Dynamically Evolving Annotation-Based Variability for Runtime Adaptation

Software applications are increasingly desirable to adapt to varying execution environments and diverse user requirements to support runtime adaptation. Behavioral variations are generally modularized to handle variability. Depending on activating or deactivating relevant variations dynamically due to continuous context changes, a required behavioral variant specific to contexts is subsequently generated and has to evolve over time, to remain higher availability. However, since the potential amount of behavioral variants to be formed is huge, it is substantially unlikely to ensure all of them are type safe due to the unnecessary variants which are not used at all. To address this problem, we propose an annotation-based approach to variability. It can type check the entire program consisting of base program and many variations only once, without resulting from each variant individually. We extend the Featherweight Java to formally prove that all the expected variants are still well typed. Moreover, the type system guarantees type safety of the updated program as one variant evolves to another and type correctness that no run-time errors occur.