RFR: 8288899: java/util/concurrent/ExecutorService/CloseTest.java failed with "InterruptedException: sleep interrupted" [v49]

[Viktor Klang]

On Thu, 12 Oct 2023 16:53:12 GMT, Doug Lea <dl at openjdk.org> wrote:

>> src/java.base/share/classes/java/util/concurrent/ForkJoinPool.java line 3153:
>> 
>>> 3151:     public <T> ForkJoinTask<T> submit(Callable<T> task) {
>>> 3152:         ForkJoinTask<T> t =
>>> 3153:             (Thread.currentThread() instanceof ForkJoinWorkerThread) ?
>> 
>> @DougLea Won't we need to check if the submitting thread belongs to the same pool here?
>
> No, but I can imagine this being the subject of a Puzzler someday, because either choice might be unexpected by someone! To maximize compatibility wrt incompatible specs (initial FJ and ES), calls to submit from workers are by default noninterruptible, even if they are client calls into another pool. The opposite choice might be more defensible but could break existing code.

@DougLea Ok, that's good to know. 👍

>> src/java.base/share/classes/java/util/concurrent/ForkJoinPool.java line 3324:
>> 
>>> 3322:                 poolSubmit(true, f);
>>> 3323:             }
>>> 3324:             for (int i = futures.size() - 1; i >= 0; --i)
>> 
>> @DougLea Might not be necessary for performance, but using an ArrayDeque and using `pollLast()` might be cleaner?
>
> This was a good enough idea that I started doing it until realizing/remembering that ArrayDeque doesn't implement List so can't be used here (without making a copy). Oh well.

@DougLea Ouch—I completely missed that it is also returned. Apologies :(

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/14301#discussion_r1357174453
PR Review Comment: https://git.openjdk.org/jdk/pull/14301#discussion_r1357172338