Experiences in using Prolog to develop a practical constraint solver

Abstract

Sophisticated Prolog implementations like ECL/sup i/PS/sup e/, with automatic memory management and facilities for data-driven computation, offer many advantages for software development. The question is whether such automated systems can be used to deliver applications with a practical performance. We discuss our experiences with implementing an incremental version of the Simplex algorithm in ECL/sup i/PS/sup e/ (as part of a constraint logic programming system). Our application consists of approximately 2000 lines of ECL/sup i/PS/sup e/ Prolog code. In this paper, we present the highlights of our implementational experience and show how the application takes advantage of the meta-level control facilities and the garbage-collector. After taking performance profiles, we show how we modified the algorithm to increase performance. In previous systems, such as CHIP, CLP(/spl Rscr/) and Prolog-III, the solver was written in C, following the conventional wisdom that this is necessary to achieve good performance. We show that intricate coding is not necessary, since automated tools such as ECL/sup i/PS/sup e/ can already be used to produce realistic applications (and ECL/sup i/PS/sup e/ is being improved on continuously). Our results are quantified using a series of benchmarks against existing C-coded implementations. The Simplex solver is being used within Esprit project CHIC to develop financial applications.<>