RE: JavaMemoryModel: New Unified JMM Description

From: David Holmes (dholmes@dltech.com.au)
Date: Thu Apr 15 2004 - 02:04:38 EDT


I wrote:
> I don't follow. Just because there is a happens-before edge between y =
> default and r1 = y it doesn't mean there is a happens-before edge
> between x
> = 1 and y = default, or between r1 = y and r2 = x. The order of statements
> in both or either threads could be switched around.

Okay I was quite confused. Jeremy has spent a lot of time today trying to
help remedy this - with limited success I'm afraid.

Basic point: actions that have a happens-before relationship need not happen
in that order. They are allowed to be reordered provided that the reordering
is not visible to anything that has a happens-after relationship with the
reordered actions.

What I'm getting stuck on is a set of rules for reasoning about whether
certain reorderings are allowed, such that I can reason about lock-free
algorithm using volatiles.

Getting back to Doron's original issue about default initialization, the key
point here is that the behaviour is as-if every object that would be created
in a program was created with its default initialized values prior to the
program starting. Hence every action in the program must happen-after that
initialization. So given:

  class Y { int y; }

then this code:
   x = 1;
   new Y();

might appear to be the same as this:

  x = 1;
  y = 0;

but it is not because "y=0" effectively happened before the program started.

The way this is currently described is not clear.

David Holmes

-------------------------------
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