On 9.10.2015 20:05, Martin Buchholz wrote:
On Thu, Oct 8, 2015 at 11:51 PM, Jaroslav Bachorik <jaroslav.bacho...@oracle.com <mailto:jaroslav.bacho...@oracle.com>> wrote: On 8.10.2015 18:56, Martin Buchholz wrote: Hi Jaroslav, we all keep writing finalization code like this... welcome to the club! I think it would be better : - never use currentTimeMillis to measure elapsed time; use nanoTime instead Ok. I suppose this would be because currentTimeMillis() is dependent on the OS time, right? - why use complex Phaser when simple CountDownLatch will do? The logic is more complex than just waiting for the finalization to happen. I need to make sure the finalization happened due to GC.run_finalization command and not because of an ordinary GC run or JVM shutdown. I will update the test comments to make this clear. Oh, now I see what you're doing - you need to block the regular finalizer thread to make sure there will be objects available for the secondary finalizer thread to process. Although Phaser works for this, I like using simple latches - CountDownLatch(1) - because they are easier to understand. CountDownLatch done = new CountDownLatch(1); in primary finalizer thread, call done.await in secondary finalizer thread, call done.countDown to release the primary finalizer thread
Ok, I took a look at the test from distance and simplified it a bit. Did a test run of 500 iterations in tight loop without failure.
http://cr.openjdk.java.net/~jbachorik/8135188/webrev.02 -JB-
