DCG translation not steadfast

Was running this test case:

p --> \+ q.
q --> {write('did you call me?'), nl}.

In Tau Prolog 0.3.2 (beta) I get:

p([a],[]).
false.

On the other hand SWI-Prolog gives me:

?- p([a],[]).
did you call me?
false.

[triska]

@gilbert: “steadfast” is hard to define in general. A “classical” example of a definition that is not steadfast is usually given for max/3, which we can wrongly define as:

max(X, Y, X) :- X > Y, !.
max(_, Y, Y).

With this definition, which seems to make intuitive sense, we have:

?- max(1, 0, X).
   X = 1.

It works as expected, right? But on the other hand, the maximum of 1 and 0 is unexpectedly 0 if we ask like this:

?- max(1, 0, 0).
   true.

The predicate is not steadfast in the sense that, by instantiating arguments in the query, we can trick it to yield an answer that we actually did not intend to define. The (1980’s) textbook way to solve this issue in this concrete case is to delay the unification of X and the third argument to occur only after the !/0:

max(X, Y, Z) :- X > Y, !, X = Z.

It is clear that nobody wants to write such code and that moreover, nobody wants to risk running into such unexpected problems, and we would therefore of course simply define, via a single clause:

max(X, Y, Z) :- Z #= max(X, Y).

and thus get a shorter and significantly more general definition, at least as we reason about integers, as is normally also the case.

In the case under discussion, “steadfastness” is, somewhat analogously, a guarantee that certain parts of the grammar rule are executed no matter the instantiation of arguments: We cannot “trick” the definition to yield unexpected results by instantiating parts of the query.

[Jean-Luc-Picard-2021]

Steadfast in connection with DCG is defined in the DCG Draft.

This document is available for everybody: https://www.complang.tuwien.ac.at/ulrich/iso-prolog/dcgs/dcgsdraft-2019-06-03.pdf

If we interpret execution as including side effects… but I didn’t check what 7.7.1 says.