Alexander,
Thank you for the contribution. The final update:
http://cr.openjdk.java.net/~ssadetsky/8130735/webrev.02/
--Semyon
On 7/21/2015 2:31 PM, Alexander Scherbatiy wrote:
On 7/21/2015 1:40 PM, Semyon Sadetsky wrote:
In the run() the delayed timer is obtained from timer, so it is
always the resulting delayed timer.
Just to use the delayed timer from the queue and not from the timer?
TimerQueue.run(){
DelayedTimer delayedTimer = queue.take(); // note that this
delayed timer is not the same as timer.delayedTimer
delayedTimer.getTimer().lock();
if(delayedTimer.removed){
// skip it
}
// ...
delayedTimer.getTimer().unlock();
}
Thanks,
Alexandr.
How do you propose to detect that timer was renewed exactly during
the time interval between the timer is taken from the queue and the
lock is captured in the run() thread?
On 7/21/2015 12:53 PM, Alexander Scherbatiy wrote:
On 7/21/2015 11:36 AM, Semyon Sadetsky wrote:
Hi Alexander,
The remove() method set delayedTimer field to null, and it is
checked in run() to be not null. So it acts in the same way as flag
you've proposed.
The issue is about how to detect consequent remove() ans add() in
run(). If you use some flag in remove() then you need to clear it
in add()
I suppose that you are talking about some flags in Timer. The
proposed change suggests to add the flag to the DelayedTimer. So
after stopping a timer (timer.delayedTimer = null) the delayedTimer
taken from the queue contains information that it has been removed.
Thanks,
Alexandr.
because after the add() the timer should run normally with the new
period. And you cannot clear such flag in the first consequent
run() after add() because you cannot determine the moment the add()
was called. So it is just the same problem.
--Semyon
On 7/21/2015 10:35 AM, Alexander Scherbatiy wrote:
What about to add removed flat to DelayedTimer? Something like:
TimerQueue.removeTimer(Timer timer){
timer.lock();
// ...
timer.delayedTimer.removed = true;
queue.remove(timer.delayedTimer);
timer.delayedTimer = null;
timer.unlock();
}
TimerQueue.run(){
DelayedTimer delayedTimer = queue.take();
delayedTimer.getTimer().lock();
if(delayedTimer.removed){
// skip it
}
// ...
delayedTimer.getTimer().unlock();
}
Thanks,
Alexandr.
On 7/17/2015 12:28 AM, Sergey Bylokhov wrote:
Hi, Semyon.
I see that the chance to reproduce the problem is very very
small, because we should call addTimer, when we are at lines
171/172. So the bug is about really small timings. So the related
question: Is it possible in the fixed version to call addTimer
when we remove DelayedTimer from the queue via queue.take(), but
before we assign its value to the runningTimer?
On 14.07.15 12:51, Alexander Zvegintsev wrote:
still looks good to me.
Thanks,
Alexander.
On 07/14/2015 12:41 PM, Semyon Sadetsky wrote:
Hi Alexander,
I added the double check
:http://cr.openjdk.java.net/~ssadetsky/8130735/webrev.01/
--Semyon
On 7/13/2015 1:24 PM, Alexander Zvegintsev wrote:
Hello Semyon,
the fix looks good to me.
P.S. Just a side note, as I can see we could possibly start
two threads instead of one in startIfNeeded():
96 void startIfNeeded() {
97 if (! running) {
98 runningLock.lock();
99 try {
100 final ThreadGroup threadGroup =
AppContext.getAppContext().getThreadGroup();
101 AccessController.doPrivileged((PrivilegedAction<Object>)
() -> {
102 String name = "TimerQueue";
103 Thread timerThread = new
ManagedLocalsThread(threadGroup,
104 this, name);
!running check is missing after try. It is not the case with
current code base, but it may be changed in future.
Thanks,
Alexander.
On 07/09/2015 08:08 PM, Semyon Sadetsky wrote:
Hello,
Please review fix for JDK9:
bug: https://bugs.openjdk.java.net/browse/JDK-8130735
webrev: http://cr.openjdk.java.net/~ssadetsky/8130735/webrev.00/
The root cause is the setting larger expiration time for the
timer which is already inserted into the delay queue. So all
timers behind the timer cannot be executed earlier than its
expiration time. This happens very rare only for repeated
timers and only if user uses the Swing timer API inaccurately
(call start() without stop()).
The fix eliminates this possibility by introducing a check if
the timer was already restarted concurrently.
It is difficult to write test because I could not reliably
reproduce the issue for a reasonable time.
--Semyon
--
Best regards, Sergey.
