views:

96

answers:

2

My question concerns safe publication of field values in Java (as discuessed here http://stackoverflow.com/questions/801993/java-multi-threading-safe-publication).

As I understand it, a field can be safely read (meaning access from multiple threads will see the correct value) if:

  • read and write are synchronized on the same monitor
  • field is final
  • field is volatile

If my understanding is correct the following class should not be thread-safe since the initial value is written without these characteristics. However I find it hard to believe that I need to make first volatile even though it is only accessed from a synchronized method.

public class Foo {

    private boolean needsGreeting = true;

    public synchronized void greet() {
        if (needsGreeting) {
            System.out.println("hello");
            needsGreeting = false;
        }
    }
}

Am I missing something? Is above code correct and if so why? Or is it necessary in such cases to make first volatile or use a final AtomicBoolean or something like that in addition to accessing it from a synchronized method.

(Just to clarify, I am aware, that if the initial value was written in a synchronized method, it would be thread-safe even without the volatile keyword.)

+3  A: 

Strictly speaking, I can't see that it is safe to assume, that needsGreeting is set to true, when greet is called.

For this to be true, there would have to be a happens before relation between the initial write (occurring when the object is constructed) and the first read (in the greet-method). Chapter 17 in the JLS however, states the following about happens-before (hb) constraints:

17.4.5 Happens-before Order Two actions can be ordered by a happens-before relationship. If one action happens-before another, then the first is visible to and ordered before the second.

If we have two actions x and y, we write hb(x, y) to indicate that x happens-before y.

  • If x and y are actions of the same thread and x comes before y in program order, then hb(x, y).
  • There is a happens-before edge from the end of a constructor of an object to the start of a finalizer (§12.6) for that object.
  • If an action x synchronizes-with a following action y, then we also have hb(x, y).
  • If hb(x, y) and hb(y, z), then hb(x, z).

Furthermore, the only way to introduce a synchronized-with relation, that is, a synchronization order is to do something of the following:

Synchronization actions induce the synchronized-with relation on actions, defined as follows:

  • An unlock action on monitor m synchronizes-with all subsequent lock actions on m (where subsequent is defined according to the synchronization order).
  • A write to a volatile variable (§8.3.1.4) v synchronizes-with all subsequent reads of v by any thread (where subsequent is defined according to the synchronization order).
  • An action that starts a thread synchronizes-with the first action in the thread it starts.
  • The write of the default value (zero, false or null) to each variable synchronizes-with the first action in every thread. Although it may seem a little strange to write a default value to a variable before the object containing the variable is allocated, conceptually every object is created at the start of the program with its default initialized values.
  • The final action in a thread T1 synchronizes-with any action in another thread T2 that detects that T1 has terminated. T2 may accomplish this by calling T1.isAlive() or T1.join().
  • If thread T1 interrupts thread T2, the interrupt by T1 synchronizes-with any point where any other thread (including T2) determines that T2 has been interrupted (by having an InterruptedException thrown or by invoking Thread.interrupted or Thread.isInterrupted).

It says nowhere that "the construction of an object happens before any calls to methods on the object. The happens-before relation however, states that there is a happens-before edge from the end of a constructor of an object to the start of a finalizer (§12.6) for that object., which may be a hint about that there is not an happens-before edge from the end of a constructor of an object to the start of an arbitrary method!

aioobe
"[...] which sort of implies that there is not an happens-before edge from the end of a constructor of an object to the start of an arbitrary method!". This "implication" is not correct.
Grodriguez
Oh, sure, it's not a formal implication. However, the fact that they state that the end of the constructor happens before the start of the finalizer method, shall we say "hints about" that this may not be true for arbitrary methods.
aioobe
@aioobe no, that is just there for cases like the following code: `new SomeObject()`, i.e. calling the constructor but not storing the reference. Then the object can immediately be garbage collected and the finalizer can immediately be called. The sentence you're referring to just makes sure that the constructor still finished before the finalizer runs.
Thomas Lötzer
Ok, I see your point. I reformulated the post slightly. Still though, I can't see that there is a guarantee that there is a happens-before relation between the end of the constructor and the start of a method call.
aioobe
Why the down vote? aioobe says there seems to be not mention in the JLS, that says that the code is correct. If he is wrong, why not reference the passage (or indicate some other resource) that gives us this guarantee?
Joe23
The happens-before relationship is clearly spelled out, and unless specified otherwise it is assumed there is no happens-before. If the end of a constructor call had a happens-before relationship with method invocations, the double-checked locking pattern would be safe.
sjlee
Downvoters, care to explain?
aioobe
+3  A: 

There is no happens-before relationship between the end of a constructor and method invocations, and as such it is possible for one thread to start constructing the instance and make the reference available and for another thread to acquire that reference and start calling the greet() method on a partially constructed object. The synchronization in greet() does not really address that issue.

If you publish an instance via the celebrated double-checked locking pattern, it becomes easier to see how. If there was such a happens-before relationship, it should have been safe even if DCLP is used.

public class Foo {
    private boolean needsGreeting = true;

    public synchronized void greet() {
        if (needsGreeting) {
            System.out.println("Hello.");
            needsGreeting = false;
        }
    }
}

class FooUser {
    private static Foo foo;

    public static Foo getFoo() {
        if (foo == null) {
            synchronized (FooUser.class) {
                if (foo == null) {
                    foo = new Foo();
                }
            }
        }
        return foo;
    }
}

If multiple threads call FooUser.getFoo().greet() at the same time, one thread might be constructing the Foo instance, but another thread may find a non-null Foo reference prematurely, and call greet() and find needsGreeting is still false.

An example of this is mentioned in Java Concurrency in Practice (3.5).

sjlee
So this is true, even though the assignment to `foo` is done *after* `new Foo()` has been fully evaluated? You have a reference confirming this?
aioobe
The assignment to foo is done after new Foo() is called *only* in that thread in which it is executed. From other threads' perspective, it may not be (or may not appear to be). The compilers and processors are permitted to reorder executions and not flush changes as long as it is consistent in the thread in which the code is executed. That's what visibility is all about after all. And that's why DCLP is broken: visibility. The Java Memory Model section of the language spec would be a good reference. Correct visibility is provided only with proper happens-before relationship.
sjlee
This also might be useful: http://www.cs.umd.edu/~pugh/java/memoryModel/DoubleCheckedLocking.html
sjlee
To my dismay it looks like you are completely right. I have deleted my own answer and am upvoting this.
Grodriguez