(...)/0 as not implemented

[hurufu]

Is it a good idea to add built-in predicate (...)/0 that will just throw an exception analogously to Perl? It plays nicely with top-level answers, because with such predicate disjunction A=0; A=1; (...). is a valid program and has expected behavior.

%% (...).
%
% Throws not implemented error.
%
% Given a program:
%     q :- (...).
%
% Following query produces:
%     ?- q.
%        error(existence_error(procedure,_A/_B),[culprit-(...)]).
%
% Also you can use it in DCGs:
%     d --> {...}.
%
(...) :-
    throw(error(existence_error(procedure, _/_),[culprit-(...)])).

I have no idea what to throw in this case though, but term _/_ implies that procedure doesn't exist and it is unknown – which kinda makes sense for "not implemented" error.

[triska]

It's already there, isn't it?

?- ... .
   error(existence_error(procedure,... /0),... /0).

[hurufu]

Yes and no, you can say same thing for every undefined predicate, I thought about explicitly defined predicate that throws an exception.

[UWN]

It plays nicely with top-level answers, because with such predicate disjunction A=0; A=1; (...). is a valid program and has expected behavior.

Maybe what you mean is that an answer should always be usable as a query such that we get again that answer back. So that answers are effectively fixed points. This does hold in many situations, but it cannot hold everywhere.

Think of the ellipsis used to abbreviate terms or side effects.

?- length(L,21).
   L = [_A,_B,_C,_D,_E,_F,_G,_H,_I,_J,_K,_L,_M,_N,_O,_P,_Q,_R,_S,_T|...].
?- put_char(x).
x   true.

The ... is just an answer description. And for side effects we do need some extra formalism. In particular outputs/1.

?- put_char(x).
   outputs("x"), true.

but term _/_ implies that procedure doesn't exist

No. It is just a non-conforming term. 7.7.7 b error says it must be a predicate indicator of the current activator which in this case would be ... /0.

(Not sure where the round brackets you use come from, they suggest that ... is an operator.).

[hurufu]

Thanks, so a more systematic approach is needed (?) also with regard to output and simple ... doesn't add much to what we have already.

but term _/_ implies that procedure doesn't exist

No. It is just a non-conforming term. 7.7.7 b error says it must be a predicate indicator of the current activator which in this case would be ... /0.

This was my worry – that term isn't conforming – so can't be included as built-in.

(Not sure where the round brackets you use come from, they suggest that ... is an operator.).

I have my library of different predicates I just copy-pasted from it – so no mystery here :) not_implemented.zip And it is defined as operator there.

[UWN]

that term isn't conforming – so can't be included as built-in.

I cannot follow that conclusion. The term _/_ just does not make any sense in that context. Nothing can be concluded from that w.r.t. validity of a definition. It might make sense to prevent the accidental definition of ... by just providing a dummy definition. Like for (:-)/2). Not sure.

And it is defined as operator there.

That is definitely a bad idea. Just as bad as, say, making one of the control constructs that are atoms an operator. Like true, !, fail, false.

[hurufu]

I would intuitively think that predicate indicator in form _/_ inside error term can be interpreted as: "predicate that was meant to be called, is yet to be implemented and we don't know the name or arity of it, so anything can unify with it in principle...". But I have never read the spec its all based on my gut feeling.

Regarding ... operator – it hasn't bitten me yet, but yeah it might surprise me in a sufficiently large Prolog program.

[UWN]

Regarding ... operator – it hasn't yet bitten me, but yeah it might surprise me in a sufficiently large Prolog program.

It should have. If you paste back the answer to repeat, you should have received a syntax error, because an atom that is itself an operator cannot be an operand.

?- op(100,fx,...). % not recommended!!
   true.
?- repeat.
   true
;  ... .
?- true ; ... .
error(syntax_error(incomplete_reduction),read_term/3:1).