软件自动修复技术修复内涵分析研究

发布时间：2018-07-20 09:28

【摘要】：随着时代的发展,软件的开发规模与软件复杂度日渐提高,由于开发人员的因素或者软件本身的因素导致软件中的缺陷越来越多,并且缺陷种类繁多,软件在社会中引发的故障所造成的损失也越来越大,开发人员的目光必须重点放在对软件修复的研究上,然而人工进行软件调试不仅给维护人员增加了繁重的工作量,还可能延缓软件恢复正常运行的时间。所以学术界把研究重点放在了如何能有效减缓调试人员的压力上,为此开发了自动调试技术,例如软件自动修复技术。虽然近些年来在关于程序自动修复上的研究取得了很大成果,但是仍然存在很多待解决问题,自动修复的成功率和效率依然有待提高。一般而言,软件自动修复技术都是基于缺陷容忍与缺陷排除的方式来实现软件错误自动修复的,可是目前的自动修复技术只关注于修复补丁是否能够满足测试用例集的验证,却忽略了对修复效果的研究分析。本文主要是对自动修复工具在修复效果方面的探讨研究,通过对两种具有代表性的自动修复工具GenProg与CETI修复结果的收集,分析两种工具的修复效果,以此来判断自动修复工具在修复时是采取缺陷容忍修复还是缺陷排除修复,在修复时工具更倾向于哪种修复手段。从本文的实验结果可得知大部分情况软件自动修复工具在修复时会使用缺陷容忍的方式,只有小部分情况会使用缺陷排除的方式,缺陷排除修复占总修复量的两成左右。本文通过对缺陷涉及代码行数统计,以及缺陷到失效所经过传播链的长度,提出了一个关于错误复杂程度的理念,由于自动修复软件在修复时必然会涉及到缺陷代码以及错误传播链,本文进而对错误复杂度与自动软件修复倾向进行了研究,通过实验表明,错误复杂度越高时软件自动修复工具越倾向于缺陷容忍修复或者只能使用缺陷容忍来进行修复,当缺陷复杂度低时,自动修复软件才会考虑使用缺陷排除来进行修复。本文的研究成果明确了现有自动修复工具的修复效果以缺陷容忍为主,并发现了错误复杂度与修复手段之间的关联关系,为提升当前主流软件自动修复工具的修复效率提供了依据和基础。
[Abstract]:With the development of the times, the scale of software development and software complexity are increasing. Due to the factors of the developer or the software itself, there are more and more defects in the software, and there are many kinds of defects. The damage caused by the failure caused by software in the society is more and more great. Developers must focus on the research of software repair. However, manual debugging of software not only increases the heavy workload for maintainers, It may also delay the time the software is back to normal. Therefore, the academic research focuses on how to effectively reduce the pressure of debugging staff, and developed automatic debugging technology, such as software automatic repair technology. In recent years, great achievements have been made in the research of automatic program repair, but there are still many problems to be solved, and the success rate and efficiency of automatic repair still need to be improved. Generally speaking, software automatic repair technology is based on defect tolerance and defect removal to achieve automatic software fault repair, but the current automatic repair technology only focuses on whether the patch can satisfy the verification of test case set. However, the research and analysis of the effect of restoration were ignored. This paper mainly discusses the effect of automatic repair tools, through the collection of two representative automatic repair tools GenProg and CETI repair results, analyzes the repair effects of the two tools. In order to determine whether the automatic repair tool should be repaired by defect tolerance or defect removal, the tool is more inclined to repair the tool. From the experimental results of this paper, we can know that in most cases, the software automatic repair tools will use defect tolerance while only a small number of cases will use defect elimination, defect exclusion and repair accounts for about 20% of the total repair amount. By counting the number of lines of code involved in the defect and the length of the propagation chain from the defect to the failure, this paper presents an idea of the complexity of the error. Due to the defect code and error propagation chain will be involved in the repair of automatic repair software, the error complexity and the tendency of automatic software repair are studied in this paper, and the experimental results show that, The higher the error complexity, the more prone the software automatic repair tool is to defect tolerance or defect tolerance. When the defect complexity is low, the automatic repair software will consider using defect elimination to repair. The research results of this paper show that the defect tolerance is the main effect of the existing automatic repair tools, and the relationship between the error complexity and the repair means is also found. It provides the basis and foundation for improving the repair efficiency of the current mainstream software automatic repair tools.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP311.53

【相似文献】