A detailed algorithm testing guards over feature trees

Abstract

We give the detailed formulation of an algorithm testing guards over feature trees. Such an algorithm is needed in constraint logic programming with entailment-based coroutining and in concurrent constraint programming; feature-tree constraint systems model extensible record descriptions (for example, in LIFE). We call this algorithm “Beauty-and-Beast” algorithm because it uses, for each variable X and for each guard test being suspended on X, a special-purpose data structure (the “beast”) which encodes the data relevant for that guard test (essentially, a unifier). A variable X has a multiple-binding list where each of its bindings to a beast is indexed by the particular guard test. This is how we achieve that the algorithm is incremental when it is applied for repeated resumptions of suspended tests. Its online time complexity is almost-linear (which is the same as for the best existing offline algorithms for the problem). Our implementation technique of multiple-bindings is applicable to all constraint systems over trees (finite trees, rational trees, etc.); the beast construction is motivated in particular by the implementation of record-descriptions with extensible lookup-tables.

This work was done while both authors were at Digital Paris Research Laboratory. A short description appeared in [17].