[Yap-users] Strange tabling behavior with dif/2

[Edward S. <esc...@ce...>]

Hi all,

I am experiencing some strange behavior when I combine tabling and dif/2.

Here's a very simple prolog program for computing transitivity:

% 1, 2 and 3 are on the same class.
same(1, 2).
same(1, 3).
% 4, 5, and 6 are the same class.
same(4, 5).
same(4, 6).
:- table ruleSame/2.
% Root from real facts.
ruleSame(X, Y) :-
    same(X, Y).
% Commutative.
ruleSame(X, Y) :-
    dif(X, Y),
    ruleSame(Y, X).
% Transitive.
ruleSame(X, Y) :-
    dif(X, Y),
    dif(Y, Z),
    dif(X, Z),
    ruleSame(X, Z),
    ruleSame(Z, Y).

Basic facts are ok:

   ?- ruleSame(1,2).
yes
   ?- ruleSame(1,3).
yes
   ?- ruleSame(2,3).
yes

But then some unexpected facts are generated...

   ?- ruleSame(2,4).
yes

If I debug the evaluation of these facts, I get an extremely long trace.

If I move the dif sub-goals after the ruleSame sub-goals, the program
works as expected.

Is this a bug?  I think it is, but hopefully I haven't overlooked
something.  It seems like maybe the deferred goals from dif/2 are
interacting with tabling in a non-trivial way.

I tested this both on YAP 6.2.2 and the current development git branch
of 6.3.  Both had the same behavior.

Thanks for any help,

Ed