Toward a Feasible Functional Logic Compiler Employing Fair Search

Traditionally, implementations for functional logic languages (FLP) have used depth-first search to implement narrowing, a well-known search strategy. While this approach yields good performance, it is unnecessarily restrictive in the sense that some programs lead to non-termination (the search computes forever in one part of the search space, failing to consider the rest) when clearly they should yield results. As a result, a number of recent FLP research efforts have focused on designing efficient FLP systems based on breadth-first search strategies, which hold an important fairness property: if a solution is known to exist, it will (eventually) be found. The work presented in this paper addresses issues of efficiency in the context of FLC, a FLP compiler that employs fair search and a novel application of first-class stores in its implementation of narrowing. We show that in order to be tractable, FLC needs an efficient data representation for first-class stores and an efficient garbage collector. We discuss and compare several data representations for first-class stores, and show that a hashing-based data structure exhibits desirable performance. We show that proper collection of an efficient first-class store data representation in FLC involves additional work in the collector, to collect some reachable values which are known to be garbage. We describe the implementation of a complete generational copy collector that correctly handles collection of first-class store data structures. Finally, we evaluate the performance of the complete system, showing favorable results.