Collaborative Research: SHF: Medium: Improving Software Quality by Automatically Reproducing Failures from Bug Reports

点击翻译按钮获取中文摘要

The large demand for mobile device based services emphasizes the importance of software quality for mobile applications (apps). Because testing and other verification techniques cannot generally detect all bugs, it is common for app users to experience failures during normal operation. Developers rely on users reporting these bugs in issue-tracking systems to understand and resolve the failures. However, in current practice the process of reproducing the reported bugs must be done manually by developers, making app maintenance inefficient. This project will develop a family of techniques and tools that can extract relevant information for steps to reproduce from bug reports, dynamically search for reproducing sequences in the app to successfully reproduce the reported failure, and improve the quality of information used for failure reproduction. The products of these research initiatives will be used in several diverse software-engineering applications, including bug-report mining, bug-report reproduction, dynamic GUI exploration, and static analysis. This project aims to transform the way developers debug, reproduce, and understand software bugs from bug reports, and thus lead to more reliable software. The overall goal of this project is to improve the process of resolving mobile-app failures by automating the task of reproducing, creating, and generating tests from bug reports. The analytical components of this project involve: (1) a novel approach for accurately extracting steps to reproduce and their contextual information, (2) a novel GUI exploration technique to automatically search for reproducing event sequences, (3) a novel static analysis to help the reproduction search avoid locally-optimal but globally sub-optimal searches and lead to better overall and more successful reproductions. The integration of static and dynamic analyses, machine learning, and natural-language processing constitutes a novel reproduction framework that promises to provide not only practical solutions, but also theoretical advances in the field of software mining. The techniques developed in this project will be evaluated for effectiveness via large-scale experiments on real-world mobile apps.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.