Attack of the clones: an investigation into removing redundant source code

Abstract

Long-term maintenance of code will often lead to the introduction of duplicated or 'cloned' code. Legacy systems riddled with these clones have large amounts of redundant code and are more difficult to understand and maintain. One option available to improve maintainability and to increase software reuse, is to re-engineer code clones into reusable components. However, before this can be achieved detection and removal of this redundant code is necessary. There are several established clone detection tools for software maintenance and this thesis aims to investigate the similarities between their output. It also looks at how maintainers may best use them to reduce the amount of redundant code in a software system. This will be achieved by running clone detection tools on several different case studies. Included in these case studies will be a novel tool called Covet inspired by research of Mayrand [May96b] which attempted to identify cloned routines through a comparison of software metrics generated from each routine. It was found that none of the clone detection tools achieved either 100% precision or 100% recall. Each tool identified very different sets of clones. Overall MOSS achieved the greatest precision and CCFinder the greatest recall. Also observed was that the use of automatically generated code increased the proportion of clones found in a software system.