Transparent system support for software fault tolerance reduces performance in general and precludes application-specific optimizations in particular. In contrast, explicit support – especially at the language level – allows application-specific tailoring. However, current techniques that extend languages to support software fault tolerance lead to interwoven code addressing functional and non-functional requirements. Reflection promises both significant separation of concerns and a malleability allowing the user to customize the language toward the optimum point in a language design space. To explore this potential we compare common software fault tolerance scenarios implemented in both standard and reflective Ada. Specifically, in addition to backward error recovery and recovery blocks, we explore the application of reflection to atomic actions and conversations. We then compare the implementations in terms of expressive power, portability, and performance. Keywords. Reflection, ...
Patrick Rogers, Andy J. Wellings