Path: utzoo!attcan!uunet!munnari!murtoa.cs.mu.oz.au!munnari.oz!lee
From: lee@munnari.oz (Lee Naish)
Newsgroups: comp.lang.prolog
Subject: Re: Committed Choice
Message-ID: <1305@murtoa.cs.mu.oz.au>
Date: 20 Mar 89 02:08:47 GMT
References: <11500012@hpldola.HP.COM> <733@gould.doc.ic.ac.uk>
Sender: news@cs.mu.oz.au
Reply-To: lee@munmurra.UUCP (Lee Naish)
Organization: University of Melbourne, Comp Sci Dept
Lines: 53

In article <733@gould.doc.ic.ac.uk> cdsm@doc.ic.ac.uk (Chris Moss) writes:
>There are 2 reasons for choice of control structures in Prolog:
>efficiency and clarity. Certainly from a theoretical point of view
>(the "power" of the system) if then-else is adequate, but there are
>lots of reasons why it's given problems. See for instance the appendix
>in Ehud Shapiro's book "Algorithmic Program debugging" which uses 
>conditionals far more than his later book "Art of Prolog". (He had an 
>argument with David Warren and lost!).

As I understand it, the reason why the argument was lost was that *in
DEC-10 Prolog* '->' is not compiled and is therefore much slower.  I
don't know of any other Prolog system where this is the case.  In fact,
-> is significantly more efficient than ! in some systems for some
simple conditions.  For example:  ( A < B -> ... ; ...) does not create
a choice point in Quintus or NU-Prolog.  The equivalent code using cut
does (this may not be the case forever).

>Well for one thing there are different variations even of if-then-else!
(and cut, as Chris has pointed out in the past)
>For example, does an if-then without an else fail or succeed. It appears
>that the Europeans on the standardization committee think it should
>succeed: in "Edinburgh" systems it fails.

I am embarassed to say that the Melbourne group only discovered this
last year.  In MU-Prolog and (still) in NU-Prolog (fail -> xyz)
succeeds (I hope this will be changed to the standard behaviour some
day).  I suggest that people using {MN}U-Prolog avoid using -> without
the else part.

While I think its not such a good idea to change the "standard"
behaviour at this stage, I think the whole construct must have been
designed after a good many pints of 70 shilling on Grassmarket:

1) -> in logic means implication, or if-then, so the syntax is
  suggestive of logic.  However, in logic false->xyz is true!
2) Making -> fail when the condition fails make -> without ; virtually
  useless.  a->b is the same as once(a),b which is much more clear.  The
  only common use is (a->b;c->d;otherwise->e), which behaves the same
  way with either semantics for ->.  Also if -> succeeded when the
  condition failed "otherwise->" could be dropped, making the code
  simpler and more efficient.
3) With the non-standard semantics -> is very useful for checking error
  conditions and other exceptions: (var(X) -> error("Nonvar expected")),...
4) Why use ; for else?  It overloads ; and makes its definition (which
  could be nice, simple and logical) into a non-logical mess (remember
  that a->b;c is ((a->b);c)).
5) What is cut meant to do inside a condition?  This is a great way to
  break Prolog systems and when -> was defined the use or scope of !
  inside contitions should have been restricted (the problem is that the
  -> creates a choice point, ! removes it, then -> tries to cut back to
  it).

	Lee Naish