> > Hmmmmm, are you _sure_ that's a bug ? The cast says that the pointed
> > value is volatile, not the pointer itself. So the compiler can, I
> > think, still legally assume that the memory zone assigned through the
> > pointer will be destroyed after the return.
>
> Oh bog.. I do indeed have the volabloodytile keyword in the wrong place
> yes. Try it as he suggested and lets see
>
> > *((u32 * volatile)&eflags)=val; should make it behave though. I'm not
> > really sure. That's quite a tricky part of the standard.
I just tested the (u32 * volatile) version. It still generates incorrect
code with 2.8.0. It seems that gcc 2.8.0's doesn't check properly for
volatile in the ADDRESSOF optimization.
-Andi