Debian version of MLton

[Stephen Weeks]

> As to having to use SML/NJ or MLton to bootstrap up, is that really true?

> Would, for instance, Moscow ML really be too horribly slow even for just the
> bootstrap? 

Yes.  They have terrible space and time performance problems in their module
elaborator.  Like exponentially bad.

I'm not sure the Kit is as bad, but I wouldn't be surprised.  The MLton style of
writing signatures is not common, and you have to do a lot of tricks in the
elaborator to handle it efficiently.  SML/NJ does 'em, and so will MLton
whenever I write it.

> That reminds me, I meant to ask you about Hamlet.  I saw that you had used it
> for something (maybe a benchmark).  Did you try to give it MLton to compile
> or would that really take infinite time?

I didn't try it.  But there is no way he put the necessary tricks into his
elaborator.  And even if that were fixed, the factor of probably >20 for running
G0 makes it infeasible.

I'd say the next most likely candidate is the ML Kit, but it doesn't seem worth
spending time on.  I'd rather move to a state where we are self sufficient, and
can compile new versions of MLton with old ones.