(Sending along some comments arising from deep within application code...) I don't remember having seen the rules for the usage of @Volatile spelt out in the X10 manual. We should.
If I (mistakenly) mark an AtomicReference as @Volatile, the x10c++ post-compiler gives an error: x10c++: "./xrudra/util/XchgBuffer.h", line 134.113: 1540-0264 (S) The non-volatile member function "x10::util::concurrent::AtomicReference<xrudra::TimedWeight *>::getAndSet(TimedWe\ ight *)" is called for "volatile x10::util::concurrent::AtomicReference<xrudra::TimedWeight *>". Not sure how to make sense of it as an X10 programmer. What am I supposed to do to fix this problem at the level of the X10 program? The wording of the error message seems to indicate that if I invoke some X10 method defined on type T on a field datum declared as @Volatile datum:T, I might have a problem...? In this case, I recognized the problem as a left-over edit. Earlier I was using an AtomicBoolean to guard a value of type T that was being read/written by two different threads, so I marked the value as @Volatile. Then I changed the code so that the value was of type AtomicReference[T], and mistakenly left the @Volatile in place. It is not needed -- removing it was adequate to fix this problem. Minor related note: Why is the static factory method on AtomicReference not called make, as in other parts of the X10 libraries? (It is called newAtomicReference(...).) ------------------------------------------------------------------------------ _______________________________________________ X10-users mailing list X10-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/x10-users
