[gecode-users] Question about branching

[Vincent Barichard]

Hi Christian,

Le vendredi 13 février 2009 10:32:36 Christian Schulte, vous avez écrit :
> Hi Vincent,
>
> would the following help: There will be two iterators Space::Propagators
> and Space::Branchings that iterate over all propagators or branchings of a
> space? Would be enough if they would just give const access?
>

A const access should be sufficient. It will be very convenient to be able to 
parse propagators and branchings :-)

> Your patch is wrong. The update function does everything thats needed. Or
> did you have a problem there?
>

Indeed, I encountered a problem. I designed a propagator which uses advisors. 
Sometimes afer cloning, Gecode crashed. The debugger tells me that the crash 
occured in the Council<A>::update function of my propagator and Valgrind said 
that there is some uninitialized variables. 
After reading the code, I saw that the field named "advisors" in the Council<A> 
class is set to NULL in only one constructor. Furthermore, there is no recopy 
constructor, so when cloning, a bad value may be copied in the "advisors" field 
of the clone, and the variable "c.advisors" in the Council<A>::update function 
may  not been correctly initialized.
After putting "co(home)" in the recopy constructor of my propagator, 
everything go well, so I thought it was the same problem for the 
SpeedLinBoolInt propagator. I probably thought wrong, sorry.

Cheers,
Vincent


> Cheers
> Christian
>
> --
> Christian Schulte, www.ict.kth.se/~cschulte/
>
>
> -----Original Message-----
> From: users-bounces at gecode.org [mailto:users-bounces at gecode.org] On Behalf
> Of Vincent Barichard
> Sent: Tuesday, February 10, 2009 10:03 PM
> To: Christian Schulte
> Cc: users at gecode.org
> Subject: Re: [gecode-users] Question about branching
>
> Hi,
>
> Le mardi 10 février 2009 14:52:37 Christian Schulte, vous avez écrit :
> > Hi,
> >
> > no there is no way to find the x, in particular as some branchings
> > might not have an array of variables.
>
> I wasn't sure but I suspected it.
>
> > There is not even the notion of a current branching (there are two
> > current
> > branchings: one that is used for status() and the other that is used
> > for
> > commit() during recomputation).
> >
> > Branchings are designed to just implement a particular interface that's
>
> it.
>
>
> It is not completely clear for me. I will look more further in the code.
>
> > Maybe you tell us what you want to do?
>
> I have created a new variable type and I want to do backjumping on the
> search tree. So I need to access to the variable of the BranchingDesc to
> know if I must to skip the BranchingDesc or not. After reading your mail, I
> decided to create sub-class of ViewValBranching and PosValDesc to get the
> information I need. Here comes a piece of the code I wrote in my search
> engine:
>
>               const BranchingDesc * bd;
>               int noAlt;
>               int noClosedDesc;
>               QBool::QBoolQuant quant;
>               do
>               {
>                 bd = rcs.nextDesc< QBool::QBoolPosValDesc<2>
>
> >(*this,noAlt,noClosedDesc);
>
>                 if (bd == NULL) return NULL;
>                 quant = static_cast<const QBool::QBoolPosValDesc<2>*>(bd)-
>
> >quantifier();
>
>               } while (quant != QBool::QBoolVarImp::EXISTS);
>
>               delete cur;
>               cur = rcs.recompute<false>(d,*this);
>               EngineCtrl::current(cur);
>
> Sometimes, the line:
> "              } while (quant != QBool::QBoolVarImp::EXISTS);"
> changes, so I have to dynamically ask the variable of the BranchingDesc
> what is its quantifier.
>
> The problem here is that I have to assume that it is a
> QBool::QBoolPosValDesc<2> BranchingDesc and to cast it.
> I would like to test if it is a QBool::QBoolPosValDesc<2> and if it is the
> case, to test its quantifier.
>
> > Then, reflection will be gone (see my next mail as to why) and there
> > is currently no way to get the actors (unless you iterate over the
> > list of actors, check how the function propagators() counting the
> > number of propagators is implemented).
>
> I looked at propagators(), but It needs "pc" being public or protected in
> order to use it in our subclass of Space.
>
> Another thing, I am not sure it's a small bug or not, but I found that the
> "co" field of SpeedLinBoolInt is not initialized during cloning. As a
> result, the update function of Council<A> may not ends. I attached a patch
> in the mail.
>
> Thank you for your help.
> Cheers,
> Vincent

-- 
Vincent Barichard        Université d'Angers (LERIA)
Tel: 02 41 73 52 06      Département Informatique
Fax: 02 41 73 50 73      H203