Peyton-Jones

[Stephen Weeks]

> I propose that we tell him we'll start working on it, but that we
> can't commit to a fixed timeline.  At the very minimum, Henry and I
> (and possibly Matthew) can devote some number of cycles to this so
> that Stephen can work on other stuff, and act as a reviewer
> initially.

OK.  Go ahead and send him mail, and please cc mlton.  If you guys
actually write some stuff, I will certainly read it, and may be
inspired to write more.  Have a look at opus/notes for my initial
ideas on how the paper should go.  

My thoughts as to the main approach of the paper are that it should
neither be a formal paper with lots of ILs and transformations nor a
benchmark paper with lots of static counts, dynamic counts, and
analysis and run times.  It should be a paper about how whole program
compilation allows one to translate SML to a very simple IL, SSA, and
do most optimizations there.  I think the paper should show lots of
example SML code, and the corresponding dot graph, and show how
optimizations of SSA got us the optimizations we need.

The only numbers that I think we need to show are some performance
numbers (compile time, run time, code size) just to prove it all works
and blows away other compilers.  I would put those up front, as
motivation to read the rest of the paper.