Constraint Propagation and Explanation over Novel Types by Abstract Compilation

Abstract

© Graeme Gange and Peter J. Stuckey. The appeal of constraint programming (CP) lies in compositionality - the ability to mix and match constraints as needed. However, this flexibility typically does not extend to the types of variables. Solvers usually support only a small set of pre-defined variable types, and extending this is not typically a simple exercise: not only must the solver engine be updated, but then the library of supported constraints must be re-implemented to support the new type. In this paper, we attempt to ease this second step. We describe a system for automatically deriving a native-code implementation of a global constraint (over novel variable types) from a declarative specification, complete with the ability to explain its propagation, a requirement if we want to make use of modern lazy clause generation CP solvers. We demonstrate this approach by adding support for wrapped-integer variables to chuffed, a lazy clause generation CP solver.