Riccardo Mottola <rollei@tiscalinet.it> writes:

> However in my persona programmers experience, it is very rare for a
> compiler to fail when NOT optimizing. If such things occour it was
> almost always a problem of my code (pointer problem, initializations
> problems). SO I point out that there is some flakey code out there.

There's definitely still the same problem with the non-altivec
pmap_zero_page() in pmap_subr.c as there was with the altivec version;
it disables the DMMU but then executes some C code, which at low
levels of optimization will blow up trying to access stack variables
when the stack is no longer mapped. But that's not a "sometimes"
failure; if that's going to blow up at all, it'll do so very quickly.

        - Nathan