Leveraging Light-Weight Analyses to Aid Software Maintenance

Abstract

Software maintenance can account for up to 90% of a system's life cycle cost. As a result, many automated techniques have been developed to reduce the overall effort necessary to sustain software over time. While many of these tools work well under certain circumstances, we believe that they could be improved by taking advantage of large untapped sources of unstructured information that result from natural software development. As the size and complexity of systems increases, taking advantage of such previously-overlooked information is of paramount importance.The proposed research will design lightweight analyses to extract latent information encoded by humans in software development artifacts and thereby reduce the costs of software maintenance. We will design analyses that apply throughout the maintenance process, focusing on three areas: (1) triaging large collections of exposed defects; (2) guiding the search for automatic defect repairs; and (3) ensuring the continued consistency of system documentation over time.For each area of proposed research we aim to improve upon existing approaches to reduce the total cost of software maintenance while requiring minimal additional developer effort. We will evaluate the proposed work using these criteria on a diverse set of real world programs totaling millions of lines of code to concretely show the reduction of maintenance cost and improvement upon existing processes.