3 ms·

Firstly the interface isn't a layer of abstraction - it's a contract, a boundary. An abstraction is an arbitrary concept which may or may not apply to an interf

• w0utert 14y ago
  >> Firstly the interface isn't a layer of abstraction - it's a contract, a boundary. An abstraction is an arbitrary concept which may or may not apply to an interface. That's just semantics, let's concentrate on why I would always want to have an interface class instead of using the public class interface as the contract between a class and the rest of the system using it. Remember that many classes in any software design don't have (or need to have) base subclasses, composed classes, abstract interfaces, whatever. They are often nothing more than pieces of data with related methods, and nothing else. Writing lots of fluff around them to make such classes extensible, flexible, replaceable, loosely coupled, easy to stub for testing, whatever, does not add any value for you (as the programmer), your colleagues (when reading the code) or the customer (when running it). You'll know it when you'll need to do all of this anyway, and the moment you know it is early enough to think about it, no need to have contingency plans for things your classes are not ever going to be used for. >> Secondly, the abstract base concept is actually a deadly form of coupling. As many people have suggested since the dawn of the problem, composition is better than inheritance and what sits at the composition boundaries? Interfaces! Interfaces, not interface classes. We have many classes in our codebase that composite other classes, and none of them have interfaces at all, because the whole codebase is C++, which doesn't even have the concept of interface classes. Yet, our codebase is very clean and robust, fully covered by unit and integration tests, and rarely (if ever) contains bugs that are caused by bad or inflexible architecture. It does happen, on occasion, that some part of the code or its design doesn't satisfy new requirements anymore, but when that happens we simply refactor it, and if needed (which is extremely rare) write a facade around it to support its old public interface. Note that I'm not saying I'm happy with the fact C++ doesn't have interface classes, because there definitely are situations I would want to use it, for example where a component-based design would best fit the problem at hand. In those cases we usually introduce a pure-abstract class that mimics an interface class, which isn't ideal but does the job. I'm not sold on the 'abstract base class is a deadly form of coupling' at all. This may hold in some situations where you have many classes interacting in an architecture or framework that has a high probability of changing/evolving over time, but as I've said before: most code isn't like that. Most code is actually pretty plain and boring, without much need for architecture at all. Two or three levels in your class hierarchy and not more than a handful of class dependencies can solve a surprising amount of programming problems. Indiscriminately discarding inheritance or abstract base classes as 'evil' is naive at best, ignorant at worst. I can come up with many programming problems that are a perfect fit for concrete and/or polymorphic classes deriving from (semi-) abstract base classes, that would most definitely not be better implemented using composition. >> I have spent 2 weeks refactoring one into something which doesn't stick a fork in your eye every time you change something Similarly, I've spent a whole lot more than 2 weeks refactoring big balls of mud created by developers who threw as many classes against the wall as they could come up with, introducing interfaces, delegates, composites, strategy classes, and whatever other design pattern they found in the GoF book, only to arrive at the conclusion that 80% of the code was boilerplate, and the solution that was spread out over all those classes could be implemented by 1/10th the lines of code, concentrated in one or two classes. YMMV depending on the application domain you work in, but having worked on large business logic applications, web applications, and (now) scientific simulation software, my experience is that most complexity in software systems is only there because someone introduced it, not because it is necessary to solve the problem it was supposed to solve. >> So your tests are chock full of empty stub classes rather than mocks? Yuck. Our test cases aren't full of stub classes, and they arent full of interface classes either. Most classes can be perfectly tested in isolation by simply instantiating them and feeding them with static inputs. Some others are tightly coupled and covered by integration tests. None of this is rocket science, we're at over 1M SLOC and we have no problems covering them in our tests.