An Ecient Term Representation for CHR Indexing

The overhead of matching CHR's multi-headed rules is al- leviated by constraint store indexing. The attributed variable interface provides ecient means of indexing on logical variables. Current state- of-the-art indexing strategies for ground terms use hash tables. However, the hash tables incur considerable performance overhead, especially when frequently computing hash values for large terms. We propose a high-level approach which improves the eciency of ground term indexing. In this approach, we introduce a new data representa- tion for ground terms, inspired by attributed variables, that avoids the overhead of hash-table indexing. The experimental evaluation establishes the usefulness of our representation, but indicates a high cost of map- ping between this representation and Prolog's standard terms. Thus, we reuse previously implemented post-processing program transformations to compensate for this overhead. We compare our approach with the current state of the art, and give measurements of its eectiveness in the