Take the 2-minute tour ×
Programmers Stack Exchange is a question and answer site for professional programmers interested in conceptual questions about software development. It's 100% free, no registration required.

Possible Duplicate:
What is the difference between all-static-methods and applying a singleton pattern?

In C# Static methods has long served a purpose allowing us to call them without instantiating classes. Only in later year have we became more aware of the problems of using static methods and classes.

  • They can’t use interfaces
  • They can’t use inheritance
  • They are hard to test because you can’t make mocks and stubs

Is there a better way ? Obviously we need to be able to access library methods without instantiated classes all the time otherwise our code would become pretty cluttered

One possibly solution is to use a new keyword for an old concept: the singleton. Singleton’s are global instances of a class, since they are instances we can use them as we would normal classes. In order to make their use nice and practical we'd need some syntactic sugar however

Say that the Math class would be of type singleton instead of an actual class. The actual class containing all the default methods for the Math singleton is DefaultMath, which implements the interface IMath. The singleton would be declared as

singleton Math : IMath
{
   public Math
   {
      this = new DefaultMath();
   }
}

If we wanted to substitute our own class for all math operations we could make a new class

MyMath that inherits DefaultMath, or we could just inherit from the interface IMath and create a whole new Class. To make our class the active Math class, you'd do a simple assignment

Math =  new MyMath();

and voilá! the next time we call Math.Floor it will call your method. Note that for a normal singleton we'd have to write something like Math.Instance.Floor but the compiler eliminates the need for the Instance property

Another idea would be to be able to define a singletons as Lazy so they get instantiated only when they're first called, like

lazy singleton Math : IMath

What do you think, would it have been a better solution that static methods and classes? Is there any problems with this approach?

Addendum

Some points have been raised, that one of the main benefits of static methods is having methods that are "stateless" and thus side-effect free to some extent from a concurrency point of view. I wholeheartedly agree that that's a valid point. However we're mixing two different issues and problems here: One is making some methods invokable and globally accessible without having to explicitly create an instance. The other is having methods that are stateless.

The second problem could be solved by method having something like a stateless keyword that similarily to static prevented them from calling this or perhaps do even more to enforce no side-effecs. With singletons rather than static classes and something like stateless classes and methods I think you'd have the following pro's and con's

Pro's

  • "Static" methods in other classes and framework classes could easily be designed to be overridable
  • Classes would be easier to test
  • Using instances instead of static classes means design patterns work better (things like factories)
  • No limitation on inheritance and polymorphism in contrast to static

Cons

  • Perhaps slightly worse performance?
  • Bad programmers will but everything in singletons to have them globally accessible instead of using dependency injection, perhaps you should only be able to access singleton methods and not properties/fields to avoid global variables :)
  • ?

Perhaps Math was a bad example, but imagine if the .Net string methods were inefficient, you could easily replace them with your own using this method. Or some third-party class has has a singleton method that you wanted to alter slightly, you could inherit and alter that method

share|improve this question
1  
+1: Good thought provoking question –  Binary Worrier Mar 10 '11 at 13:39
1  
I'm interested to hear about the situation in which you need to globally redefine Math.Floor() to do something other than what you'd expect it to do. –  Ant Mar 10 '11 at 14:16
2  
You can test global functions just fine without resorting to interfaces, instances, and mocks. In fact I would question the practice, and some call me a TDD zealot. If calling a function on Math is part of the algorithm of the object I have under test, then I am testing the correctness of the algorithm-- not whether it calls certain functions on Math. –  Berin Loritsch Mar 10 '11 at 14:47
    
Math.Floor might not be the best example, but there are other framework methods you might want to override to give better performance or alter behavior slightly. –  konrad Mar 10 '11 at 16:11
2  
If you're trying to micro-optimise to the extent that the .NET string methods are too inefficient for you, you really shouldn't be using .NET in the first place. I think the overhead associated with inheritance chains and looking up methods within objects would probably outweigh any benefits you can make by implementing your own string class. It sounds to me like you're desperately trying to implement monkey patching in C#. –  Ant Mar 11 '11 at 15:38
show 3 more comments

marked as duplicate by ChrisF Mar 23 '12 at 10:03

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

3 Answers

There is no difference in the dangers by changing your static methods into instance methods on a singleton. They are effectively the same.

Sometimes you simply have functions. A function in mathematical terms, contains no state. It is simply the calculations to apply to an input to determine your output. For example, Math.Sqrt does not contain any state. It performs a calculation on the value you provide, and returns a response that is derived from that value. In C, C++, Ruby, Perl, and other languages that support simple functions you don't have to tie that function to any class, static or otherwise. In Java and C# all functions have to be tied to some class--which is technically a hack.

It's important to note that true functions are perfectly fine to remain a static method. For instance, everything in the Math class is properly defined. I don't fancy doing something stupid like Math.Instance.Sqrt(x) just to satisfy a "no static method" standard.

The truth is, your Instance property in this case would be the type of static method that causes the most problems. The reason? It's not a function--it maintains state, and changes its behavior based on the internal state. On the first access, it determines that the global instance doesn't exist so it creates it. On subsequent accesses it reuses that instance. In multi-threaded environments this can create a race condition which can potentially create multiple instances of the Math class. Eventually one will win out and remain stored, while the others get garbage collected. With the Math class the race condition would simply be an affordable inefficiency, but if the singleton needed to maintain state from all it's clients then that will be a problem. If you add locks around the singleton access, you've introduced a major performance bottleneck for multi-threaded applications that only gets worse with the number of threads. This is particularly because the lock is only needed until the instance is created.

Let your functions be as close to functions as possible, and you never need to worry about these multi-threaded singleton issues. In languages that support functions, implement them as functions. In languages that require a static method, make them a static method. Singletons introduce a number of issues that only come to light when you dig deeper into your project.

share|improve this answer
2  
@MKO, I disagree. The multithreading issues only come to play because you've changed the static methods to instance methods on a singleton. Why do we need a stateless keyword? What can the language do (practically) to enforce the statelessness of a method? –  Berin Loritsch Mar 10 '11 at 16:17
1  
@MKO, I still don't see that we're gaining anything by going through all these changes for simple stateless functions. It's already a language hack to have to tie those functions to a class. Adding yet one more and incorporating a host of problems that can otherwise be avoided doesn't seem like a good trade-off. I fail to see how inheritance plays a part in a simple function. Perhaps a much more concrete example that clearly demonstrates the type of problem you are trying to solve. The Math example is just a horrendous over-engineering to force that in your model. –  Berin Loritsch Mar 10 '11 at 19:06
1  
Yes, for the problem you specified: Math related functions, pure functions would be better. For more general needs where you need to maintain state, and do proper OO, then yes tying methods to classes is appropriate. The biggest takeaway is that you have to use the right tool for the job. What you were suggesting for the example you provided was not the right tool. What you suggested could be proper for another problem altogether. However, you can do it without a statically accessible singleton--which is no better than static methods tied to a class, and is in fact worse. –  Berin Loritsch Mar 11 '11 at 12:58
1  
@MKO, I'm still lost as to what the downside of static is in this case. I've done nothing but demonstrate how they are superior for this case. I even demonstrated that there are more downsides to instance methods on a singleton. At this point, I think we will have to agree to disagree. –  Berin Loritsch Mar 11 '11 at 14:26
1  
@Berin Loritsch While I'd agree that being forced to put static methods on an object can at times seem like a hack, when the class itself is also static it mostly becomes a module or a namespace. Terminology aside, I don't think a static method on a static class is much different from a module full of a library methods in another language. And surely you'd want your free-standing functions grouped together somehow for ease of use and maintenance. –  CodexArcanum Mar 11 '11 at 16:05
show 9 more comments

If the question is "What do we use to replace static classes?"

Singletons are not the answer.

Dependancy injection is.

That said, I'm using singletons to replace static classes in a legacy code base, in this instance it's far more achieveable to make this change than it is to add DI, and allows us get old code under test much faster. See this stack overflow question for more on this

Update
Q1: How would you use dependency injection on a framework level for things like Math and File functions?

At the framework level? Since we're allowed to add language support for Singleton we instead add support in the framework for DI in namespaces like Math, IO etc.
The machine config will provide sections that - by default - add the framework flavours of these, hooks which we can remove in app.config files and add in our own (or other ones for tests)

Q2: Also, if you could scope what class a singleton would use for different classes, wouldn't that be de facto dependency injection?

Sounds confusing to me mate, so instead of your singleton representing a single object, it now represents multiple objects and serves up different ones depending on the context of how it's called? Wouldn't that a) make it a multipleton and b) be a freaking nightmare to try and keep track of what any given Math method is going to do depending on how/where you call it.

share|improve this answer
    
How would you use dependency injection on a framework level for things like Math and File functions? –  konrad Mar 10 '11 at 13:23
    
Also, if you could scope what class a singleton would use for different classes, wouldn't that be de facto dependency injection? –  konrad Mar 10 '11 at 13:24
    
b)Normally you wouldn't much around with singleton context, it might be only a feature from testing purposes, and you don't really need it since you simply can set the singleton to what you want before and after testing –  konrad Mar 10 '11 at 16:13
    
a) machine config's doesn't seem like an elegant solution for a programming language –  konrad Mar 10 '11 at 16:15
1  
I actually think singleton instead of statics would be perfect for dependency injection. Just avoid using the singleton and instead pass the interface or base class of the classes you want to use to the classes. Since Math would no longer be static that would be much more flexible and you could create an instance of DefaultMath or create your own. –  konrad Mar 10 '11 at 17:52
show 2 more comments

The question of interfacing and inheriting static classes kept me busy for a while, too, although on a theoretical level more than on a practical.

There are also a couple of questions on SO dealing with the subject.

Finally I worked out a solution that essentially consists of

  • the separation of the "real class" part and the "static class" part into 2 classes
  • a class attribute assigning a class its pseudo-static class
  • a (static) helper class retrieving and instantiating the pseudo-static class for a class

For more details, read this mini series on my blog.

share|improve this answer
add comment

Not the answer you're looking for? Browse other questions tagged or ask your own question.