caroma@ai.mit.edu (01/02/91)
In article <95437@aerospace.AERO.ORG> srt@aerospace.aero.org (Scott "TCB" Turner) writes: >I don't know off-hand if anyone has brought AKCL up on a NeXT, but if >they have I'd recommend it over T. AKCL is an excellent >implementation of Common Lisp, very correct and fast. I've programmed >the same large system in both T and AKCL, and I much prefer AKCL. >There are things I like about T, particularly the object-oriented >aspects, but it is just too slow and kludgy compared to AKCL. Could you characterize your application and platforms a little more, so readers can get a better idea of what your datapoint represents? I have an application which is VERY call intensive, so I wrote some little test programs which basically test a few calling strategies. On these (little test) programs, T came out on top about 10 times faster than AKCL; AKCL was at the bottom of the heap of the Lisps I tested, including T 3.1, Lucid, IBCL, and AKCL (yes, with safety=0, speed=3). I tested on Sun3's, which are 68030's I believe, and therefore might have similar performance on NeXT's. Admittedly, these were small programs. Also, T seemed to invoke the GC alot, and slowed down quite a bit when it did so. So AKCL may be better for some systems; I'm just countering the impression that the AKCL would be faster that T for all applications. For the original poster's AI class, AKCL would probably be better if it can be found just because it is Common Lisp, which is a widely available standard so the exercises and the student's work will be portable. T is a dialect of Scheme, so it would take a bit of work to convert programs to T from CL or vice versa, but it would be better for learning Scheme (popular for CS Lisp programming courses). Cheers, caroma@ai.mit.edu