ref cell?

[Stephen Weeks]

> I'm curious about Shao's comment on the ref cell.  Was the ray tracing program
> really bashing on a ref cell?  (I can't imagine why.) 

Yes.  The original OCAML program used a ref cell and for loops in matrix
multiply.  As much as possible, I wanted to avoid rewriting, so the original SML 
version did too.

> If so, MLton is going
> to have an advantage because our GC is worse.

This makes no sense.  Do you mean MLton has an advantage because it doesn't have 
a generational gc?  If so, I sort of agree, although not much, because even with 
a generational gc, we would know that it is an int ref and thus not have do
anything tricky on ref update.