🚀 go-pugleaf

In article <983@cresswell.quintus.UUCP> ok@quintus.UUCP (Richard A. O'Keefe) writes:
>in order to perform optimisations like this
>automatically a compiler needs to be licensed to treat predicates
>according to their declarative reading rather than according to the
>strict behaviour of a Prolog interpreter.

NU-Prolog allows procedures to be declared "pure" for precisely this
reason.  One of the problems with Prolog as a logic programming language
is that it is impossible to avoid over-specification when writing a
program.  Whatever logic you write also has a fixed procedural
interpretation.  This means that even with very smart compilers,
parallel machines etc, the scope for optimization is quite limited.
It also means that programmers must consider the control component of
their programs.

Logical NU-Prolog constructs normally have both declarative and
procedural meanings: "A;B" means "A or B" declaratively and "execute A
and, on backtracking, execute B" procedurally.  If a procedure is
declared pure, only the declarative meaning is defined.  Currently, the
transformations done by the compiler are quite simple.  However, the
compiler can be extended and the declarations can be used for other
tools which transform programs (eg, adding control information, magic
sets, fold/unfold,...).

One of the problems with the idea is that we cant have 100% confidence
in such a system.  The undecidability of the halting problem means that
no matter how smart logic programming systems are, programmers will
still need to consider procedural factors in some cases.  I think the
difference between logic programming and automated theorem proving is
that in logic programming, programmers are responsible for infinite
loops.

	Lee Naish