Agent-based class design – a step towards betters customizations points

Abstracting a set of functionalities into a class which provides a higher level interface often requires tough design decisions. Users who do not have the exact requirements for which the abstraction is optimized will suffer a syntactic or runtime overhead as a result. Alexandrescu’s famous “policy-based design” provides a mechanism to allow the user to extend and customize an existing abstraction in order to fine-tune its functionality for many different use cases. This is however limited to use cases where each policy more or less represents a compile-time strategy pattern.

Alas, not everything is a strategy pattern. In this talk I will explore the viability of a more agent-pattern-like paradigm where each policy knows its requirements and publishes its capabilities. In this paradigm, glue code connecting any valid set of policies is automatically generated using template metaprogramming. This allows much more powerful customizations while maintaining static linkage.