Specifying and Exploiting Advice-Execution Ordering using Dependency State Machines∗

In this paper we present Dependency State Machines, an annotation language that extends AspectJ with finite-state machines that define the order in which pieces of advice must execute to have a visible effect. Dependency State Machines facilitate the automatic verification and optimization of aspects, but also program understanding. In this work we present the syntax and semantics of Dependency State Machines and one possible use case of Dependency State Machines: program understanding. We explain how a set of three static program analyses can exploit the information that Dependency State Machines carry to remove advice-dispatch code from program locations at which dispatching the advice would have no effect. Dependency State Machines hereby help to abstract from the concrete implementation of the aspect, making the approach compatible with a wide range of aspect-generating monitoring tools. Our extensive evaluation using the DaCapo benchmark suite shows that our approach can pinpoint to the user exactly the program locations at which the aspect’s execution matters in many cases. This is particularly useful when the aspect’s purpose is to identify erroneous execution sequences: in these cases, the program locations that our analysis pinpoints resemble possible points of program failure.