Path: utzoo!utgpu!jarvis.csri.toronto.edu!clyde.concordia.ca!uunet!mcsun!ukc!icdoc!qmw-cs!mmh
From: mmh@cs.qmw.ac.uk (Matthew Huntbach)
Newsgroups: comp.lang.prolog
Subject: Re: Legitimate uses of "assert"?
Message-ID: <1708@sequent.cs.qmw.ac.uk>
Date: 28 Feb 90 12:11:29 GMT
References: <8205@pt.cs.cmu.edu>
Reply-To: mmh@cs.qmc.ac.uk (Matthew Huntbach)
Organization: Computer Science Dept, Queen Mary and Westfield College, U of London, UK.
Lines: 41
Summary:
Expires:
Sender:
Followup-To:
Distribution:
Keywords:

In article <8205@pt.cs.cmu.edu> fp@PROOF.ERGO.CS.CMU.EDU (Frank Pfenning) writes:
>I am looking for papers/reports/code/information on "legitimate" uses
>of "assert" and "retract" in logic programming, that is, uses which are
>considered to be in good taste by the connoisseurs of the field.

"assert" and "retract" are best used to get round what I
believe is a very awkward problem with Prolog, the fact that
computations on the OR-branches of search trees cannot
otherwise communicate.

e.g

pred(<args1>) :- body1
pred(<args2>) :- body2

Suppose body1 fails, but some of the partial calculations done
within it are useful in body2. Then you have to use "assert"
to pass them on.

I still don't think this is in good taste. I would prefer
something like:

pred(<args>) :- body1',body2'.

where body1' passes on any information required to body2' and
has a flag which if set to "true" causes body2' to immediately
halt.

This is, of course, an argument against backtracking in Prolog,
and an argument for the non-backtracking languages like Parlog.

Relevant paper:
"Compiling OR-parallelism into AND-parallelism"
Codish & Shapiro, 3rd Int. Conf. Log. Prog. 1986.

A good example of this is something like branch-and-bound
search where body1 might produce bounds which restrict the
amount of search done in body2. An unpublished paper of
mine discusses this in great detail.

Matthew Huntbach