[gecode-users] Problem with constraint

[Ujhelyi Zoltan]

Hello again,

I have been experimenting with the questions raised in my previous mail.

I created a small Gecode space to be able to test the problem in a  
smaller scale, where I did use two variables (variable1, variable2,  
both IntVars with the domain {0,5}) and three constraints: variable1  
equals variable2, and variable2 cannot have the values 2 and 4.

In all of my tests the latter two constraints are described like that:

post(space, new BExpr(space.variable1, IRT_NQ, 2));
post(space, new BExpr(space.variable1, IRT_NQ, 4));

For the third constraint I had several tries. When I was using the  
rel() clause to write the constraint, I got the expected result:

Variable #1: {0..1,3,5}
Variable #2: {0..1,3,5}

But when using a similar construct I would like to use in the static  
verifier, I got no result for the second variable, it can has all  
values from it's domain.

post(space, new BExpr(space.variable1, IRT_EQ, space.variable2));

Reading about different propagators, I have been able to get the  
expected result with the use of the ICL_DOM propagator, but only if I  
was posting the constraint directly, like that:

post(space, new Expr(space.variable1), IRT_EQ, new  
Expr(space.variable2),org.gecode.GecodeEnumConstants.ICL_DOM);

But if I tried to translate the a BExprs to that form, I did not  
succeed:

	    Expr expression = new Expr(new  
Expr(space.variable2).minus(space.variable1));
	    post(space, new Expr(expression), IRT_EQ, 0,  
org.gecode.GecodeEnumConstants.ICL_DOM);

Can you tell me a way how to enable this strong ICL_DOM propagators  
meanwhile using the BExpr form, if it is possible? It would help our  
project a lot, it would simplify it's code.

Thank you for your help,
Zoltán Ujhelyi
On 2008.08.09., at 18:30, Ujhelyi Zoltan wrote:

> Hello all,
>
> we are currently trying to build a static checker system on top of
> Gecode/J 2.1.1. (in an Eclipse environment), and we have found an
> interesting case.
>
> In order to allow us greater flexibility we try to use the Expression
> API (it allows us to build constraints in a hierarchical manner).
>
> I was testing the implementation, and found an interesting final
> result: there were to IntVars, which were supposed to be the same, did
> not have the same final ranges: one of them had {2,4}, but the other
> had {2,3,4}. I tried to trace back the code to look whether the
> constraints were wrong or not, and found the following expression
> building term:
>
> BExpr expression = new BExpr(new Expr(variable1),relation, new
> Expr(variable2));
>
> Tracing the variables variable1 was a new variable, it hadn't been
> constrained yet, while variable2's domain had been limited to the
> values 2 and 4 (and they were the variables mentioned before in this
> letter), while the relation had the value IRT_EQ.
>
> I tried a post of the expression right after it's construction to see
> whether it helps or not:
> org.gecode.Gecode.post(space, expression);
>
> It did not solve the problem, but when I replaced the post method to
> the following rel call, the constraint handling was working as I
> expected it before.
>
> rel(space, variable1, IntRelType.IRT_EQ, variable2);
>
> I would like to know, whether this phenomenon is the intentional
> (maybe I do not understand something under the hood), or not. And if
> you could tell me a way to avoid it, I would be very glad.
>
> Thank you very much in advance for your help,
> Zoltán Ujhelyi
> _______________________________________________
> Gecode users mailing list
> users at gecode.org
> https://www.gecode.org/mailman/listinfo/gecode-users