[MLton] RE: VC++ as the C compiler for MLton? (Was: Compiler dependencies)

[Stephen Weeks]

> Does practical build infrastructure exist for using MLton in this way?
> i.e. spend most of one's time building in SML/NJ, once in awhile build
> MLton.  

Yes.  This is how most people who use MLton develop, I suspect.  It's
certainly how we do it to develop MLton.  It's really not too onerous.
These days, with MLton's nice new front end, I do most type checking
with MLton.  Then, I build with SML/NJ for quick testing, while
running a self compile on another machine.

> Yes, in industrial practice MLton could only be used as a 'final
> stage' optimizer.  Large programs require modules that can be
> separately compiled, otherwise you wait around all day for things to
> compile.

We've worked very hard on MLton to make things much better than all
day.  A self compile on a non-exotic (1.6GHz) machine takes <10
minutes.  MLton is a 128-thousand line program, so things are really
very livable with 10K and 20K line programs, which in SML can do quite
a lot.

As you can see from the thread re: HOL, we're still working on
improving compilation of large programs.  But I don't see anything
unsolvable there.

I wouldn't argue that MLton is livable on its own, but it's certainly
quite usable with another system around for fast testing.