PEREIRA@SRI-AI.ARPA (04/22/83)
I'm re-sending this article via a new route, to test it.
Could you please tell me when it arrives? I've sent it also
via UDEL -- Fernando
---------------
Date: Thu 21 Apr 83 09:21:41-PST
From: PEREIRA@SRI-AI
Subject: more "Prolog compared with Lisp"
To: bts.unc@UDEL-RELAY
Richard O'Keefe from Edinburgh mentioned the lamentable Gutierrez article
and his reply on SigPlan, of which I have only seen an early version.
For the benefit of those on this newsgroup that have seen neither article,
and that may be inclined to infer from the exchanges that this is yet
another "holy war", I will give my own view of the debate
to the best of my recollection. Needless to say, I agree with Richard!
But first of all a bit of history.
In the mid-seventies, when David Warren designed the first version of
the Edinburgh compiler, nobody would believe that Prolog (after all
just a "dumb theorem prover") could be used efficiently for practical
programming tasks. We then ran some simple benchmarks between simple
(almost) pure Lisp and pure Prolog (ie. no assert & friends), for some
simple tasks such as list concatenation, sorting and ranking and
differentiation. We reported the results of these experiments
in "Prolog - the language and its implementation compared with Lisp"
which appeared in SigPlan of Aug. 77. In the area of performance
comparison, that paper concluded the speed of pure Prolog programs
compiled with that version of David's compiler was with a factor of
two of that of analogous as pure as possible Lisp programs compiled
with the Lisp 1.6 compiler. Since then, there has been a new improved
Prolog compiler that makes the same sort of programs comparable
with their counterparts compiled with the MacLisp compiler.
Gutierrez's task was some sort of theorem prover. It was implemented
in Lisp using a rather low level style (ie. SETQ, RPLAC? and loops
instead of recursion and consing). His Prolog version is possibly
the worst written Prolog program I have ever seen. The whole
operation of the program is made of 'assert' and 'retract' and
other evaluable predicates, with hardly any declarative content.
That is the program is a benchmark of the Prolog database operations,
not of Prolog. In fact, I was surprised that such a program was only
twice as slow as its Lisp counterpart!
One writes nondeclarative Prolog programs for various reasons:
(1) poor understanding of logic programming
(2) not knowing how to formulate the whole task in pure Prolog
(3) circumventing limitations of the Prolog system or machine
(4) expressive limitations of definite clauses
Gutierrez's program seems to be the result of (1) and (2). I have
variously suffered from the four complaints. As with any other
language, it takes a long time for anyone to develop a style that fits
the "expressive profile" of the language, and clearly Gutierrez hadn't
been through that.
--- And now a message from our sponsor ----
Let me use this article to push one of my pet concerns: before
you use those 'assert's, or even cut, think, THINK VERY HARD.
You may well be rewarded with a much cleaner and enlightening
description of the task.
--- end of message ---
The original comparison paper was necessary because few people
accepted that Prolog was a useful and efficient language. Today,
such comparisons are no longer needed: witness the number of people
putting up with rather slow interpreters, people who use Prolog primarily
for its expressive power, not for speed. For the future, the important
benchmarks and studies of efficiency are those between different
implementation techniques for Prolog and logic programming in general,
and those related to new computer architectures for logic programming.
Fernando Pereira
-------
-------