#= may fail in one direction but not the other

[jeshan]

I've found a case where A#=B fails while B#=A does not. The issue is more readily observed with a custom propagator:

:- use_module(library(clpz)).

:- multifile clpz:run_propagator/4.

try :-
    A+B #= _,
    clpz:make_propagator(my_propagator(A), Prop),
    clpz:init_propagator(C, Prop),
    clpz:init_propagator(B, Prop),
    % clpz:trigger_once(Prop),
    write(start),
    nl,
    C = 0.                      % nontermination

my_propagator_impl(A, _) :-
    clpz:fd_get(A, _, fd_props(_,[propagator(pplus(_,B_,_),_)],_)), % trying to access the plus constraint created above
    _ #= B_. % swap sides to observe difference

% also this:
clpz:run_propagator(my_propagator(Arg), MState) -->
    { my_propagator_impl(Arg, MState) }.

I run with:

sicstus -l demo.pl --goal "try."

swap _ #= B_ with B_ #= _ to observe that the issue is gone.

What gives? Should I not use #= in a custom propagator?

Also: note that C isn't even useful in this very simplified scenario

[UWN]

The issue is more readily observed with a custom propagator:

Is this also observed just with library(clpz) alone?

[jeshan]

Yes, I have updated the code above. you meant you weren't able to reproduce?

[UWN]

I read your sentence "The issue is more readily observed with a custom propagator" as implying that you have a direct clpz program (without any propagator) that still exposes this problem. (And I would be very interested to see it)

[jeshan]

understood. I meant to say there probably exists a simpler scenario as I thought that I had found one earlier. I tried to simplify it further to remove the propagator but couldn't.