Tim -

If I'm not mistaken, Ian made it so that you can indeed run VMs which 
pretend to be loaded "on the wrong side" of the 2GB barrier. It would be 
slower since there is some extra address translation involved but I 
think it's doable to compile a VM for test purposes that exposes the 
problem if it's still there.

Cheers,
   - Andreas

Tim Rowledge wrote:
> The '2gb' problem is where the address space allocated to squeak's
> object memory goes over 2gb - ie the point where signed/unsigned
> becomes critical. Lots of the code in the VM was written without any
> thought for this issue; it was after all written a long time ago when
> 16Mb was a big machine.
> 
> The latest code is much better though probably not totally immune. The
> problem can appear to be random, dpending somewhat upon the memory
> allocation stratgey of the OS (version) you are running. On my machine
> it can depend on what applications have run before I start Squeak,
> since that affects where available memory will be.
> 
> tim
> --
> Tim Rowledge, tim at rowledge.org, http://www.rowledge.org/tim
> Death is a nonmaskable interrupt.
> 
>