A Comparison of Three Occur-Check Analysers

A well known problem of most Prolog interpreters and compilers is the lack of occur-check in the implementation of the uniication algorithm. This means that such systems are unsound with respect to rst-order predicate logic. Static analysis ooers an appealing approach to the problem of occur-check reduction, that is, how to safely omit occur-checks in uniication. We compare, for the rst time, three static methods that have been suggested for occur-check reduction, two based on assigning \modes" to programs and one which uses abstract interpretation. In each case, the analysis or some essential part of it had not been implemented so far. Of the mode-based methods, one is due to Chadha and Plaisted and the other is due to Apt and Pellegrini. The method using abstract interpretation is based on earlier work by Plaisted, SSndergaard and others who have developed \groundness" and \sharing" analyses for logic programs. The conclusion is that a truly global analysis based on abstract interpretation leads to markedly higher precision and hence fewer occur-checks at run-time. The analysis is a compile-time investment, and given the potential run-time speedups, a precise analysis would be well worth the extra time. Abstract A well known problem of most Prolog interpreters and compilers is the lack of occur-check in the implementation of the uniication algorithm. This means that such systems are unsound with respect to rst-order predicate logic. Static analysis ooers an appealing approach to the problem of occur-check reduction, that is, how to safely omit occur-checks in uniication. We compare, for the rst time, three static methods that have been suggested for occur-check reduction, two based on assigning \modes" to programs and one which uses abstract interpretation. In each case, the analysis or some essential part of it had not been implemented so far. Of the mode-based methods, one is due to Chadha and Plaisted and the other is due to Apt and Pellegrini. The method using abstract interpretation is based on earlier work by Plaisted, SSndergaard and others who have developed \groundness" and \sharing" analyses for logic programs. The conclusion is that a truly global analysis based on abstract interpretation leads to markedly higher precision and hence fewer occur-checks at run-time. The analysis is a compile-time investment, and given the potential run-time speedups, a precise analysis would be well worth the extra time.