Meta-interpreting !/0

[triska]

Dear all,

as hinted in #2720 (comment), several tools may benefit from being able to meta-interpret !/0, at the very least for analytic purposes and maybe in the future also for automated suggestions and rewrites using more recent constructs with more desirable algebraic properties.

I think it is possible to meta-interpret !/0 with the standard control constructs catch/3 and throw/1. Declaratively, !/0 is true, and any faithful emulation must succeed exactly once.

I therefore suggest the following predicate to emulate !/0 with exceptions:

cut.
cut :- throw(cut).

This succeeds, and throws an exception on backtracking.

We can catch the exception with catch/3. For example:

?- catch(((X=1;X=2),cut,member(Y,"abc")), cut, false).
   X = 1, Y = a
;  X = 1, Y = b
;  X = 1, Y = c
;  false.

This suggests, as a rough outline for how meta-interpretation could be performed:

mi((Before,!,After)) :-  % needs work
        catch((mi(Before),cut,mi(After)), cut, false).

I hope this helps!

All the best,
Markus

[jjtolton]

What part needs work? Looks good to me!

[triska]

There are two remaining problems:

The first, and that is what the comment meant to indicate, is due to the defaulty representation of Prolog code, meaning that Before can either be an atomic goal or a goal that itself contains control structures (such as conjunction). For instance, Before may be a term like (G1,!,G2). This mean that either the representation must be fixed (using or converting to a clean representation) or the Before part must be more carefully analyzed to recognize where the first !/0 in the body actually occurs.

The second issue is that the MI now itself throws and catches the ball cut/0, which may incorrectly interpret Prolog programs that use this term for their own exceptions.

[bakaq]

For the second issue, you could meta-interpret exceptions. This obviously adds a lot of complexity though.