[gecode-users] Variable update problem

[Guido Tack]

Jérémie Vautard wrote:
> I am coding a little stuff that needs to extract some potential  
> partial solutions from a constraint problem. To do this, I have a  
> (pointer to a)Gecode::Space S with an array of IntVar, and I do the  
> following :
> - I have initially propared a set of IntVarArgs, each containing a  
> given subset of the variables of S.
> - I create S2 to be a clone of S ( via S2 = S->clone(); )
> - I choose one of the IntVarArgs I initially created, let's call it  
> V, and I create a new IntVarArgs V2(V);
> - I update all the variables of V2 so that they "become" variables  
> of S2 : for (int i=0;i<V2.size();i++)   V2[i].update(S2,true,V[i]);
> - I post a branching in S2 with the variables of V2. e.g. :   
> Gecode::branch(S2,V2,INT_VAR_SIZE_MIN,INT_VAL_MIN);
>
> However, when I search on S2 (with the DFS engine), the returned  
> solution do not have the corresponding variables domain reduces to a  
> singleton. (I check this via an IntVar public member of my  
> Gecode::Space descendant class that I did not forget to update in  
> the copy constructor :-) )
>
> I join an small example that illustrates my issue : for this problem  
> with one variable and no constraints, I obtain 2 solutions (this  
> number is correct), but, for both of them, the 'v' member is still  
> [0..1]...
>
> I most likely have missed something, but I don't understand what....

The problem is that you cannot use update outside the Space's copy  
constructor (or any method it calls in turn).  Copying uses forwarding  
pointers and other "magic" behind the scenes, so when you clone a  
space, some clean-up is done after all the updates.
I think for the problem you're describing, you might want to use  
arrays of indices for describing the subsets instead of using  
IntVarArgs directly.  Then you don't have to mess with update, but you  
can just create an IntVarArgs and initialize it directly from the  
IntVarArray in the space, using the indices.

Cheers,
	Guido

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.ps.uni-sb.de/pipermail/users/attachments/20090127/907767a6/attachment.htm>