Strange warning and computation result in constraint logic program - prolog

Question

First, sorry for posting the whole program, but as I don't know were the problem is I don't know which parts are irrelevant. These are two slightly different implementations of the same logic puzzle in SWI-Prolog, the first one succeeds the second one fails and I can't find the reason for the failure.
The puzzle:
4 persons are having a diner:
Donna, Doreen, David, Danny
the woman (Donna,Doreen) are sitting vis-a-vis.
the men (David,Danny) are sitting vis-a-vis.
Each of them picked a unique meal and beverage.
1) Doreen sits next to the person that ordered risotto.
2) the salad came with a coke.
3) the person with the lasagna sits vis-a-vis the person with the milk.
4) david never drinks coffee.
5) donna only drinks water.
6) danny had no appetite for risotto.
who ordered the pizza?
I choose the following approach
table with positions:
1
4 O 2
3
domain: positions{1,2,3,4}
variables: persons, meals, beverages
First the inefficient succeeding implementation:
solution(Pizza, Doreen, Donna, David, Danny) :-
% assignment of unique positions to the variables
unique(Doreen,Donna,David,Danny),
unique(Lasagna,Pizza,Risotto,Salad),
unique(Water,Coke,Coffee,Milk),
% general setting
vis_a_vis(Donna,Doreen),
vis_a_vis(David,Danny),
% the six constraints
next_to(Doreen,Risotto),
Salad = Coke,
vis_a_vis(Lasagna,Milk),
\+ David = Coffee,
Donna = Water,
\+ Danny = Risotto.
unique(X1,X2,X3,X4) :-
pos(X1),
pos(X2),
\+ X1 = X2,
pos(X3),
\+ X1 = X3, \+ X2 = X3,
pos(X4),
\+ X1 = X4, \+ X2 = X4, \+ X3 = X4.
right(1,2).
right(2,3).
right(3,4).
right(4,1).
vis_a_vis(1,3).
vis_a_vis(3,1).
vis_a_vis(2,4).
vis_a_vis(4,2).
next_to(X,Y) :- right(X,Y).
next_to(X,Y) :- right(Y,X).
pos(1).
pos(2).
pos(3).
pos(4).
This works and gives the right result. But when I try to reorder the clauses of the solution procedure to be more efficient (this is the second implementation)
solution(Pizza, Doreen, Donna, David, Danny) :-
% general setting
vis_a_vis(Donna,Doreen),
vis_a_vis(David,Danny),
% the six constraints
Salad = Coke,
vis_a_vis(Lasagna,Milk),
\+ David = Coffee,
Donna = Water,
\+ Danny = Risotto,
% assignment of unique positions to the variables
unique(Doreen,Donna,David,Danny),
unique(Lasagna,Pizza,Risotto,Salad),
unique(Water,Coke,Coffee,Milk).
%% all other predicates are like the ones in the first implementation
I get a unassigned variable warning when trying to load the file:
Warning: /home/pizza.pl:28:
Singleton variable in \+: Coffee
and the computation returns false. But shouldn't it return the same result?
I see no reason for the difference...

Answer 1

the warning is due to the fact that Coffe and Risotto are unbound when the negation is executed. If you replace \+ David = Coffee, by David \= Coffee, you will avoid the warning, but the solution cannot will not be computed. Should be clear indeed that since Coffee is unbound, David \= Coffee will always fail. You can use dif/2, the solution will work and will be more efficient. I've named solution1/2 your first snippet, and solution2/5 this one (using dif/2):
solution2(Pizza, Doreen, Donna, David, Danny) :-
% general setting
vis_a_vis(Donna,Doreen),
vis_a_vis(David,Danny),
% the six constraints
next_to(Doreen,Risotto), % note: you forgot this one
Salad = Coke,
vis_a_vis(Lasagna,Milk),
dif(David, Coffee),
Donna = Water,
dif(Danny, Risotto),
% assignment of unique positions to the variables
unique(Doreen,Donna,David,Danny),
unique(Lasagna,Pizza,Risotto,Salad),
unique(Water,Coke,Coffee,Milk).
a small test:
?- time(aggregate_all(count,solution1(P,A,B,C,D),N)).
% 380,475 inferences, 0.058 CPU in 0.058 seconds (100% CPU, 6564298 Lips)
N = 8.
?- time(aggregate_all(count,solution2(P,A,B,C,D),N)).
% 10,626 inferences, 0.002 CPU in 0.002 seconds (100% CPU, 4738996 Lips)
N = 8.