ICFP

[Stephen Weeks]

> I think that sounds like a good idea.  Is there anything left to formalize
> on the algorithm?  I think the analysis portion is pretty much set, but
> would it make sense to formalize the actual transformation?  If we allow
> mutual recursion for local functions, I guess it's fairly trivial.  If we
> stick to the current CPS IL, it's a little more complicated (=? interesting).

I think we should formalize the transformation, but use a modified version of
the CPS IL that allows mutual recursion.  The transformation is still not
entirely trivial, because of the changes of continuations at call sites.
Proving correctness of the transformation is probably too much for the
submission, but we can think about it for the final version.

I agree with Suresh that this will also offer a nice chance to publish MLton's
CPS IL.