RFR: ForkJoinTask leaks exceptions

[Mike Duigou]

Looks fine. Is it practical to create a regression test for this case?

Mike


On Oct 14 2013, at 08:52 , Chris Hegarty wrote:

> This is a request to pull a small change to ForkJoinTask into jdk8, from the jsr166 CVS. This was first discussed over on c-i [1] a few weeks back.
> 
> diff -r 077237e4613f src/share/classes/java/util/concurrent/ForkJoinTask.java
> --- a/src/share/classes/java/util/concurrent/ForkJoinTask.java  Mon Oct 14 11:47:54 2013 +0100
> +++ b/src/share/classes/java/util/concurrent/ForkJoinTask.java  Mon Oct 14 16:44:01 2013 +0100
> @@ -439,11 +439,13 @@ public abstract class ForkJoinTask<V> im
>         final Throwable ex;
>         ExceptionNode next;
>         final long thrower;  // use id not ref to avoid weak cycles
> +        final int hashCode;  // store task hashCode before weak ref disappears
>         ExceptionNode(ForkJoinTask<?> task, Throwable ex, ExceptionNode next) {
>             super(task, exceptionTableRefQueue);
>             this.ex = ex;
>             this.next = next;
>             this.thrower = Thread.currentThread().getId();
> +            this.hashCode = System.identityHashCode(task);
>         }
>     }
> 
> @@ -605,9 +607,9 @@ public abstract class ForkJoinTask<V> im
>     private static void expungeStaleExceptions() {
>         for (Object x; (x = exceptionTableRefQueue.poll()) != null;) {
>             if (x instanceof ExceptionNode) {
> -                ForkJoinTask<?> key = ((ExceptionNode)x).get();
> +                int hashCode = ((ExceptionNode)x).hashCode;
>                 ExceptionNode[] t = exceptionTable;
> -                int i = System.identityHashCode(key) & (t.length - 1);
> +                int i = hashCode & (t.length - 1);
>                 ExceptionNode e = t[i];
>                 ExceptionNode pred = null;
>                 while (e != null) {
> 
> 
> -Chris
> 
> [1] http://jsr166-concurrency.10961.n7.nabble.com/ForkJoinTask-leaks-exceptions-td10134.html