After reflecting on this, and in the light of Doug's comments about the use
of atomic locks, it seems to me that my 'thread model implementation' is
also capable of being recast so that it can use atomic locks. So there are
at least two such solutions.
It is far from clear to me that either solution then represents a
noticeable cost over implementations that rely on the pthread_cond_wait or
its equivalent. Indeed, if the underyling pthread implementation is using
the notifyAll mechanism to avoid lost notifications, then these higher
level solutions may represent a cost saving where interupts are used.
Sylvia.
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:55 EDT