Statistical Analysis On Modification Of Files In Open-source Software Repository

Open and collaborative development of open-source software may change the traditionalmode for software development. Mining the evolutionary rules of source code files in a SVN(Subversion) repository is conducive to detecting potential bugs and then to improving thequality of software. Through mining the information of developer’s commits, on one hand,researchers with the community of software engineering can investigate evolutionary aspects ofOSS projects and analyze developer’s behaviors and collaboration, which might prove useful inquantifying and understanding the dynamics of human behavior on a collective scale.In this thesis we conducted statistical analyses on commit activity for4OSS projects, andfound that (1) the number of changes for class files follows roughly a power-law distribution;(2)the modification difference between a pair of adjacent versions of frequently-changed classesalso follows approximately a power-law distribution, and there are significantly positivecorrelations between the number of changes and both source lines of code and the number ofimported classes, implying that the function and structure of these classes tend to become morecomplex;(3) the commit size in terms of new definitions follows a power-law distribution, andexhibits self-similarity in time dimension;(4) there are5common zones for the distribution ofcommit activity across various releases in terms of our indicator, and there exists an interesting“deadline effects” in the last zone and developers do prefer to fix bugs in the stage of rushingdeadline, perhaps due to deadline pressure.The findings of this paper provide new insights into the research on evolution, refactoring,and task allocation of maintenance for open-source software. Moreover, this would offerdedicated guidance for OSS developers to draw up project schedule and release a new versionbetter.