Things that are not in the ISO core standard are more
likely to behave differently across Prolog systems?

GNU Prolog behaves already differently to SWI-Prolog in
numbervars/3. I find this discrepancy:

/* GNU Prolog 1.5.0 */
?- X = f(A,X), numbervars(X, 0, _), write_canonical(A), nl.
_24
cannot display cyclic term for X

/* SWI-Prolog 9.1.16 */
?- X = f(A,X), numbervars(X, 0, _), write_canonical(A), nl.
'$VAR'(0)
X = f(A, X),
A = A.

LoL

So what goodies are there in library(compat) as well?

Check this out:

/* SWI-Prolog 9.1.16 */
?- numlist(1,100000,A), numlist(1,100000,B),
time((between(1,1000,_), subsumes((f(_),X,X),(_,A,B)),
fail; true)), fail.
% 2,999 inferences, 2.375 CPU in 2.498 seconds (95% CPU, 1263 Lips)
false.

/* Dogelog Player 1.1.3 for Java (new!) */
?- numlist(1,100000,A), numlist(1,100000,B),
time((between(1,1000,_), subsumes((f(_),X,X),(_,A,B)),
fail; true)), fail.
% Zeit 2 ms, GC 0 ms, Lips 1550000, Uhr 18.10.2023 17:40
fail.

Dogelog Player for Java came out a few days ago. To get
subsumes/2 you have to use ensure_loaded(library(compat)).
I am using Richard O’Keefes algorithm from 1984 for subsumes/2.

Mild Shock schrieb am Mittwoch, 18. Oktober 2023 um 14:59:26 UTC+2:
> Things that are not in the ISO core standard are more
> likely to behave differently across Prolog systems?
>
> GNU Prolog behaves already differently to SWI-Prolog in
> numbervars/3. I find this discrepancy:
>
> /* GNU Prolog 1.5.0 */
> ?- X = f(A,X), numbervars(X, 0, _), write_canonical(A), nl.
> _24
> cannot display cyclic term for X
>
> /* SWI-Prolog 9.1.16 */
> ?- X = f(A,X), numbervars(X, 0, _), write_canonical(A), nl.
> '$VAR'(0)
> X = f(A, X),
> A = A.
>
> LoL

I am not extremly convinced by Richard O'Keefes approach
for subsumes/2. Especially how I have implemented it in
Dogelog Player, avoiding term_variables/2 all togtether!

Possibly I can produce some drastic test cases where it
doesn't run very well. On the other hand Jan W. is conceptually
promoting this bootstrapping of subsumes/2:

subsumes(General, Specific) :-
term_variables(Specific, SVars),
General = Specific,
term_variables(SVars, SVars).

The above cannot fast fail like Richard O'Keefes approach
can do. The further boostrapping is identical:

subsumes_term(General, Specific) :-
\+ \+ subsumes(General, Specific).

Mild Shock schrieb am Mittwoch, 18. Oktober 2023 um 17:56:52 UTC+2:
> So what goodies are there in library(compat) as well?
>
> Check this out:
>
> /* SWI-Prolog 9.1.16 */
> ?- numlist(1,100000,A), numlist(1,100000,B),
> time((between(1,1000,_), subsumes((f(_),X,X),(_,A,B)),
> fail; true)), fail.
> % 2,999 inferences, 2.375 CPU in 2.498 seconds (95% CPU, 1263 Lips)
> false.
>
> /* Dogelog Player 1.1.3 for Java (new!) */
> ?- numlist(1,100000,A), numlist(1,100000,B),
> time((between(1,1000,_), subsumes((f(_),X,X),(_,A,B)),
> fail; true)), fail.
> % Zeit 2 ms, GC 0 ms, Lips 1550000, Uhr 18.10.2023 17:40
> fail.
>
> Dogelog Player for Java came out a few days ago. To get
> subsumes/2 you have to use ensure_loaded(library(compat)).
> I am using Richard O’Keefes algorithm from 1984 for subsumes/2.
> Mild Shock schrieb am Mittwoch, 18. Oktober 2023 um 14:59:26 UTC+2:
> > Things that are not in the ISO core standard are more
> > likely to behave differently across Prolog systems?
> >
> > GNU Prolog behaves already differently to SWI-Prolog in
> > numbervars/3. I find this discrepancy:
> >
> > /* GNU Prolog 1.5.0 */
> > ?- X = f(A,X), numbervars(X, 0, _), write_canonical(A), nl.
> > _24
> > cannot display cyclic term for X
> >
> > /* SWI-Prolog 9.1.16 */
> > ?- X = f(A,X), numbervars(X, 0, _), write_canonical(A), nl.
> > '$VAR'(0)
> > X = f(A, X),
> > A = A.
> >
> > LoL

But Jan W. conceptualization can be used to develop
a funny variant/2 predicate. If we start from this here,
we will have two unifications, in each subsumes_term/2:

variant2(A,B) :-
subsumes_term(A,B),
subsumes_term(B,A).

We can also use only one unification:

equalizer(A, B) :-
term_variables(A, L),
term_variables(B, R),
A = B,
term_variables(L, L),
term_variables(R, R).

variant3(A, B) :-
\+ \+ equalizer(A, B).

Seems to work:

A B variant2(A,B) variant3(A,B)
f(X,g(Y,X),Y) f(Z,g(T,Z),T) true true
f(X,g(Y,X),Y) f(Z,g(Z,Z),T) false false
f(X,g(Y,X),X) f(Z,g(T,Z),T) false false
f(A,A,B) f(B,A,A) false false
X+Y Y+Z false false

Mild Shock schrieb am Donnerstag, 19. Oktober 2023 um 19:02:23 UTC+2:
> I am not extremly convinced by Richard O'Keefes approach
> for subsumes/2. Especially how I have implemented it in
> Dogelog Player, avoiding term_variables/2 all togtether!
>
> Possibly I can produce some drastic test cases where it
> doesn't run very well. On the other hand Jan W. is conceptually
> promoting this bootstrapping of subsumes/2:
>
> subsumes(General, Specific) :-
> term_variables(Specific, SVars),
> General = Specific,
> term_variables(SVars, SVars).
>
> The above cannot fast fail like Richard O'Keefes approach
> can do. The further boostrapping is identical:
>
> subsumes_term(General, Specific) :-
> \+ \+ subsumes(General, Specific).