Hi, Clemens, the minimal patch looks fine.
The full patch also looks OK. Yesterday I thought in exactly the same direction and here is my observation. EDT.pumpEventsForFilters(id) can return the "false" value only if the thread is interrupted or stopped (ThreadDeath is caught). Therefore when the pumpEvents() call in EDT.run() exits, "shutdown" is always true. That's why I wrote about detaching the event dispatch thread unconditionally in my previous email.
Did I miss anything? Thanks, Artem On 8/24/2011 2:05 PM, Clemens Eisserer wrote:
Hi, Please find the proposed fix at: http://cr.openjdk.java.net/~ceisserer/7081670/ The "minimal" patch just tries to restore the behaviour when monitors were used, trying to be low-risk. The "full" patch does no longer rely on async exeptions (but still honors them if they happen), and calls detachQueue unconditionally on shutdown (but still takes care of pending events of only stopDispatch() has been called). I didn't change AppContext.dispose(), as I think calling Thread.stop() may be still benefitial to kill application-specific threads which don't honor interrupt(). I took the liberty to do a small amount of low-risk refactoring: - Replaced Vector with ArrayList, as it is only accessed after acquiring its monitor - Replaced event-queue id gerenation code with AtomicInteger - Made waitForID volatile, as it is accessed by multiple threads. Thanks, Clemens
