[eclipse-clp-users] Bug in lex_lt with occurrences?

Hi!

I think I've encountered a nasty bug in ic_global.

Version 6.0 #199 (x86_64_linux), Fri Nov  2 22:47 2012

:- lib(ic).
:- lib(ic_global).
:- lib(ic_global_gac).

model_1(Curr, Next) :-
    length(Curr, N), length(Next, N),
    Next :: 0..9,
    ( for(I, 1, 9), param(Curr, Next) do
        occurrences(I, Curr, Di),
        occurrences(I, Next, Di) ),
    ic_global:lex_lt(Curr, Next),
    labeling(Next).

model_2(Curr, Next) :-
    length(Curr, N), length(Next, N),
    Next :: 0..9,
    ( for(I, 1, 9), param(Curr, Next) do
        occurrences(I, Curr, Di),
        occurrences(I, Next, Di) ),
    ic_global_gac:lex_lt(Curr, Next),
    labeling(Next).

model_3(Curr, Next) :-
    length(Curr, N), length(Next, N),
    Next :: 0..9,
    same(Curr, Next),
    ic_global:lex_lt(Curr, Next),
    labeling(Next).

model_4(Curr, Next) :-
    length(Curr, N), length(Next, N),
    Next :: 0..9,
    same(Curr, Next),
    ic_global_gac:lex_lt(Curr, Next),
    labeling(Next).


Model description:
Curr is a list of digits.
I want Next to be a list of the same length as Curr, with the same
number of non-zero digit, and Curr is lexicography smaller than Next.
Another formulation of the same model: Next is a permutation of Curr,
and Curr is lexicography smaller than Next.

I prepared 4 models:
1: ic_global:occurrences and ic_global:lex_lt
2: ic_global:occurrences and ic_global_gac:lex_lt
3: ic_global_gac:same and ic_global:lex_lt
4: ic_global_gac:same and ic_global_gac:lex_lt

I think all 4 models are equivalent.

But model_1 works only for some cases:

:- model_1([0, 5, 9, 4, 9, 1, 9], X).
X = [0, 5, 9, 4, 9, 9, 1]

:- model_1([0, 5, 9, 7, 7, 3, 3], X).
X = [0, 7, 3, 3, 5, 7, 9]

But for many other cases I get exactly the same list from model_1
despite lex_lt constraint:

:- model_1([0, 5, 9, 7, 7, 3, 2], X).
X = [0, 5, 9, 7, 7, 3, 2]

:- model_1([0, 4, 9, 7, 6, 3, 3], X).
X = [0, 4, 9, 7, 6, 3, 3]

So I think there is a bug in ECLiPSe when both ic_global:occurrences
and ic_global:lex_lt are used.

Sergey.