Design is choice. Most often, the struggle is not that there is no way to solve a design problem, but that there are too many ways that apparently solve the problem. You must know which collection of solutions solves the problem in a satisfactory manner. The need to choose propagates from the largest architectural levels down to the smallest unit of code. Furthermore, choices can be combined, which confers on design an evil multiplicity.
To fight the multiplicity of design with a reasonably small amount of code, a writer of a design-oriented library needs to develop and use special techniques. These techniques are purposely conceived to support flexible code generation by combining a small number of primitive devices. The library itself provides a number of such devices. Furthermore, the library exposes the specifications from which these devices are built, so the client can build her own. This essentially makes a policy-based design open-ended. These devices are called policies, and the implementations thereof are called policy classes.
The mechanics of policies consist of a combination of templates with multiple inheritance. A class that uses policies—a host class—is a template with many template parameters (often, template template parameters), each parameter being a policy. The host class "indirects" parts of its functionality through its policies and acts as a receptacle that combines several policies in a coherent aggregate.
Classes designed around policies support enriched behavior and graceful degradation of functionality. A policy can provide supplemental functionality that propagates through the host class due to public inheritance. Furthermore, the host class can implement enriched functionality that uses the optional functionality of a policy. If the optional functionality is not present, the host class still compiles successfully, provided the enriched functionality is not used.
The power of policies comes from their ability to mix and match. A policy-based class can accommodate very many behaviors by combining the simpler behaviors that its policies implement. This effectively makes policies a good weapon for fighting against the evil multiplicity of design.
Using policy classes, you can customize not only behavior but also structure. This important feature takes policy-based design beyond the simple type genericity that’s specific to container classes.
Policy-based classes support flexibility when it comes to conversions. If you use policy-by-policy copying, each policy can control which other policies it accepts, or converts to, by providing the appropriate conversion constructors, conversion operators, or both.
In breaking a class into policies, you should follow two important guidelines. One is to localize, name, and isolate design decisions in your class—things that are subject to a trade-off or could be sensibly implemented in other ways. The other guideline is to look for orthogonal policies, that is, policies that
don’t need to interact with each other and that can be changed independently.