Leveraging Test Suites for Better Program Analysis

点击翻译按钮获取中文摘要

State-of-the-art tools for understanding programs include static and dynamic program analyses, which are powerful tools for predicting and reasoning about program behaviours. Developers use static and dynamic program analysis to improve their software. For example, program analysis can locate undesirable behaviours and help enforce desired properties, including security properties; and it can answer developer questions during program development. Any program analysis, whether static or dynamic, could benefit from additional information about the program. Modern programming methodologies expect developers to maintain automated test suites along with their software. Test suites encode developer understanding of correct program behaviours. Consider the following analogy. At its outset, a game of chess allows a vast number of possibilities. Each move restricts the available possibilities for the players---endgames are simple enough for even beginners to analyze. Similarly, each test case focusses attention on specific parts of the program. Restricting the problem scope enables analyses to make strong guarantees about program behaviour within that scope. The key idea behind the proposed research is to leverage test suites to enable more effective program analyses. The overall long-term objective of the proposed research programme is to develop a theory of combined static and dynamic analysis. This theory will support the design of program analyses that integrate information from source code (as done today) and its test cases (which is novel). Novel hybrid analyses will leverage the strengths of both static and dynamic analyses, using test suites to drive the dynamic analysis and provide information for the static analysis. Dynamic analysis can help refine the approximations inherent to static analysis and reduce the number of reported false positives. Static analysis can broaden the applicability of the results obtained from dynamic analysis. Specific applications of this theory include, for instance, automatically detecting inconsistencies in test suites. The key insight, that test suites can provide additional information about expected program behaviours, is one that can easily be incorporated into others' research programmes, contributing to the proposed research's academic impact. There is widespread industry interest in improving test practices. Developers write enormous amounts of test code, and expend substantial effort on software maintenance, which is critically supported by test suites. Results that improve the quality and usability of test suites can be quickly adopted by Canadian industry.