Meta-S - Combining Solver Cooperation and Programming Languages

Meta-S is a constraint solver cooperation system which allows the dynamic integration of arbitrary external (stand-alone) solvers and their combination with declarative languages. We sketch the main aspects of Meta-S including solver and language integration as well as its strategy definition framework for specifying solver cooperation and language evaluation strategies by means of an example. 1 Motivation Constraint solvers offer problem solving algorithms in an encapsulated way. Many solvers have been designed and implemented, covering several different constraint domains, for example finite domain problems, linear arithmetic or interval arithmetic. However, many real world problems do not fit nicely into one of these categories, but contain constraints of various domains. Writing specialized solvers that can tackle particular multi-domain problems is a time consuming and error prone task. A more promising approach is the integration of existing constraint solvers into a more powerful overall system for solver cooperation. Meta-S – our flexible and extendable constraint solver cooperation system with support for integration of arbitrary declarative languages – implements this idea. 2 Meta-S Constraint Solver Cooperation Figure 1 shows the general architecture of our constraint solver cooperation framework [3]. The system consists of several constraint solvers, each maintaining its own store. The collaboration is coordinated by the meta solver that establishes an exchange of information between the connected solvers. This meta solver maintains a pool of constraints to be solved. The communication between the meta solver and the individual solvers is done solely through two interface functions (readily supported by many preexisting solvers). A propagation function adds a constraint from the constraint pool into a solver’s store, hereby ensuring consistency of the resulting constraint store. The second interface function handles projection, i.e. it infers knowledge implied by a constraint store in form of constraints that are put into the pool and passed to other solvers. WCLP 2005 159 Ulmer Informatik-Berichte 2005-01