
oilpan enabled). So we cannot rely on kGCTypeProcessWeakCallbacks any more.

+            ThreadState::current()->scheduleV8FollowupGCIfNeeded(BlinkGC::V8MajorGC);
         break;
     case v8::kGCTypeIncrementalMarking:
         TRACE_EVENT_END1("devtools.timeline,v8", "MajorGC", "usedHeapSizeAfter", usedHeapSize(isolate));
@@ -356,9 +358,6 @@ void V8GCController::gcEpilogue(v8::Isolate* isolate, v8::GCType type, v8::GCCal
         if (isMainThread()) {
             TRACE_EVENT_SET_NONCONST_SAMPLING_STATE(V8PerIsolateData::from(isolate)->previousSamplingState());
         }
-        // TODO(haraken): Remove this. See the comment in gcPrologue.
-        if (ThreadState::current())
-            ThreadState::current()->scheduleV8FollowupGCIfNeeded(BlinkGC::V8MajorGC);
         break;
     default:
         ASSERT_NOT_REACHED();