Prevalence of Botched Code Integrations

Integrating code from different sources can be an error-prone and effort-intensive process. While an integration may appear statically sound, unexpected errors may still surface at run time. The industry practice of continuous integration aims to detect these and other run-time errors through an extensive pipeline of successive tests. Using data from a continuous integration service, Travis CI, we look into the prevalence of integration errors. We find code integration causes failure less often than regular commits. Repairing is usually done the same day and takes less than ten lines of code, largely in the source code. These results indicate that applying proper practices mitigates many issues associated with code integration.