/*
- * Copyright (C) 2012 Apple Inc. All rights reserved.
+ * Copyright (C) 2012, 2013 Apple Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
#include "config.h"
#include "ArrayAllocationProfile.h"
-#include "Operations.h"
+#include "JSCInlines.h"
namespace JSC {
void ArrayAllocationProfile::updateIndexingType()
{
- if (!m_lastArray)
+ // This is awkwardly racy but totally sound even when executed concurrently. The
+ // worst cases go something like this:
+ //
+ // - Two threads race to execute this code; one of them succeeds in updating the
+ // m_currentIndexingType and the other either updates it again, or sees a null
+ // m_lastArray; if it updates it again then at worst it will cause the profile
+ // to "forget" some array. That's still sound, since we don't promise that
+ // this profile is a reflection of any kind of truth.
+ //
+ // - A concurrent thread reads m_lastArray, but that array is now dead. While
+ // it's possible for that array to no longer be reachable, it cannot actually
+ // be freed, since we require the GC to wait until all concurrent JITing
+ // finishes.
+
+ JSArray* lastArray = m_lastArray;
+ if (!lastArray)
return;
- m_currentIndexingType = leastUpperBoundOfIndexingTypes(m_currentIndexingType, m_lastArray->structure()->indexingType());
+ m_currentIndexingType = leastUpperBoundOfIndexingTypes(m_currentIndexingType, lastArray->indexingType());
m_lastArray = 0;
}