A subtle difference between XSB and SWI-Prolog about mode directed tabling `po(PI)`

Discussion
, ,
1

Firstly, the document of po(PI) is confusion.

po (PI)

Partial Ordering . The new answer is added iff call(PI, +Old, +Answer) succeeds. For example, po('<'/2) accumulates the largest result. In SWI-Prolog the arity (2) may be omitted, resulting in po(<) .

IMO, the po('<'/2) should accumulate the smallest result instead of the largest result.

For example,

:- table
  path(_,po('<'/2)).

path(X,P) :-
  (   X=1,P=1
  ;   X=1,P=3
  ).

?- path(X,P).
X = P, P = 1.

Clearly, it is accumulating smallest result, otherwise the result should be X=1,P=3.

Secondly, there is a subtle difference between XSB and SWI-Prolog about mode directed tabling po(PI).

For example,

:- table
  path(_,po(myls/2)).

myls(1,2).
myls(3,4).

path(X,P) :-
  (   X=1,P=1
  ;   X=1,P=2
  ).

In SWI-Prolog

?- path(X,P).
X = P, P = 1.

In XSB

| ?- path(X,P).
X = 1
P = 1;
no

That is correct.

However, the following example,

:- table
  path(_,po(myls/2)).

myls(1,2).
myls(3,4).

path(X,P) :-
  (   X=1,P=1
  ;   X=1,P=3
  ).

In SWI-Prolog

?- path(X,P).
X = 1,
P = 3.

In XSB

In XSB
| ?- path(X,P).

X = 1
P = 3;

X = 1
P = 1

I think the XSB result is correct, because po(PI) means partial ordering, not total ordering. However, the po(PI) in SWI-Prolog means total ordering.

2

That is just a documentation typo, no? Thanks for spotting. Fixed.

This seems really fundamental. Answer subsumption was supposed to be a deterministic process. In this example it is not. I don’t think that is easily fixed in the current design for answer subsumption and thus I’m afraid this should be on the pile of remaining open issues :frowning: Of course, unless someone fixes it or makes the resources available to get this fixed.

1 Like