Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!linus!decvax!harpo!seismo!hao!hplabs!sri-unix!Narain@Rand-Unix
From: Narain@Rand-Unix@sri-unix.UUCP
Newsgroups: net.lang.prolog
Subject: Problem With "Not"
Message-ID: <4230@sri-arpa.UUCP>
Date: Mon, 15-Aug-83 19:41:00 EDT
Article-I.D.: sri-arpa.4230
Posted: Mon Aug 15 19:41:00 1983
Date-Received: Thu, 18-Aug-83 05:49:48 EDT
Lines: 36

In browsing through the Prolog utilities today I came across one
that advised using \+ instead of not ( or something ) and noticed
the following example:

bachelor(X):-not(married(X)),male(X).

If we add:

married(a).
married(b).
male(c).

then :-bachelor(Y). fails. This is what the example said or implied.
But if we rewrite the rule as:

 bachelor(X):-male(X),not(married(X)).

 then :-bachelor(Y). succeeds with Y=c.

I think this is serious since the order of literals should not affect
what the program computes, only its efficiency if at all.
The problem arises since at runtime the argument to "not" contains a
variable.  This changes the meaning of not(X) to:

 there exists a Y such that not(married(Y)). In which case Y may be
 bound to any element in the set:

Herbrand-Universe for the program - {Y such that married(Y) is
derivable} a potentially infinite set.

But the call to not(married(Y)) must not fail.

No doubt this has been observed before and I would greatly appreciate
any comments on how to avoid the problem.

-- Sanjai