How People Debug, Revisited: An Information Foraging Theory Perspective

Many theories of debugging have been proposed over the years, but most were formulated using very small programs, in an era of programming environments that were much simpler than today’s environments. Further, because these theories rely on complex mental constructs such as hypotheses, they offer little practical advice to builders of software engineering tools. In this paper, with a practical perspective in mind, we reconsider the notion of how people go about debugging in today’s large collections of source code using a modern programming environment. Specifically, we present an information foraging theory of debugging-focused navigation that describes behavior in terms of environmental cues. The theory treats programmer navigation relative to a bug as being analogous to an animal using scent and topology when navigating in the wild in search of prey. It further proposes that these constructs of scent and topology provide enough information to describe and predict programmers’ navigation during debugging, without reference to mental states such as hypotheses. To evaluate this premise, we used an executable model of the theory to predict ten professional programmers’ navigation of real-world open source programs while they debugged, and analyzed the results in light of what these programmers talked aloud about while they debugged. We found that the programmers’ verbalizations as they debugged were overwhelmingly about following scent, pervasively across six debugging modes, much more than their verbalizations of (non-scent) hypotheses. We also found that, although the programmers used numerous resources, source code was the artifact in which most scent processing took place, suggesting that relatively inexpensive algorithms based on static analysis of source code alone are viable in supporting programmers’ information seeking activities as they debug. In fact, scent and scent plus topology were both more effective predictors of programmers’ navigation than the programmers’ stated hypotheses, suggesting that reasoning about artifacts in the environment is not only a more practical mechanism in practice than drawing from verbalizations of mental state, it may also be more accurate. Finally, we discuss our results’ implications for ways to enhance software engineering tools according to information foraging principles.