Hmmm, CMUCL does it already in the compiler and since its compiled code is more debuggable than XEmacs interpreted code, I feel that I can point to it as an example. See http://www.mindspring.com/~rtoy/software/cmu-user/node140.html#SECTION006500 00000000000000 The thing I like about scheme is that theres not defvar, defparameter, defconstant, defun. Theres just define. Well thats true both ways. I beleive that either language would be cool, but theres going to be a bit of pain involved during the transition. :-) I agree with this on some points. For example, I wish that sequences were an object so that I could add new things with sequence semantics and have the existing common lisp sequence functions work on them. But, the things that have gone into the language are the result of many man-years of people using lisp derivatives and then making compromises so that people could port to it without dying from old age in mid-port. The problem with scheme is that while its a great abstraction building tool, the baseline may be a bit too low. I remember reading that Jeffrey Siskind moved to scheme and it took him five years to get his abstractions working to his liking. How long will it take me? I consider myself to be nowhere near his skill level. OK. I may lean more toward scheme. Here is what would make for a kick ass system. All IMHO of course. A module system like SML; Higher order modules included. CLOS like object system + a MOP like STk. With this you really don't need structures and you can use the MOP to properly expose the object-based C code in the lower layers. A condition system with conditions being objects like Common Lisp. Generational garbage collection with excess space being given back to the OS after a long period of non-use. I like to run my emacs for months on end without restarting it. :-) A fast scheme engine. This is the tough one. Scheme48 is currently a bit too slow for this. Of course clisp has a lot of hand coded C/assembly so maybe making a better prescheme compiler would do the trick. I really like the concept of scheme48 so it would be great to devise some way to get large speed benefits while retaining all of the good layering stuff. Gambit is good, but goes to C not bytecode. Bigloo makes too many compromises to scheme for my taste especially wrt. call/cc. If you are going to use scheme, I would want call/cc, delay, force and the syntax-case macro system. Maybe you can talk Kent Dybvig into making Chez Scheme open source. :-) This probably runs into the realm of fantasy. On the other hand, you may have it all implemented on your home machine. -Reggie