[K42-discussion] race condition in clustered object call on unoptimized build

[Michal Ostrowski]

On Sun, 2006-02-12 at 09:00 -0500, Bryan S Rosenburg wrote:
> 
> You've missed my point.  We want every "DREF(ref)" to dereference ref
> exactly once.  The problem that's been reported is that the compiler
> is dereferencing ref more than once for a single DREF instance.
>  Burying the dereference in a volatile asm would keep the compiler
> from doing that. 


Yes.

Actually, if you enforced certain compiler optimizations you could
ensure that given:

DREF(obj)->x();
DREF(obj)->y();


The result of both DREF()'s would always be the same; so that within a
given functional scope you could never see "obj" becoming deleted.

Michal