Simple BlinkGC heap compaction.

third_party/WebKit/Source/wtf/HashTable.h

Index: third_party/WebKit/Source/wtf/HashTable.h
diff --git a/third_party/WebKit/Source/wtf/HashTable.h b/third_party/WebKit/Source/wtf/HashTable.h
index 2afcfba70189a2ce23dcbd70bc11168a516c8317..3fdc563502b1bba15ccd898309815e51fb4114ad 100644
--- a/third_party/WebKit/Source/wtf/HashTable.h
+++ b/third_party/WebKit/Source/wtf/HashTable.h
@@ -2081,8 +2081,12 @@ void HashTable<Key,
   // needed.
   if (Traits::weakHandlingFlag == NoWeakHandlingInCollections) {
     Allocator::markNoTracing(visitor, m_table);
+    // If the backing store will be moved by sweep compaction, register the
+    // table reference pointing to the backing store object, so that the
+    // reference is updated upon object relocation.
+    Allocator::registerBackingStoreReference(visitor, &m_table);
   } else {
-    Allocator::registerDelayedMarkNoTracing(visitor, m_table);
+    Allocator::registerDelayedMarkNoTracing(visitor, &m_table);

[haraken, 2016/11/30 06:29:53]

Why don't we need to register the backing store of weak hash tables?

[sof, 2016/11/30 06:52:44]

See above answer.

     // Since we're delaying marking this HashTable, it is possible that the
     // registerWeakMembers is called multiple times (in rare
     // cases). However, it shouldn't cause any issue.