RFR(s) #2: 6344935: (spec) clarify specifications for Object.wait overloads
Martin Buchholz
martinrb at google.com
Mon Aug 21 15:21:36 UTC 2017
My attempt to write a good wait loop led me to want only to use a
nanosecond wait method. And in fact we have existing method
Condition.awaitNanos for inspiration. For a moment I considered adding
Object#waitNanos but then I came to my senses. But we could add a static
method Objects#waitNanos and that could be the recommended method to call
inside all wait loops? Maybe it's time to replace JVM_MonitorWait (that
takes millis) with a version that takes nanos? Object.c has not been
touched since 2003!
On Sun, Aug 20, 2017 at 8:05 PM, Martin Buchholz <martinrb at google.com>
wrote:
>
>
> On Sat, Aug 19, 2017 at 1:53 PM, Martin Buchholz <martinrb at google.com>
> wrote:
>
>> Now I see that the code snippet in TimeUnit.timedWait is also in need of fixing. Hmmmm ....
>>
>> public synchronized Object poll(long timeout, TimeUnit unit)
>> throws InterruptedException {
>> while (empty) {
>> unit.timedWait(this, timeout);
>> ...
>> }
>> }
>>
>> It's surprisingly hard to write a high quality wait loop, and we've been
> doing our users a disservice by punting on providing good models in the
> javadoc. You probably want to work entirely in nanoseconds, to avoid
> rounding errors, and because you'll end up calling nanoTime anyways. You
> don't want to call nanoTime unless you are sure to wait. OTOH you want to
> throw NPE on null TimeUnit even if there's no need to wait. You must
> beware of overflow when timeout is Long.MIN_VALUE.
>
> Below is my attempt to fix the sample loop in timedWait. Calling
> Object.wait directly is obviously even harder. How about just not
> providing a sample loop that calls Object.wait directly (effectively,
> gently deprecate timed Object wait in favor of the loop below) (I'm also
> pretending that the platform has better than millisecond granularity).
>
> public E poll(long timeout, TimeUnit unit)
> throws InterruptedException {
> long nanosTimeout = unit.toNanos(timeout);
> synchronized (lock) {
> for (long deadline = 0L; isEmpty(); ) {
> long remaining = (deadline == 0L) ? nanosTimeout
> : deadline - System.nanoTime();
> if (remaining <= 0)
> return null;
> if (deadline == 0L) { // first wait
> deadline = System.nanoTime() + nanosTimeout;
> if (deadline == 0L) // unlikely corner case
> deadline = 1L;
> }
> NANOSECONDS.timedWait(lock, remaining);
> }
> return dequeueElement();
> }
> }
>
>
More information about the core-libs-dev
mailing list