I think I've seen comments here that assume that code that manipulates only
volatiles must be correctly synchronized, and would therefore be SC.
Consider
volatile v, w, initially 0
Thread 1
r1 = v;
r2 = w;
Thread 2
w = 2;
v = 1;
Behaviour in question, r1 = 1, r2 = 0
Since r2 = 0, there cannot be a happens-before edge from w = 2 to r2 = w,
therefore r2 = w is ordered before w = 2 in the synchronization order. r1 =
v must be ordered before r2 = w in the synchronization order because
program order must be respected for the thread. Ditto w =2 is ordered
before v = 1. So r1 = v must be ordered before v = 1.
However, there is no happens-before edge from r1 = v to v =1, so there is
nothing to stop r1 = v from seeing the write v =1.
At the moment, it seems to me that behaviour in question is permitted, even
though it is not SC.
This program contains data races, so is not correctly synchronized and is
not required to be SC.
Sylvia.
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:01:04 EDT