Design Heuristics for Reliable and Flexible Virtual Machine Subsystems

This paper describes the process of designing a logging subsystem for a Smalltalk virtual machine written in C to illustrate a software design methodology focused on promoting reliability and flexibility. The design approach begins with identifying responsibilities, as well as sources of change and risk. Once these are determined, the implementation space is carved so that the areas affected by these sources of change and risk are encapsulated by interface boundaries that simultaneously fulfill the required responsibilities. Aligning responsibilities with sources of change and risk during this process results in clearer, less ambiguous interface boundaries. Interface clarity improvement manifests in a number of ways. For example, change mitigation implementation approaches should not be exposed by, or propagated through, their enclosing interface. Concentrating interface focus on fulfilling responsibilities reduces behavioral ambiguity. Moreover, by aligning responsibilities with sources of change and risk, the scope of responsibilities affected by future modifications is constrained. The resulting interfaces are more stable over time, and therefore reduce the background source code maintenance overhead. Consequently, these interfaces gain clarity from their stability. As an almost unintended side effect of this design methodology, the logging facility gained flexibility beyond initial expectations. Specifically, the more accurate boundaries allowed extending the logging subsystem's functionality in several beneficial, unanticipated ways without affecting its users or inducing code duplication. Hence, this design methodology also seems valuable as a heuristic approach to handle new feature requirements.