Experimental evaluation of a generic abstract interpretation algorithm for Prolog

Abstract

The practical value of research involving the abstract interpretation of Prolog programs was examined experimentally. The design and implementation of the generic abstract interpretation algorithm originally proposed by B. Le Charlier (1991), its instantiation in a sophisticated abstract domain containing modes, types, sharing and aliasing, and its evaluation in terms of performance and accuracy are described. The overall implementation (over 5000 lines of Pascal) was systematically analyzed on a variety of programs. The experimental results, given the abstract domain and the programs analyzed, indicate that: the number of iterations of the algorithm is bounded by 7.5*N and is in most cases smaller than 3*N, where N is the size of the analyzed program (e.g. the number of program points); the CPU time in seconds is bounded by N and is in most cases smaller than 0.6*N; the algorithm explores few elements (less than 11% and often none) outside the subset of the fixpoint required to answer the query and hence is close to optimality; and the results are quite accurate and could be used in a Prolog compiler. >