On Tue, 3 Jan 2006 20:55:39 -0500
kpneal@pobox.com wrote:

> On Mon, Jan 02, 2006 at 06:50:22PM -0500, der Mouse wrote:
> > Is there anything akin to memtest86+ for SPARCs?  I did a little poking
> > around and I saw some indications that some of the Linux people have
> > developed something like that for Linux; anyone know of anything
> > similar either standalone or NetBSD?
> 
> You know, for the longest time my favorite memory test has been
> to run a large compile from memory. Keep disk access to a minimum
> (no swapping) for best results. 
> 
> I once diagnosed incorrect memory (3-chip vs 9-chip SIMMS) in a 486
> by doing a kernel build (with gcc). And not too long ago a guy at
> work diagnosed an actual single bit error that caused a compiler
> crash on Sunday nights. 
> 
> I don't know what about compilers causes memory problems to surface,
> but it seems to work remarkably well.

I actually saw the case where an edgy box could compile software in an
endless loop without ever crashing / core dumping, but the binaries 
were not identical. Nothing wrong could be detected on the box with
other software than gcc...
It is of course better to choose software without a build date embedded
in the binary during the build process.

khaqq