Inferring rich input structure for software debugging and defence

点击翻译按钮获取中文摘要

Software systems are pervasive and critical to modern life, yet at the same time they are often complex, error prone, and high valued targets for hackers. As a result techniques that can help improve the quality of software can provide significant value to both industry and society. Improved tools for testing and debugging can lower the costs and increase productivity within software development. These lower costs and increased productivity are important for maintaining Canada's growing software industry. These same techniques can also improve the reliability of software, providing better assurances that computer programs do what they are intended to and are more resilient to malicious users. Improving the reliability and cost-effectiveness of software development has been a primary focus over the last decade for key software companies like Google, Apple, Microsoft, Facebook, and IBM. Existing state of the art approaches for automating debugging and testing are limited by the quality of the models for valid inputs that they use to guide analyses. They focus primarily on considering the syntactic validity of inputs. This proposal instead focuses on automatically constructing models of semantically valid inputs for a program. These models can be produced without human interaction and help enable the automated construction of customized tools for testing and debugging software.