July 22, 2017
Code Analysis for Dependency Inversion
As work continues on the next book and training course, I'm thinking about how we could analyse our code for adherence to the Dependency Inversion Principle (the "D" in S.O.L.I.D.)
The DIP states that "High-level modules should not depend upon low-level modules. Both should depend upon abstractions. Abstractions should not depend upon details, details should depend upon abstractions."
This is a roundabout way of saying dependencies should be swappable. The means by which we make them swappable is dependency injection (often confused with Dependency Inversion, and the two are very closely related.)
Dependency injection is simply passing an objects collaborators in (e.g., through a constructor) instead of that object instantianting them itself. When we directly instantiate an object, we bind ourselves to its exact type. This makes it impossible to swap that collaborator with a different implementation without modifying the client code, making our design inflexible and difficult to adapt or extend.
In practice, what this means is that most of our objects are composed from the outside.
For example, in my Reading Ease calculator, the Program class - the entry point for this console app - creates all of the objects involved in doing the calculation and "plugs" them together via constructors.
I've used the analogy of Russian dolls to describe how we compose simpler collaborations into more complex collaborations (collaborations within collaborations). This means that the lowest-level objects in the call stack typically get created first.
Inside those lower-level classes, there's no direct instantiation of collaborators.
So, when we analyse the dependencies, we should find that classes that have clients in our code - classes that are further down the call stack - don't directly instantiate their collaborators.
More simply, if things depend on you, then don't use new.
There are, of course, exceptions. Factories and Builders are designed to instantiate and hide the details. Integration code - e.g., opening database connections - is also designed to hide details. We can't very well pass our database connections into those, or we'd be spreading that knowledge. Typically what we're talking about here is dependencies on our own classes. And what a kerfuffle it would be to try to apply DIP to strings and ints and collections and other core library types all the time. Though, again, there are situations where that may be called for.
If I was measuring adherence to the Dependency Inversion Principle, then, I'd look at a class and ask "Do any other of my classes depend on this?" If the answer is "yes", then I'd check to see if it creates instances of any other of my classes. I might also check - and this would be language-dependent - if those dependencies are on abstract types (abstract classes, interfaces).
Posted 2 years, 9 months ago on July 22, 2017