[gecode-users] wrapping SetVar

[Guido Tack]

> SetView::variable(void) works for me!

Ok, good. The only problem is that you can't get the pointer back to  
something you can use... You'd have to add another constructor to  
SetView that takes a SetVarImp*.

>
>>
>>> // this function could be called something like
>>> // LubRanges_from_SetVar
>>> {
>>> ...
>>> SetVar* set;
>>> SetView view(*set);
>>> return new LubRanges ranges(view); // (this would not work I  
>>> suppose....)
>>> }
>>
>> This will work!
>
> :) And this one two!
>
> But concerning the SetView, when the function returns I think the  
> destructor is called and then it is deallocated from the stack,  
> right? This does no harm since the LubRanges is working on the  
> underlying implementation.

SetView and SetVar (and all other views and vars) are nothing but  
wrappers around a single pointer. This pointer is never affected by  
any allocation or deallocation of a var or view. That's why nothing  
goes wrong here. And the pointer is normally kept internal because  
you cannot do anything useful with it - except maybe for storing it  
somewhere.

> But in general, shouldn't I be able to create an Object like  
> SetView, or e.g
> Gecode::Int::OffsetView with a function and return it to Lisp  
> directly?
>
> I imagine the case, where I could pass an Int::OffsetView to a  
> propagator, freeing me from the need to create a new IntVar to only  
> represent a constant addition of 1 for example.
> Maybe I am even not supposed to do something like that?

No, you're not. You can't call a propagator with an OffsetView  
anyway! The posting interface for propagators only accepts SetVars  
(or IntVars or BoolVars). The offset is not represented in the actual  
variable data structures, only in the view wrapper. This is an  
internal mechanism for getting more propagators out of a single  
implementation (we wrote a paper about this, if you're interested in  
what can be done with views).

Cheers,
	Guido