Change impact analysis for large-scale enterprise systems

This paper is concerned with analysing the potential impact of changes to enterprise systems, and in particular on how regression testing following such changes can be minimised. The target scope of the approach we describe in this paper is systems containing hundreds of thousands of classes and millions of methods. It is extremely difficult and costly to apply regular regression testing techniques to such systems. Retesting everything after a change is very expensive, and in practice often unnecessary. Selective retesting is dangerous if the impacts of change are not understood, and analysing such systems to understand what is being changed and what the impacts are is difficult. This paper proposes a way to perform a change impact analysis which makes it possible to do efficient, targeted regression testing of enterprise systems. Our approach has been tried on a large system comprising 4.6 million methods with 10 million dependences between them. Using our approach, maintainers can focus on a smaller, relevant subset of their test suites instead of doing testing blindly. We include a case study that illustrates the savings that can be attained.