Minerando exceções Runtime não documentadas em bibliotecas Java a partir do GitHub: um estudo exploratório

Abstract

Uncaught exceptions are not an exceptional scenario in today’s systems. Uncaught exceptions are estimated to account for up to two-thirds of the crashes on Java systems. Such exceptions may be thrown by problems in the application code, the underlying hardware, or even by a library used. Very often, documentation about exceptions flagged by library methods is missing or incomplete. As a consequence, the developer usually discovers these exceptions when they occur in the production environment, causing the application crash. Alternatively, static analysis tools can be used to identify such exceptions. However, these tools have a high computational cost and may be imprecise. In this scenario, this paper reports an exploratory study that extracted the stack traces inserted in issues of GitHub to discover the unchecked and undocumented exceptions of libraries. Overall, the issues of 2,970 Java projects hosted on GitHub were mined, of which 66,118 stack traces were extracted. In this study, the outstanding interfaces of the 85 most popular Maven libraries were investigated. The results show that static analysis captures only a small part of the exceptions that have been reported with GitHub’s stack traces in issues, and that the information available in these stack traces can, in fact, be used to discover undocumented exceptions released by library methods, thus enriching the data obtained with static analysis tools.