Let me add to my previous mail that I've now upgraded from 18d to 18d+ (built 2002-09-13) and the situation has significantly improved. While the CLOS-generated closures are still somewhat slower the difference isn't as important anymore - it's 3 to 4 percent as opposed to the 60 percent I experienced with 18d. That's certainly something I can live with.
Also, the whole program is noticeable faster with 18d+ - my hat's off again to the CMUCL developers. Thank you! The bad thing is that the disassemblies of CLOS-generated closures now look completely different than those of closures created by normal functions and they're also _much_ longer, i.e. I'm really confused - but that's probably not that important... :) (I'd still be interested in explanations if someone wants to shed light on this issue.) Cheers, Edi.
