x86 snapshot

[Matthew Fluet]

On the interesting/amusing side, in cleaning up the use of the "future"
list, I generated the second strangest bug that I've encountered: a
G0 version of the compiler that sucessfully compiled raytrace.sml, but
produced an executable that would only produce black and white images.
Stranger yet, they were the "right" black and white images -- for
fractal.gml I got circles in the right places, black shadows and all.
Very annoying -- I actually prefer segfaults, at least you know precisely
where something went wrong.

The strangest bug I've encountered was a while ago: a G0 version of the
compiler that successfully passed all the regression tests and benchmarks,
could do a self compile to produce a G1 version of the compiler that would
also pass the regression tests, yet would not recognize the -o switch.
All the other switches worked, just not -o.