> David, time for another scan over the time gathering code I think...
> 
> I've just done a bootstrap of gcc (from the trunk) on a NetBSD snapshot 
> that I built at the weekend and got the following output times:
> 
> User: -2147438122.41
> System: -2147473797.07
> Elapsed: 54569.25
> 
> A normal build will give something like
> 
> User: 45921.03
> System: 11102.27
> Elapsed: 56272.91
> 
> Which is much more reasonable, but still not quite right...
> 
>    User + System > Elapsed
> 
> and this is a single processor machine that was also doing other things 
> (though not very much) :-)
> 

Hmm, interesting.

-2147438122 as a hex number is 0x8000b1d6 and 0xb1d6 (the bottom four 
digits) in decimal is 45526 which is roughly what I would have expected 
the User time to be.

So my guess is that this might be a 32->64 bit time roll-over race 
somewhere...

R.