February 2, 2015
What's The Problem With Static Methods?Quick brain dump before bed.
When I'm pairing with people, quite often they'll declare a method as static. And I'll ask "Why did we make that static?" And they'll say "Well, because it doesn't have any state" or "It'll be faster" or "So I can invoke it without having to write new WhateverClassName()".
And then, sensing my disapproval as I subtly bang my head repeatedly against the desk and scream "Why? Why? Why?!", they ask me "What's the problem with static methods?"
I've learned, over the years, not to answer this question using words (or flags or interpretive dance). It's not that it's difficult to explain. I can do it in one made-up word: composibility.
If a method's static, it's not possible to replace it with a different implementation without affecting the client invoking it. Because the client knows exactly which implementation it's invoking.
Instance methods open up the possibility of substituting a different implementation, and it turns out that flexibility is the whole key to writing software with great composibility.
Composibility is very important when we want our code to accomodate change more easily. It's the essence of Inversion of Control: the ability to dynamically rewire the collaborators in a software system so that overall behaviour can be composed from the outside, and even changed at runtime.
A classic example of this is the need to compose objects under test with fake collaborators (e.g., mock objects or stubs) so that we can test them in isolation.
You may well have come across code that uses static methods to access external data, or web session state. If we want to test the logic of our application without hitting the file system or a database or having to run that code inside a web server, then those static methods create a real headache for us. They'd be the first priority to refactor into instance methods on objects that can be dependency injected into the objects we want to unit test.
The alternatives are high maintenance solutions: fake in-memory file systems, in-memory databases, fake web application servers, and so on. The dependency injection solution is much simpler and much cleaner, and in the long run, much, much cheaper.
But unit testability is just the tip of the iceberg. The "flex points" - to use a term coined in Steve Freeman and Nat Pryce's excellent Growing Object Oriented Software Guided By Tests book - we introduce to make our objects more testable also tend to make our software more composable in other ways, and therefore more open to accommodating change.
And if the static method doesn't have external dependencies, the unit testing problem is a red herring. It's the impact on composibility that really hurts us in the long run.
And so, my policy is that methods should be instance methods by default, and that I would need a very good reason to sacrifice that composibility for a static method.
But when I tell people whose policy is to make methods static if they don't need to access instance features this, they usually don't believe me.
What I do instead is pair with them on some existing code that uses static methods. The way static methods can impede change and complicate automated testing usually reveals itself quite quickly.
So, the better answer to "What's the problem with static methods?" is "Write some unit tests for this legacy code and then ask me again."
* And, yes, I know that in many languages these days, pointers to static functions can be dependency-injected into methods just like objects, but that's not what I'm talking about here.
Posted 2 weeks, 2 days ago on February 2, 2015