A brief study on build failures in continuous integration: Causation and effect

Abstract

Continuous Integration (CI) has successfully tackled the problem of bug fixing owing to which it has gained immense popularity among software developers. CI encourages to commit on the go so that each bug can be traced to its source and handled accordingly. However, CI remains a practice at its core, and only a part of it can be implemented. Anything which does not follow good CI practice would pave the way for a greater number of build fails. CI’s continuous nature may cause a clutter in a big team, leading to one developer’s build failing the other. Numerous consecutive build fails can put the project on a standstill till the build is made clean which may cause developers to lose interest eventually. We investigate, in this paper, causation and effect of build failure in CI. We first see whether a large team size contributes to more build failure and second, whether an increasing number of consecutive build failures have any impact on the productivity of developers. We have used data provided in TravisTorrent and analysed the 3,702,595 Travis builds which mostly contain Java and Ruby as the programming language used. For both the languages, we have made a comprehensive analysis of the problem we address.