On 4/11/2013 1:42 PM, Martin Buchholz wrote:
On Sun, Nov 3, 2013 at 5:09 PM, David Holmes <david.hol...@oracle.com>wrote:
Locking access to a CountDownLatch just seems inherently wrong. I get that
it is the atomicity of the two calls that we want, but this still seems
unpleasant. I've looked at Martin's suggested fix and confess that I am
struggling to understand what exactly are we trying to achieve with this
synchronization ??
I was scratching my head trying to understand the author's intent as well.
The idea IIRC was to have the last cohort of pool-size threads all start
executing "as concurrently as possible" to try to tickle any races.
In that case Tristan's locking scheme is likely to be counter-productive
due to the forced serialization.
David
Perhaps that was indeed able to help repro some bug back in 2007.