In my experience the driving force to create interfaces didn't occur until I start doing unit testing with a mocking framework. It became abundantly clear that using interfaces was going to make mocking much easier (since the framework depended on the methods being virtual). Once I started I saw the value of abstracting away the interface to my class from the implementation. Even if I don't create an actual interface, I try now to make my methods virtual (providing an implicit interface that can be overridden).

我发现还有许多其他原因可以加强重构到接口的良好实践，但是单元测试/模拟的事情提供了最初的“顿悟时刻”的实践经验。

EDIT: To clarify, with unit testing and mocking I always have two implementations -- the real, concrete implementation and an alternate mock implementation used in testing. Once you have two implementations, the value of the interface becomes obvious -- deal with it in terms of the interface so you can replace the implementation at any time. In this case I'm replacing it with a mock interface. I know that I can do this without an actual interface if my class is constructed properly, but using an actual interface reinforces this and makes it cleaner (clearer to the reader). Without this impetus, I don't think I would have appreciated the value of interfaces since most of my classes only, ever have a single concrete implementation.

它解决了一个具体的问题:

你有a b c d四种不同类型。在你的代码中，你可以这样写:

a.Process();
b.Process();
c.Process();
d.Process();

为什么不让他们实现IProcessable呢

List<IProcessable> list;

foreach(IProcessable p in list)
    p.Process();

当你添加50种类型的类，它们都做同样的事情时，这种伸缩性会更好。

另一个具体问题是:

你有没有看过System.Linq.Enumerable?它定义了大量的扩展方法，可以对实现IEnumerable的任何类型进行操作。因为任何实现IEnumerable的东西基本上都在说“我支持无序foreach类型模式中的迭代”，所以你可以为任何可枚举类型定义复杂的行为(Count、Max、Where、Select等)。

使用接口有很多目的。

Use in polymorphic behavior. Where you want to call specific methods of a child class with an inteface having a reference to the child class. Having a contract with classes to implement all of the methods where it is necessary, like most common use is with COM objects , where a wrapper class is generated on a DLL which inherits the interface; these methods are called behind the scenes, and you just need to implement them but with the same structure as defined in the COM DLL which you can only know through the interface that they expose. To reduce memory usage by loading specific methods in a class. Like if you have three business objects and they are implemented in a single class, you can use three interfaces.

例如IUser, IOrder, IOrderItem

public interface IUser()
{

void AddUser(string name ,string fname);

}

// Same for IOrder and IOrderItem
//


public class  BusinessLayer: IUser, IOrder, IOrderItem

{    
    public void AddUser(string name ,string fname)
    {
        // Do stuffs here.
    }

    // All methods from all interfaces must be implemented.

}

如果你只想添加一个用户，可以这样做:

IUser user = new (IUser)BusinessLayer();

// It will load  all methods into memory which are declared in the IUser interface.

user.AddUser();

扩展一下Larsenal所说的。接口是所有实现类都必须遵循的契约。因此，您可以使用一种称为契约编程的技术。这允许您的软件变得独立于实现。

In my experience the driving force to create interfaces didn't occur until I start doing unit testing with a mocking framework. It became abundantly clear that using interfaces was going to make mocking much easier (since the framework depended on the methods being virtual). Once I started I saw the value of abstracting away the interface to my class from the implementation. Even if I don't create an actual interface, I try now to make my methods virtual (providing an implicit interface that can be overridden).

我发现还有许多其他原因可以加强重构到接口的良好实践，但是单元测试/模拟的事情提供了最初的“顿悟时刻”的实践经验。

EDIT: To clarify, with unit testing and mocking I always have two implementations -- the real, concrete implementation and an alternate mock implementation used in testing. Once you have two implementations, the value of the interface becomes obvious -- deal with it in terms of the interface so you can replace the implementation at any time. In this case I'm replacing it with a mock interface. I know that I can do this without an actual interface if my class is constructed properly, but using an actual interface reinforces this and makes it cleaner (clearer to the reader). Without this impetus, I don't think I would have appreciated the value of interfaces since most of my classes only, ever have a single concrete implementation.

假设你想要模拟当你试图睡觉时可能发生的烦恼。

接口前的模型

class Mosquito {
    void flyAroundYourHead(){}
}

class Neighbour{
    void startScreaming(){}
}

class LampJustOutsideYourWindow(){
    void shineJustThroughYourWindow() {}
}

正如你清楚地看到的，当你试图睡觉时，许多“事情”都可能令人讨厌。

使用没有接口的类

但是在使用这些类时，我们遇到了一个问题。他们毫无共同之处。您必须分别调用每个方法。

class TestAnnoyingThings{
    void testAnnoyingThinks(Mosquito mosquito, Neighbour neighbour, LampJustOutsideYourWindow lamp){
         if(mosquito != null){
             mosquito.flyAroundYourHead();
         }
         if(neighbour!= null){
             neighbour.startScreaming();
         }
         if(lamp!= null){
             lamp.shineJustThroughYourWindow();
         }
    }
}

带有接口的模型

为了克服这个问题，我们可以引入一个iterface

interface Annoying{
   public void annoy();

}

并在类中实现它

class Mosquito implements Annoying {
    void flyAroundYourHead(){}

    void annoy(){
        flyAroundYourHead();
    }
}

class Neighbour implements Annoying{
    void startScreaming(){}

    void annoy(){
        startScreaming();
    }
}

class LampJustOutsideYourWindow implements Annoying{
    void shineJustThroughYourWindow() {}

    void annoy(){
        shineJustThroughYourWindow();
    }
}

接口使用

这将使这些类的使用更容易

class TestAnnoyingThings{
    void testAnnoyingThinks(Annoying annoying){
        annoying.annoy();
    }
}