Using an artificial intelligence approach to build an automated program understanding/fault localization tool

Artificial intelligence techniques and architectures have played a large role in the design of a blackboard-based program understanding/fault localization tool we have been developing. We focus on a system knowledge source called the plan processor which will have artificial intelligence support for two of its major tasks. One task is to retrieve a set of program plans from a plan library using indices called signatures. To make this retrieval task more effective we propose using a genetic algorithm. We also describe a fuzzy reasoning component which supports the plan processor with a second task; ranking the retrieved plans in order of similarity to the target code. The most similar plan is then used for the complex plan/code matching required for automated program understanding. Our approach may eliminate the need for exhaustive plan library searches, and could lead to automated program understanders that scale up for use on software systems from a variety of problem domains.