ADAPT: Automated debugging in an adaptive Prolog tutor

Abstract

Automated program debugging is a difficult task which requires recognizing the algorithm employed by the programmer and detecting any errors that exist in her program. All approaches to automated program debugging require some form of plan library which consists of a collection of representative programs. The general approach to constructing these debuggers is to build a system with an initial set of representative programs, run empirical studies on novice programmers to test the sufficiency of this set of programs, and then add new representative programs and repeat the empirical studies. Because the decision to add new representative programs to the library is dictated by empirical studies, the set of programs in the library has no precise characterization. It is possible to precisely define a non-trivial class of recursive list processing Prolog programs which empirical studies have shown to be sufficient for novice Prolog programmers. When supplemented with a general equivalence-preserving transformation scheme, this class of programs can be represented by a very small set of representative normal form programs which is bounded by the number of inputs in the program. The resultant debugger is more robust than APROPOS2 [9] and provides a precise characterization of the class of programs it can recognize.