Software is everywhere in our modern technological society. People’s daily lives have become increasingly dependent on complex software systems and the results of software failures have a huge social and economic impact. Research indicates that software failures cost the American economy nearly $60 Billion annually. Clearly, an investment in software quality and dependability will pay off.
One proposed solution is automated software inspection: the use of advanced program analysis techniques to automatically examine a software system’s source code to verify that it complies with coding standards and warn upon detection of potential problems. This allows early and repeated quality assessments during development, and enables early corrections, when changes are still relatively inexpensive. As a result, the software system’s quality and reliability increase and the overall development costs decrease. Despite these promises, companies are reluctant to adopt these techniques, mainly for two reasons: (1) missing evidence: the fact if automated software inspection could have prevented known post-release defects has never before been thoroughly and systematically investigated; (2) the nature and amount of generated warnings causes information overload, which greatly reduces the technique’s effectiveness and may lead to downright rejection, especially in cases where the first warnings turn out to be not useful (e.g. false positives or trivial due to overly tight coding standards).
This project addresses these issues in two consecutive steps: (1) we will empirically assess the cost-effectiveness and efficiency of automated software inspection techniques in reducing actual field defects; and (2) we will advance the state of the art by devising novel prioritization techniques that improve existing automated software inspection techniques via extensions to focus the analysis in such a way that the most effective warnings will get higher priority.
