Streamlining Development of a Multimillion-Line Computational Chemistry Code

Computational science and engineering projects have goals and development strategies that can be quite different from those in the traditional scientific software engineering domain. However, methodologies developed by and for software engineers can be both applicable and helpful to the scientific software development process. This article presents a case study of applying the software engineering strategy of continuous integration to the Assisted Model Building with Energy Refinement (Amber) molecular dynamics code. The authors identify and discuss the key points of this strategy, their applicability to scientific codes, and how they were implemented in the Amber project. Despite difficulties in adapting existing continuous integration servers to fit the project's requirements, the implementation has streamlined the development process and continues to assist developers in identifying errors introduced on a commit-by-commit basis rather than immediately before release.