fluet

[Stephen Weeks]

> He's arriving on the 29th.  I was thinking of having him visit
> you for a week starting June 12th.  Is that a good time for you?

That sounds good.

> Can you take care of hotel accomodations?

In the sense of making them, yes.  In the sense of paying for them,
I'll have to look into it.

> Before then, we should agree on what he's supposed to do and
> how.  I was thinking he should work on a linear-scan 
> register allocator before doing code generation, but it may be
> feasible to punt register allocation for now (stack allocate
> everything), concentrating on code generation instead.  Let
> me know your thoughts.

I would strongly vote for punting register allocation and for the main
goal of the summer being to produce something that works ASAP,
especially for a large program like a self-compile.  It should be
possible to re-use a lot of the current stuff (stack allocator, data
representation, liveness analysis) and get something so we can make a
pretty direct comparison to the C backend.  In fact, it might even be
possible to use the same MACHINE signature and backend.fun that is
there right now -- just spit out machine code instead of C.

On a related point, I vote for generating machine code instead of
assembly.  We've already seen how slow the assembler is for large
programs (about half the time of the C compile IIRC).  I would hate to
just push the bottleneck down a little bit.

BTW, let's continue discussions on the MLton mailing list.  That way
Henry can chime in.  Also, go ahead and add Fluett if he's not there
already.