Uma ferramenta para predição de mudanças conjuntas baseadas em informações de repositórios de software

Abstract

Maintenance is a stage of the software life cycle recognized by demanding a lot of effort in comparison to others, such as development. The maintenance tasks involve software modification, specifically the artifacts that compose it. Modifying a particular artifact can affect other parts of the software. This phenomenon is known as change impact. Techniques and tools to support the change impact analysis are usually based on types of couplings and have been proposed by several studies in the literature, such as the Association Rules and Machine Learning using Classification. However, these techniques have not been evaluated from the practical point of view. Therefore, a tool was developed to automate the execution of these techniques to predict joint changes of artifacts and present them to the developer, as well as collecting developers’ feedback on the tool. With the developed tool, a proof of concept was performed using the tool in maintenance tasks (bugs) of the Apache free software project called CXF, by novice collaborators represented by the students of the Computer Science course. Although it depends on the performance of joint change prediction techniques, the results showed evidence that the tool can support novice collaborators and also decrease the effort to perform a software maintenance task, comparing when no technique is used.