A Generative Constraint Formalism for Configuration Problems

Traditionally, constraint satisfaction systems have been considered an especially well-suited representation to configuration problems. However, a conventional constraint system with a predefined set of variables does not capture the flexibility inherent in composing systems out of a multitude of components of varying types. We propose an extended constraint satisfaction scheme that allows the incremental extension of a constraint network in accordance with the component-oriented view of configuration. Components can be individually represented and connected, while resource constraints express non-local requirements on the interaction of components. Constraints may be generative in that they lead to introduction of new variables, and are generic in that they may be defined to hold for all components of a given type.