Efficient, effective methods of software verification and validation

点击翻译按钮获取中文摘要

As software becomes more important to our everyday lives, it becomes more important to make sure that that software is doing what we want it to do, without bugs.  The goal of software verification and validation (V&V) is to find as many bugs as we can, as fast as we can, and to collect as much information as we can that helps us to eliminate those bugs. We plan to investigate efficient, effective methods for software V&V.  The primary method we plan to investigate is randomized unit testing.  This is the practice of exercising small pieces of closely-associated program text a huge number of times in a random order, allowing V&V practitioners to perform "torture tests" on individual pieces of program text.  We have developed tools and techniques for randomized unit testing that have already found bugs in widely-available public-domain software.  We plan to extend this work and make it available to all V&V practitioners. We also plan to investigate ways in which researchers can experimentally measure the effectiveness of software test techniques.  Key among these is the practice of automatically generating faulty "mutants" of a program and seeing whether the test techniques can catch the faults.  The percentage of faults caught is a measure of the effectiveness of the test technique.  Our work concentrates on measuring exactly how closely these automatically-generated mutants mimic the behaviour of real bugs.