Stanio, not only are you right, I think you even know that you are right ;-)
We are still collecting material for some documentation. We are also planning to write a whitepaper that is more concerned with the design ideas. But time is really the issue. Christian On 4/3/08 12:54 PM, "[EMAIL PROTECTED]" <[EMAIL PROTECTED]> wrote: >> let me first elaborate a little bit how Gecode's variables and propagators >> are structured before I actually attempt to give an answer. >> >> Gecode has three important concepts regarding variables: >> - variables: they are used for modeling, and the only operations they >> provide are good for modeling. In particular, >> variables to not provide operations to update their domain. Examples are >> IntVar, BoolVar, etc. Variables are >> hence provide nothing but an interface to variable implementations >> (actually, variables are implemented as >> a pointer to variable implementations). >> - variable implementations: they are the real objects storing the domain >> and which propagators and advisors must be >> executed when the domain changes. They themselves are constructed in two >> steps: a domain independent stub that is >> automatically generated from a simple description (just a few lines). The >> stub is concerned with events, how >> propagators can be subscribed to, and linking the variable >> implementations to the Gecode kernel etc. The real >> variable implementation then inherits from the generated stub and adds >> the variable domain and so on. >> - views: like variables they are nothing but an interface for variable >> implementations. However, views are used >> by propagators and branching and provide operations to update variable >> domains, subscribe propagators to changes, >> and so on. More importantly, for the same variable implementation several >> views exist: a MinusView for a variable >> implementation x, for example, provides an interface that behaves as if >> the operations on the MinusView are executed >> on -x rather than x. Examples are IntView, BoolView, MinusView, >> OffsetView, ... For more on views there is paper >> on the Gecode webpage. >> >> Propagators themselves have two parts: a post function and the actual >> propagator. A post function (such as rel, distinct, linear, etc.) is used >> for modeling and hence takes variables as arguments. It then chooses the >> right propagator and the right views (converting variables to views by >> passing the underlying reference to the variable implementation). >> >> Propagators are typically parametric (as templates in C++) with respect to >> the view they are using: for example, the Max propagator is instantiated for >> the max posting function with an IntView whereas it is also instantiated >> with a MinusView for the min posting function. > >> Sorry for the long blurb, >> [...] > > I find such _blurb_ very useful, even only as a user. When > one knows more about the design of the system, this can > prevent many conceptual mistakes. I don't remember to have > read these explanations on the Gecode website. You might > save a bit of your time, if you put things like that in > kind of FAQ. (Or does it exist?) > > Certainly all this can be extracted from the publications > referred from gecode.org, from the doxygen docs and from the > sources, but I find the above condensed manner of > presentation very suitable for a plain text in a FAQ or > similar. > > :o) > Have a nice day. -- Christian Schulte, web.ict.kth.se/~cschulte/ _______________________________________________ Gecode users mailing list [EMAIL PROTECTED] https://www.gecode.org/mailman/listinfo/gecode-users
