Facilitate Extensibility

Accidental complexity is a real risk with trying to pursue systematic reuse . In trying to get reuse, you don’t want developers to jump through layers of indirection and complexity before they can write any useful code. Accounting for extensibility is a pragmatic approach when striving to design for systematic reuse.

Extensibility needs to be supported at various points in the design – it could be in user interface, integration, data contracts/definitions, system flows, services, and many other points. There are a plethora of choices and the question then becomes – how does one know what needs to be extensible? The answer lies in “finding what varies and encapsulating it”[1].

If a reusable component provides an extensibility hook – the consumer of the component can extend / replace default behaviour. Extensibility can be provided in a variety of ways – there ins’t a magic bullet here and each tactic needs to be considered in context. Here are various examples:

Spring Framework provides application context loader listeners that a developer can use to plugin custom logic. The listener doesn’t allow customization of how Spring constructs the beans or injects dependencies. However, it does provide a convenient mechanism for the developer to provide logic after container startup and before shutdown – place logic specific to a legacy/proprietary API that doesn’t support dependency injection. Alternatively, if there is a specific way to guarantee cleaning up resources prior to destroying context – that logic can be placed here as well.

Another extensibility strategy is from SLF4J– it is a logging facade and provides logging API methods that can be wired with concrete providers such as Log4J. Here, the idea is to free up individual classes from coupling with a specific provider. It provides much more points of extensibility than the above example.

Finally, the nature and number of extension points needs to be carefully weighed in the context of real applications. Not enough extension points hurts reusability. Similarly, too many increases learning curve, complexity, and possibly investment prior to making it usable in an application.

Last updated