Estudo do uso de vocabulários para analisar o impacto de relatórios de defeitos a código-fonte.

Abstract

Locating and fixing bugs described in bug reports are routine tasks in software development processes. A major effort must be undertaken to successfully locate the (possibly faulty) entities in the code that must be worked on. Generally, developers map bug reports to code through manual reading and inspection of both bug reports and the code itself. In practice, they must rely on their knowledge about the software architecture and design to perform the mapping in an efficient and effective way. However, it is well known that architectural and design knowledge is spread out among developers. Hence, the success of such a task is directly depending on choosing the right developer. In this paper, we present results of an empirical study we performed to evaluate whether the automated analysis of bug reports and software vocabularies can be helpful in the task of locating bugs. We conducted our study on eight versions of six mature Java open-source projects that use Bugzilla and JIRA as bug tracking systems. In our study, we have used Information Retrieval techniques to assess the similarity of bug reports and code entities vocabularies. For each bug report, we ranked ali code entities according to the measured similarity. Our results indicate that vocabularies are indeed a valuable source of information that can be used to narrow down the bug-locating task. For ali the studied systems, considering vocabulary similarity only, a Top 8% list of entities has about 75% of the target entities. We conclude that while vocabularies cannot be the sole source of information, they can certainly improve results if combined with other techniques.