Contents 1 A Constraint - Satisfaction Framework for Evaluating Program

Di erent program understanding algorithms often use di erent representational frameworks and take advantage of numerous heuristic tricks. This situation makes it is di cult to compare these approaches and their performance. This paper addresses this problem by proposing constraint satisfaction as a general framework for describing program understanding algorithms, demonstrating how to tranform a relatively complex existing program understanding algorithm into an instance of a constraint satisfaction problem, and showing how this facilitates better understanding of its performance. Plan recognition is the task of interpreting the actions of agents in the environment, in the context of the knowledge we possess about how action occurs in the world, and why. The recognition task involves constructing a mapping, possibly partial, between an existing repository of plan and domain knowledge and a set of dynamic observations of a subset of the actions taken toward a goal. Program understanding can be viewed as a special case of plan recognition, where the task is to recognize the plans programmers have used in constructing a particular piece of legacy source code. However, program understanding di ers from generalized plan recognition in that a complete set of action observations is the basis of goal determination. This paper discusses, in detail, how this di erence leads to inadequacies in applying typical plan recognition algorithms to program understanding. Program understanding can instead be viewed as a special case of plan recognition which is particularly amenable to constraint satisfaction techniques. 1