Maclisp revisited.

[gjc%bucsf.BU.EDU@bu-it.BU.EDU (George J. Carrette)]

Through the intervention of the Macsyma->Lisp translator which I did
extensive work on and with in the days when the Macsyma Consortium had
hundreds of active users, I can say I produced my share of heavy
numerical code usage in Maclisp. In fact, we Plasma Fusion hackers
knew well the trade of between a quick-turn-around hack in maclisp or
macsyma and a heavy run in Fortran on the CRAY-1 at the MFE center at
Lawrence Livermore Labs. It was not unusual for a person using that
KL-10 with Maclisp and a knowlege of applicable numerical programming
techniques to come up with a physically meaningful result before the
guy in the next office using the CRAY-1 was able to. Of course, one would
hope that we always made sure the results checked before going to publication.

It is indeed sad but true that no machine/lisp implementation I have used
in the ten years since has quite matched that performance. Lispmachines
came very close, but that is another story.

I disagree strongly however that it is actually DIFFICULT to get good
number complication in Lisp. Not great, just good. Maclisp was good,
not great, and that worked out just fine. The amount work supposedly
put into the Maclisp compiler is not a valid measure of how much work
it would be to do something equivalent today.

Anyway, in a couple months DOE-Macsyma will be producing as good numerical
code in a lisp enviroment (in lisp of course) as in the PDP-10 days.
There actually are some decent numerical lisp's out there.

-gjc