New test for volatile store+load+store in constructor

[Vladimir Sitnikov]

Hi,

What do you think of adding volatile store+load+store in constructor test
to canSeeDefault group?

I mean a test that is clone of existing [1] with additional volatile read
and store in constructor.
As far as I understand, those additional read and write should not alter
results in any way and observing of default value should still be
acceptable.

It would be great to have acceptable outcomes documented and validated
since someone might want to rely on things like "volatile load/store
barriers would flush everything and make everything visible" (see [2]).

Below is a sketch of the test [3], the key point is Shell constructor.

[1]:
http://hg.openjdk.java.net/code-tools/jcstress/file/tip/tests-custom/src/main/java/org/openjdk/jcstress/tests/init/primitives/volatiles/IntVolatileTest.java
[2]:
https://groups.google.com/forum/#!msg/mechanical-sympathy/4EDCX0F_3ow/RlDZq1MjgiMJ
[3]:
public class IntVolatileStoreLoadStoreTest implements
Actor2_Test<IntVolatileTest.State, IntResult1> {

    public static class State {
        Shell shell;
    }

    public static class Shell {
        volatile int x;
        volatile int y;

        public Shell() {
            this.x = 0xFFFFFFFF;
            this.y = this.x;
            this.x = this.y;
        }
    }

    @Override
    public State newState() {
        return new State();
    }

    @Override
    public void actor1(State s, IntResult1 r) {
        s.shell = new Shell();
    }

    @Override
    public void actor2(State s, IntResult1 r) {
        Shell sh = s.shell;
        r.r1 = (sh == null) ? 42 : sh.x;
    }

    @Override
    public IntResult1 newResult() {
        return new IntResult1();
    }

}

-- 
Regards,
Vladimir Sitnikov