Consistency Management in the Presence of Dynamically Scoped Adaptations

Ensuring that software can display different behavior in different use contexts requires adapting software at runtime in dynamically created scopes (e.g. in a thread, in a client session, in a collaboration). ContextOriented Programming (COP) offers dedicated language constructs for performing such dynamically scoped adaptations. COP supports powerful customizations such as concurrent customization where independent clients can concurrently customize the functionality of a shared component, and system-wide customizations that cross technical boundaries. However, like any dynamic software adaptation technique, COP hits a conceptual barrier when new variations of existing program entities are integrated into a running system: if applied uncarefully, dynamically scoped adaptations may lead to faulty behavior especially when layer activation occurs at stages where the affected system components themselves are in a transitional, inconsistent state. Inherently, the ways of managing state consistency is highly specific to the application at hand and therefore requires additional application-specific logic from the system itself. What is needed is a consistency management framework that guides the programmer in specifying this application-specific logic in a principled way. 1 Flavors of context-oriented programming Many COP languages exists to date. Althoug there may be different flavors of COP, a common goal of most COP languages is to enable activation of layers of behavior in a specific dynamic scope of the program, in parallel with other dynamic scopes of the program. For example, our research platform Lasagne [7] supports per-client request activation of aspectual layers. This allows different clients to customize the same component instance in parallel. Looking further however at the different flavors of COP, it can be seen that at least two types of dynamically scoped adaptations are supported [2]: loyal adaptations where layers, once activated, are continously activated in their current dynamic scope, versus prompt adaptations where layers can be flexibly activated and deactivated in the same dynamic scope. For instance, Lasagne is a loyal COP language. The notion of per-request activation implies that the