When faced with designing objects, many developers immediately try to identify the abstractions. They have a preconception that the world must be organized as a hierarchy of abstractions. They quickly run into situations that don’t fit nicely into their hierarchy. They struggle to force-fit things into this model. After a while, they ask for help.

The world is not a tree. There may be forests. But generally the world is what it is. Do not force things to be a part of a tree, when that is not what they are. That should be obvious enough. It is always the most obvious things that are difficult to realize, when one is blinded by misconceptions.

Step 1. Identify the concrete types. Start from the objects that will actually exist, not from higher level abstractions. The concretes have definite characteristics that can be analyzed. These characteristics are where the abstractions will be found later through refactoring (unit economy). Concretes are characterized by their behavior. A concrete has roles and responsibilities relative to other objects with which it collaborates. It may be necessary to decompose a concrete into smaller parts that take on those distinct roles and responsibilities (separation of concerns) more clearly.

Step 2. Refactor the characteristics into higher level abstractions. Identify characteristics and behaviors that are common across concretes. Often behaviors can be seen as actions performed on the concretes (rather than actions performed by the concretes), and there is a distinct characteristic that each action depends on. I frequently call such a characteristic a capability, and the action is known as a generic algorithm. As this refactoring proceeds, the generic algorithms and the capabilities are identified as higher level abstractions shared or applied across concretes.

Step 3. Refactor the algorithms and capabilities. As the generic algorithms are implemented against the capabilities of the concretes, the methods will often benefit from factoring out commonality (duplicate code). Parts of the algorithm may benefit from being specialized through pluggable policies. These motivations identify additional abstractions at finer granularities. The finer grained parts of algorithms will likely act upon finer grained capabilities on the concretes. Broader generalizations and narrower specializations appear in the model as a consequence of this iterative approach due to refactoring to reduce complexity (the complexity is refactored to become encapsulated within objects, which are more manageable than messy algorithms).

The resulting model is forests of trees, along with lions and tigers and bears, as well as all the other concretes that exist in the world. Things as they really are. The way it ought to be.