Code Review 6988618: JCK test setDaemon0101 hangs on specific machine
Alan.Bateman at oracle.com
Thu Nov 4 12:13:23 UTC 2010
Chris Hegarty wrote:
> David, Alan,
> There is a race, if the thread being created completes before the
> starting thread has a chance to complete its start method ( and
> decrement the groups nUnstartedThreads count ), then when the newly
> started threads exit method notifies the group that it is terminating
> the group will not be destroyed because it believes it still as an
> unstarted thread in it.
> Basically we decrement the unstarted thread count in threadStarting.
> Since the thread is added to the group the reason for the unstarted
> thread count is no longer applicable, i.e. the thread is a real member
> of the group. We rectify the unstarted thread count if start fails,
> allowing a subsequent attempt to start the thread (this is consistent
> with current behavior).
Looks good to me. Do you think it would be feasible to include a
regression test or would it be too problematic?
More information about the core-libs-dev