On Wed, Mar 13, 2002 at 04:13:34PM -0500, Anthony Hannan wrote:
> Lee Salzman <lee.salzman at lvdi.net> wrote:
> >    Squeampiler is a completey independent
> > codebase (written completely in Squeak and for Squeak, of course). 
> > However, it is vastly more portable. It does ALL optimization
> > at the level of a target independent representation (even register
> > allocation), and does a number of cool optimizations that were absent
> > in the translator that are found in the industrial grade compilers,
> > and will do even more when it is done. If someone would like to 
> > collaborate with me on making it more suitable for dynamic compilation
> > (brainstorming, code, whatever), I'd appreciate the mindshare. :)
> 
> Sounds interesting.  I use an intermediate representation also in the BC
> compiler before generating bytecodes.  It helps me find shared vars
> between contexts/closures.  I also use it for a couple of simple
> optimizations.  Maybe we can share some code ideas.  Maybe we can add
> block closures to your compiler or we can add native code generation to
> mine (or some combination there of).  You can find my code on the swiki
> at http://minnow.cc.gatech.edu/squeak/BlockClosureVersion.
> 
> Cheers,
> Anthony
> 
> PS. By the way, the block closure page is the first one listed under
> Google when searching for "block closures".  Cool!
> 
    Well, currently Squeampiler is in no way intended to obviate a lot
of the existing compilation mechanisms. If anything, it is intended to
complement them, though. :) However, making Squeampiler do some closure
optimizations would be cool, but would mean I'd have to make it more
knowledgeable of much higher level details (although I've been pondering
this lately for other things).

    Lee