A common sort of work in programming languages takes the form “concept X meets concept Y and here are the consequences.” In this case, X is bound to “constraint languages” and Y is bound to “object-oriented programming.” Constraints are defined as formulas relating components of objects, which has the desirable consequence of not requiring any sort of primitive constraint entities. Steels introduces the notion of constraint managers; they are analogous to metalevel interpreters in Prolog, since managers determine how the process of constraint satisfaction will happen. A single system could have several kinds of managers; one does simple propagation, another uses relaxation, while a third takes tasks from an agenda. Managers can also be built into a hierarchy, in which higher-level managers decide about invocation of lower-level managers. All constraint managers are objects with their own environments and defining functions, which opens up the possibility of having inheritance among types of managers.

The entire paper is rather sketchy and does not do justice to the interesting ideas that are proposed. An example of cryptarithmetic is only partially solved, which left me wondering if the described language is powerful enough to do the whole solution. The paper is also quite specialized, but it should be of interest to the small but growing constraint language community, as well as to those looking for new ideas in language design.