Developing Finite Domain Constraints — A Data Model Approach

Abstract

We describe a technique for formulating a problem for solution by a finite domain constraint solver, where the finite domains can be modelled in correspondence with an Entity-Relationship diagram or UML Class diagram. This works particularly well where data for the problem is retrieved from database(s) over a network, but we believe the modelling discipline will be more generally useful. We show how relationships are conveniently represented using the infers operator of the generalised constraint propagation (Propia) library of ECLiPSe. Further, we can then express sets of quantified constraints over the data model in the declarative Colan language, and use this to generate equivalent ECLiPSe code directly. The user then has only to maintain the declarative version of the constraints, which are much easier to read. They can also be reused in many ways by fusing them with constraints from other sources, as in the KRAFT project. An important subclass of such constraints behave as conditional constraints which need delayed application, and we discuss experience in making such constraints more active in the solving process.