Algorithmic debugging of prolog side-effects

Abstract

Prolog side-effects are typical necessary evil. Convenient logic program properties no longer hold, and upsetting practical problems sprout, namely the debugging of programs with side-effects is harder. This paper attempts to reduce the magnitude of such problems, by formulating a framework for the algorithmic debugging of logic programs with side-effects, encompassing both input/output (I/O) and internal database (DB) ones. Our debugging framework [Pereira Calejo 88] is reviewed and extended for that purpose. The new methods extend the scope and differ, significantly, from those developed in the context of [Pereira 84 86].

In what refers DB side-effects, we extend our suspect sets to include additional suspects: those contained in the partial derivations leading to ‘assert’ and ‘retract’ calls. For I/O we define an additional type of bug related to the I/O side-effect notion of segment (or trace) of a goal, and introduce a corresponding additional type of suspect set. This approach allows us to apply the same diagnosis algorithms already developed for programs without side-effects. We make some reasonable assumptions in either case, so as to render the implementation amenable and to avoid extorting excessive information from the user.

These extensions are being implemented in our HyperTracer debugger [Pereira Calejo 89a, 89b], incurring in tolerable overhead, mostly confined to actual use of the extensions.

UUCP...{mcvax,inria}!inesc!un!mc Bix: mcalejo Applelink:IT0083