Pair program with me!

Open Closed Ruby Woes

Recently I received an email from a good friend, Jessica Kerr about how the Open/Closed Principle is applied in the Ruby community. It took me way too long to get back to her, but this topic was going to take some time to properly answer.

The Open/Closed Principle is one of the most misunderstood SOLID Design Principles. Misunderstood might not be the right term. Maybe I should say that it puts most people into a nonplus state. Open/Closed isn’t something that most people strive towards when writing their code. I believe Open/Closed is a place you end up when you’ve been working on a good design for a while. After many passings of refactoring you get a class into a position where it meets all of the other principles and suddenly you notice that it is easily extended.

I guess I should take a moment to define Open/Closed in case you missed the wiki links which tend to punch holes all over my posts. Open/Closed is the concept that classes are open to extension, but closed to modification. I think you might see why this is a little harder to notice than the other SOLID principles.

Ok, now let’s get back on track like nothing happened. Once your system has matured and been consciously refactored, many times you can find classes in your system that fit this wonderful idea. This is really nice when you find it, but there is a sweet spot. This is an easy principle to take way too far, much like the Single Responsibility Principle. When is a class at just the right time to stop? When has it become nothing more than a method class. When is there too much abstraction (that should be another post)? When is this class able to be extended, but doesn’t require extension to not be trivial and a pain to utilize? Now that I’ve pummeled you with questions I’m not going to answer any of them. Enjoy!

In searching for more ideas about how to answer @jessitron I went back to the source and asked her to clarify what she really wanted to know. She talked about “has-a” vs “is-a” relationships and how we in the Ruby community utilize libraries. This made me a little sad.

A large majority of our gems and libraries utilize inheritance as in ActiveRecord or a mixin. Sit down because this is going to be a big shock, mixins are inheritance! Take a moment to get all of your arguments out of the way. You can even email me about them if you feel so inclined.

Now that you’ve taken a deep breath an considered the way that Ruby looks up methods you’ve realized that a mixin is in fact inheritance. I personally would like to see more libraries that are not utilizing this way of thinking. It is sometimes convenient and great to work with, but it can limit you and encourage monkey patching. This doesn’t come from any real gem, but I’m sure there is one out there that does just this.

class User
  include Geloationable

  # Other awesome codes

At this point our User class would have all the methods and powers to handle geolocation, and this would include the idea of latitude and longitude. It would contain methods to find distances between points and many other geographical mapping tools. We have just expanded the responsibility of our class and violated the open closed principle. I believe a much better approach would be to have a Geolocation class that handles all of these responsibilities. Some will say that is what the above does, and then delegates everything to the Geolocation class. Yes, that is possible, but it still expands the User class and gives it a whole new realm of responsibility. The User model would have to be changed to have a Geolocation inside of it, but that is a much smaller change to the original. Wait maybe instead the geolocation would have a reference to the User, and then the User classes stays Closed. That thought just occurred, right there. You just witnessed that.

The nice thing is that the geolocation class can now stand alone. It can be used with more than just a User. It has a minimal impact on the interface of a User.

Another approach would be to make a GelocationableUser class that decorated the User. Please forgive me on my class naming. Maybe I shoul d grab a thesaurus. Using this approach you can add a single method to the original User to give you back the wrapped class. This is still changing the interface of User, but again it is minimal. Gregory Brown has some more about this in his SOLID Design Principles post. I recommend reading through it.

One last little bit. Before the end of our conversation Jessica asked, “How do you divide your own code into concrete and abstract?” My mind went immediately to abstract classes in Java. I’m not sure if it applies to the post, but I’m going to give you a simple guideline that I personally like. If I am going to use inheritance and the class can stand on it’s own I will utilize a class and note a module. If the parent cannot stand alone and requires a child to implement another piece of functionality in order to be used without throwing exceptions everywhere, I will utilize a mixin. If it isn’t an object in its own right then don’t make it a class. Would you be happy if you found the Geolocation class from above and when you sent it a message it blew up with a not implemented or method missing exception? Also I don’t do the not implemented route. I like to provide a module(Test::Unit) or a shared example group(RSpec) that is included in the tests for the extending class. These mostly test that the inheriter responds to the proper methods.

Now it is up to you to find your own way. If it can stand alone is it worth of being a class? Well, that depends on the situation. You will know it when you’ve felt it enough times. You still have to strive for it to truly feel it. Should you inherit? Should you decorate? Should it be a property of another class? These I will also answer with “well…it depends.” Enjoy…

The Past

Contact Me

Binary Noggin
This Agile Life