Representation Sharing for Prolog

TL;DR

This paper explores representation sharing in Prolog, introducing practical methods like input sharing and a runtime sharer module, demonstrating their implementation in hProlog and analyzing their performance and application-dependent effectiveness.

Contribution

It presents new techniques for representation sharing in Prolog, including an easy input sharing method and a runtime sharer module adaptable to WAM-like systems.

Findings

The sharer has a cost comparable to heap garbage collection.

Effectiveness of sharing is highly application dependent.

Sharing policy needs tuning for optimal performance.

Abstract

Representation sharing can reduce the memory footprint of a program by sharing one representation between duplicate terms. The most common implementation of representation sharing in functional programming systems is known as hash-consing. In the context of Prolog, representation sharing has been given little attention. Some current techniques that deal with representation sharing are reviewed. The new contributions are: (1) an easy implementation of {\em input sharing} for {\em findall/3}; (2) a description of a {\em sharer} module that introduces representation sharing at runtime. Their realization is shown in the context of the WAM as implemented by hProlog. Both can be adapted to any WAM-like Prolog implementation. The sharer works independently of the garbage collector, but it can be made to cooperate with the garbage collector. Benchmark results show that the sharer has a cost comparable to the heap garbage collector, that its effectiveness is highly application dependent, and that its policy must be tuned to the collector. To appear in Theory and Practice of Logic Programming (TPLP)