Benchmarking CLOS

[pab@starbase.mitre.org (Paul Birkel)]

Yes, I know this is a topic fraught with contention (if not down-right
obfustication). But I'm interested in comments, anyway.

I'm new to CLOS, and am evaluating a commercial CLOS implementation called
"MOS" from AI Technology. It is not a PCL derivative, and currently is only
available on/under VAX/VMS. The Sun3 port is nearing completion. One important
aspect of the product for us is its meta-class extension which integrates
(among other things) CLOS objects with a production engine.

I'm reasonably satisfied with the production engine aspect of things, and
I'm also satisfied with the rationale for deviations ("enhancements") of
MOS from CLOS. I'm not sure yet how to best evaluate the performance of
MOS itself.

AI Technology has implemented MOS on top of the existing VAXLisp (and
soon, Lucid) compilers. Therefore they do not claim to have an "optimal"
implementation; their benchmarks (very limited) are with respect to (1)
object creation in PCL, and (2) defstruct creation, ... period. They
intend to move onto a compiler-incorporated CLOS as such become available,
so most efficiency issues will eventually be passed to the compiler designer.

In the meantime, the major issue for me is whether they stack up reasonably
well "across-the-board" against PCL, and decently against compiler-specific
comparable defstruct capabilities.

So, please help me out with some ideas (code?) for "across-the-board"
capabilities which are "must-haves" (and vs. "nice-to-haves").

Thank you.

Paul A. Birkel
MITRE, Mailstop W418
7525 Colshire Drive
McLean, VA  22102-3481

(703) 883-6399