Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!apple!olivea!samsung!munnari.oz.au!lee
From: lee@munnari.oz.au (Lee Naish)
Newsgroups: comp.lang.prolog
Subject: Re: Standards question: behavior of arg/3
Keywords: ANSI, standard
Message-ID: <5816@munnari.oz.au>
Date: 19 Oct 90 08:57:49 GMT
References: <9888@bunny.GTE.COM> <3992@goanna.cs.rmit.oz.au> <1920@tuvie> <4018@goanna.cs.rmit.oz.au>
Sender: news@cs.mu.oz.au
Reply-To: lee@munmurra.UUCP (Lee Naish)
Organization: Comp Sci, University of Melbourne
Lines: 52

> >	?- arg(N, [a,b,c], X).
> >makes perfect sense and has two solutions
> >	N = 1, X = a ;
> >	N = 2, X = [b,c] .
>
> Which Prolog systems have arg/3 defined in this way?

MU-Prolog 3.2db
1?- arg(N, [a,b,c], X).
 
N = 1, 
X = a  ;

N = 2,
X = [b, c]  ;

fail.
2?- 

As Richard pointed out, NU-Prolog delays this goal.  NU-Prolog is a WAM
based system and open codes arg/3 - it knows it is deterministic.
MU-Prolog is an interpreter (the implementation of arg/3 is
particularly inefficient - I put in a quick hack and never got around to
doing it properly, but MU-Prolog isn't a really "serious" Prolog system
anyway).

Richard also mentioned Sicstus failing with this goal.  I think its
unfortunate that Sicstus does not make more of its coroutining
facilities.  There lots of builtins which could be made to behave more
logically using these facilities.  One possible reason for not doing
this is to remain compatible with various other Prolog systems which
fail silently.  There is something to be said for this.  I have had the
"pleasure" of porting a 30,000 line program written in (an old version
of) Quintus Prolog to NU-Prolog.  One of the nastiest bugs was a call to
functor/3 which failed in Quintus because it was insuffiiently
instantiated, but delayed in NU-Prolog.  I'm not sure what the latest
version of Quintus does.  I hope more systems start producing error
messages in situations like this (you can always add a call to nonvar/1
if you really want failure).

Another way of solving the porting problem is to support two versions of
functor, arg, is, <, etc, etc, as IBM Prolog does.  You should
definitely do this for some builtins, such as the type tests (integer
etc).  However, I think Prolog vendors would be doing every?one a favour
if they made functor etc produce error messages when insufficiently
instantiated.  I don't think it would hurt to put this into the
standard(s) either, even though most existing systems would not
conform.  I haven't seen any recent draft standards - do they discuss the
general class of errors called instantiation faults?  Richard's 1984 draft
did.

	lee