How to avoid a meta argument warning in SICStus SPIDER?

This is probably related to a comp.lang.prolog-discussion.

I'm getting several warnings like this using Eclipse with the SICStus SPIDER:

The plain meta argument (Y) is passed as a closure argument
(with 0 suppressed arguments) to the callee.

Here is a code sample:

% Prologs set_of is baroque %% RS-140614  130sec runtime vs. 28sec runtime
:- meta_predicate set_of(+,:,+) .
set_of(X,Y,Z):-           %%
    setof(X,Y^Y,Z),!;     %% Trick to avoid alternatives
    Z=[].                 %% What is wrong with empty sets ?

How can I get rid of the SPIDER warnings? I'm not really interested in simply suppressing the warnings. I'm using the latest version of SPIDER IDE (0.0.51), and SICStus Prolog 4.2.3.

There are several issues in the code you show.

Bad meta argument

First, the built-in predicate setof/3 has the following properties:

?- predicate_property(setof(A,B,C),P).
   P = (meta_predicate setof(?,0,?))
;  P = built_in
;  P = jittable.

which closely corresponds to the ISO declarations in ISO/IEC 13211-1:

8.10.3.2 Template and modes

setof(?term, +callable_term, ?list)

The second argument is a goal to be executed by call/1. No extra arguments are needed. This is what the 0 tells us.

On the other hand, your code you show contains a different meta predicate declaration:

:- meta_predicate(set_of(+,:,+)) .

Here, the second argument is a :. In SICStus, YAP, and SWI, the : means: This argument will be automatically qualified with the current module, such that the module information can be passed further on. Think of asserta(:). Here, the argument is not a goal but a clause.

So what you need to fix this, is to replace : by 0. And you might indicate this fact in the variable name used. That is, Goal_0 for call(Goal_0), Goal_1 for call(Goal_1, Arg1), Goal_2for call(Goal_2, Arg1, Arg2) etc.

Bad modes

The + in the first and third argument is inappropriate. The 3rd argument is commonly an uninstantiated variable to be unified with the resulting list.

Prolog's setof/3 baroque?

% Prologs set_of is baroque

The comment probably wants to say that setof/3 contains superfluous ornaments. In fact, setof/3 is much more versatile than mentioned set_of/3. Take this recent question or that. Often you first think about a very specific situation. Say, you want the list of actors of a particular movie. Then, later on you want to ask what movies there are. It is this generalization which works very smoothly with setof/3 whereas it is extremely complex if you do not have it.

Another very useful way to use setof/3 is when you want to eliminate redundant answers:

?- (X=2;X=1;X=2).
   X = 2
;  X = 1
;  X = 2.
?- setof(t, (X=2;X=1;X=2), _).
   X = 1
;  X = 2.

Try to emulate that efficiently.

Runtime overheads

They are next to negligible. If you really believe that there are overheads, simply use setof/3 with a single goal. In this manner preprocessing is next to naught.