Postdoc: Data-Flow Testing for Difficult-To-Find Bugs

点击翻译按钮获取中文摘要

9704703 Ryder, Barbara G. Rutgers University CISE Postdoctoral Research Associates in Experimental CS :Data-Flow Testing for Difficult-to-Find Bugs Data-flow coverage criteria are heuristics for judging the quality of software test suites and determining when enough testing has been done. The analysis required for data-flow testing of programs in languages such as C is currently impractical on large programs. Achieving near 100% satisfaction of the coverage requirements constitutes the real expense of data-flow testing. However, this is not necessarily related to the real expense of testing, which is looking for the difficult-to-find bugs. With the hypothesis that difficult-to-find bugs are correlated with statement hazard level, statement hazard is introduced to measure the unexpected effects of executing a program statement. Preliminary experiments indicate that a small proportion of program statements are highly hazardous. Hazard of a side-effect is defined as the degree to which its side-effects are unexpected by the programmer. Thus, experiments to determine the relation between hazardous statements and bug detection are described. These experiments build on other successful experiments in analyzing data-flow testing effectiveness, but additionally pay special attention to those bugs that are difficult to find. Determining the correlation between hazardous statements and difficult-to-find bugs will improve the effectiveness of data-flow testing and reduce its cost, making data-flow testing feasible in an industrial environment.