Towards evidence-supported, question-directed collaborative program comprehension

Many software maintenance and enhancement tasks require considerable developer knowledge and experience in order to be efficiently completed on today's large and complex systems. Preserving explicit forms of documentation that are accessible by large development teams with regular developer turnover is a difficult problem. This problem can result in temporal and spatial miscommunication, an easily lost cognitive work context, and largely unmaintainable software. The research described in this paper hypothesizes that the problem may be addressed by a semi-structured goal-question-evidence methodology for program comprehension that has three primary aspects. First, a redocumentation system should function in parallel with the development process by integrating into the user's usual tool environment and development workflow. Second, knowledge should be dispersed throughout a development team as soon as it is discovered so that comprehension is not merely confined to the mind of one individual. Finally, the developer should be made peripherally aware of their work objectives and the surrounding collaborative environment, reducing time spent on task reorientation, context reconstruction, and duplicative work. We present an observational study conducted on pair program comprehension and use the analyzed results to drive the formation of tool requirements for a collaborative comprehension tool. A prototype tool has been developed, showing promise for the methodology.