History-Based Automated Program Repair

点击翻译按钮获取中文摘要

Software reliability and security are grand and fundamental research challenges in information technology. Producing reliable and secure software is also critically important to Canada's economy and public safety. Software bugs significantly impair software reliability and security. They cost our economy billions annually. Many bugs, even those that are known, remain in mature software for a long time due to the lack of the development resources to fix them. Therefore, techniques to help developers fix bugs are in high demand. Automatically fixing bugs could: (1) reduce the cost of software development through saving programmers' time and effort in diagnosing and fixing bugs, and (2) improve software reliability and security through fixing more bugs earlier. The long-term goal of the proposed research program is to develop novel program repair techniques that generate fixes for real-world bugs automatically to reduce the cost of software development and improve software reliability and security. In the next five years, the goal is to generate one major type of fixes---recurring fixes---automatically. Recurring fixes are fixes that are identical or similar to another fix. Studies have shown that up to 45% of fixes are the recurring type. The specific objectives are (1) to obtain a deep understanding of recurring bug fixes, (2) to learn fix patterns from past fixes automatically to generate partial recurring fixes, and (3) to generate complete recurring fixes automatically by combining finer-grained fixes to form larger complete fixes. Existing approaches for recurring fix generation work for recurring fixes with identical contexts only. However, the PI's preliminary work in this area has shown that the majority of recurring fixes have different contexts, which suggests that new approaches are needed to learn and generate fixes automatically to tolerate different contexts. The PI will propose a new context matching algorithm to address this issue, enabling automated generation of fixes that are currently unachievable with existing techniques. This research program will create high-quality research capabilities. It will broaden the impact of the large amount of widely-used bug prediction, detection, and diagnosis techniques. It will open a new research direction of incorporating semantic information into automated program repair. This research program will reduce the cost of software development and improve the software reliability and security for the more than 1,300 software companies in Canada and improve Canada's competitiveness in this vital sector. It will train Highly Qualified Personnel (HQP) to work in the field of software engineering and software reliability. The HQP will acquire a unique blend of interdisciplinary knowledge and apply that knowledge to improve software reliability and security.