Using double-checked locking on mutable objects in non-volatile fields is not thread-safe.
If the field is not volatile, the compiler may re-order the code in the accessor. For more information, see:
The canonical example of correct double-checked locking for lazy initialization is:
class Foo { /** This foo's bar. Lazily initialized via double-checked locking. */ private volatile Bar bar; public Bar getBar() { Bar value = bar; if (value == null) { synchronized (this) { value = bar; if (value == null) { bar = value = computeBar(); } } } return value; } private Bar computeBar() { ... } }
Double-checked locking should only be used in performance critical classes. For code that is less performance sensitive, there are simpler, more readable approaches. Effective Java recommends two alternatives:
For lazily initializing instance fields, consider a synchronized accessor. In modern JVMs with efficient uncontended synchronization the performance difference is often negligible.
// Lazy initialization of instance field - synchronized accessor private Object field; synchronized Object get() { if (field == null) { field = computeValue(); } return field; }
If the field being initialized is static, consider using the lazy initialization holder class idiom:
// Lazy initialization holder class idiom for static fields private static class Holder { static final Object field = computeValue(); } static Object get() { return Holder.field; }
If the object being initialized with double-checked locking is immutable, then it is safe for the field to be non-volatile. However, the use of volatile is still encouraged because it is almost free on x86 and makes the code more obviously correct.
Note that immutable has a very specific meaning in this context:
[An immutable object] is transitively reachable from a final field, has not changed since the final field was set, and a reference to the object containing the final field did not escape the constructor.
Double-checked locking on non-volatile fields is in general unsafe because the compiler and JVM can re-order code from the object's constructor to occur after the object is written to the field.
The final modifier prevents that re-ordering from occurring, and guarantees that all of the object's final fields have been written to before a reference to that object is published.