我知道有很多关于这两种模式之间差异的帖子,但有一些东西我找不到。

From what I have been reading, I see that the factory method pattern allows you to define how to create a single concrete product but hiding the implementation from the client as they will see a generic product. My first question is about the abstract factory. Is its role to allow you to create families of concrete objects in (that can depend on what specific factory you use) rather than just a single concrete object? Does the abstract factory only return one very large object or many objects depending on what methods you call?

我最后两个问题是关于一句我在很多地方都见过的引语,我不能完全理解:

两者之间的一个区别是 使用抽象工厂模式,a 类委托的责任 对象实例化到另一个对象 通过合成,而工厂 方法模式使用继承和 类依赖于子类来处理 所需的对象实例化。

我的理解是,工厂方法模式有一个Creator接口,它将使ConcreteCreator负责知道要实例化哪个ConcreteProduct。这就是使用继承来处理对象实例化的意思吗?

现在,关于引用,抽象工厂模式是如何通过组合将对象实例化的责任委托给另一个对象的?这是什么意思?在我看来,抽象工厂模式似乎也使用继承来完成构造过程,但我仍然在学习这些模式。

任何帮助,特别是最后一个问题,将非常感激。


当前回答

之前的很多回答都没有提供抽象工厂和工厂方法模式之间的代码比较。下面是我试图用Java来解释它。我希望它能帮助那些需要简单解释的人。

正如GoF所言:抽象工厂提供了一个接口,无需指定就可以创建相关或依赖的对象族 具体的阶级。

public class Client {
    public static void main(String[] args) {
        ZooFactory zooFactory = new HerbivoreZooFactory();
        Animal animal1 = zooFactory.animal1();
        Animal animal2 = zooFactory.animal2();
        animal1.sound();
        animal2.sound();

        System.out.println();

        AnimalFactory animalFactory = new CowAnimalFactory();
        Animal animal = animalFactory.createAnimal();
        animal.sound();
    }
}

public interface Animal {
    public void sound();
}

public class Cow implements Animal {

    @Override
    public void sound() {
        System.out.println("Cow moos");
    }
}

public class Deer implements Animal {

    @Override
    public void sound() {
        System.out.println("Deer grunts");
    }

}

public class Hyena implements Animal {

    @Override
    public void sound() {
        System.out.println("Hyena.java");
    }

}

public class Lion implements Animal {

    @Override
    public void sound() {
        System.out.println("Lion roars");
    }

}

public interface ZooFactory {
    Animal animal1();

    Animal animal2();
}

public class CarnivoreZooFactory implements ZooFactory {

    @Override
    public Animal animal1() {
        return new Lion();
    }

    @Override
    public Animal animal2() {
        return new Hyena();
    }

}

public class HerbivoreZooFactory implements ZooFactory {

    @Override
    public Animal animal1() {
        return new Cow();
    }

    @Override
    public Animal animal2() {
        return new Deer();
    }

}

public interface AnimalFactory {
    public Animal createAnimal();
}

public class CowAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Cow();
    }

}

public class DeerAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Deer();
    }

}

public class HyenaAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Hyena();
    }

}

public class LionAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Lion();
    }

}

其他回答

两者的区别

“工厂方法”和“抽象工厂”的主要区别在于,工厂方法是方法,而抽象工厂是对象。我想很多人都把这两个词搞混了,开始交替使用。我记得当我学习它们的时候,我很难找到它们之间的确切区别。

因为工厂方法只是一个方法,它可以在子类中被重写,因此引用的后半部分:

... 工厂方法模式使用的 继承并依赖于一个子类 来处理所需的对象 实例化。

引用假设对象在这里调用自己的工厂方法。因此,唯一可以改变返回值的是子类。

抽象工厂是一个具有多个工厂方法的对象。看看你引言的前半部分:

... 使用抽象工厂模式,一个类 委托对象的职责 实例化到另一个对象 作文……

他们说的是,有一个对象A,想要创建一个Foo对象。而不是创建Foo对象本身(例如,使用工厂方法),它将获得一个不同的对象(抽象工厂)来创建Foo对象。

代码示例

为了向你展示区别,这里有一个正在使用的工厂方法:

class A {
    public void doSomething() {
        Foo f = makeFoo();
        f.whatever();   
    }

    protected Foo makeFoo() {
        return new RegularFoo();
    }
}

class B extends A {
    protected Foo makeFoo() {
        //subclass is overriding the factory method 
        //to return something different
        return new SpecialFoo();
    }
}

这是一个正在使用的抽象工厂:

class A {
    private Factory factory;

    public A(Factory factory) {
        this.factory = factory;
    }

    public void doSomething() {
        //The concrete class of "f" depends on the concrete class
        //of the factory passed into the constructor. If you provide a
        //different factory, you get a different Foo object.
        Foo f = factory.makeFoo();
        f.whatever();
    }
}

interface Factory {
    Foo makeFoo();
    Bar makeBar();
    Aycufcn makeAmbiguousYetCommonlyUsedFakeClassName();
}

//need to make concrete factories that implement the "Factory" interface here

Understand the differences in the motivations: Suppose you’re building a tool where you’ve objects and a concrete implementation of the interrelations of the objects. Since you foresee variations in the objects, you’ve created an indirection by assigning the responsibility of creating variants of the objects to another object (we call it abstract factory). This abstraction finds strong benefit since you foresee future extensions needing variants of those objects. Another rather intriguing motivation in this line of thoughts is a case where every-or-none of the objects from the whole group will have a corresponding variant. Based on some conditions, either of the variants will be used and in each case all objects must be of same variant. This might be a bit counter intuitive to understand as we often tend think that - as long as the variants of an object follow a common uniform contract (interface in broader sense), the concrete implementation code should never break. The intriguing fact here is that, not always this is true especially when expected behavior cannot be modeled by a programming contract. A simple (borrowing the idea from GoF) is any GUI applications say a virtual monitor that emulates look-an-feel of MS or Mac or Fedora OS’s. Here, for example, when all widget objects such as window, button, etc. have MS variant except a scroll-bar that is derived from MAC variant, the purpose of the tool fails badly. These above cases form the fundamental need of Abstract Factory Pattern. On the other hand, imagine you’re writing a framework so that many people can built various tools (such as the one in above examples) using your framework. By the very idea of a framework, you don’t need to, albeit you could not use concrete objects in your logic. You rather put some high level contracts between various objects and how they interact. While you (as a framework developer) remain at a very abstract level, each builders of the tool is forced to follow your framework-constructs. However, they (the tool builders) have the freedom to decide what object to be built and how all the objects they create will interact. Unlike the previous case (of Abstract Factory Pattern), you (as framework creator) don’t need to work with concrete objects in this case; and rather can stay at the contract level of the objects. Furthermore, unlike the second part of the previous motivations, you or the tool-builders never have the situations of mixing objects from variants. Here, while framework code remains at contract level, every tool-builder is restricted (by the nature of the case itself) to using their own objects. Object creations in this case is delegated to each implementer and framework providers just provide uniform methods for creating and returning objects. Such methods are inevitable for framework developer to proceed with their code and has a special name called Factory method (Factory Method Pattern for the underlying pattern). Few Notes: If you’re familiar with ‘template method’, then you’d see that factory methods are often invoked from template methods in case of programs pertaining to any form of framework. By contrast, template methods of application-programs are often simple implementation of specific algorithm and void of factory-methods. Furthermore, for the completeness of the thoughts, using the framework (mentioned above), when a tool-builder is building a tool, inside each factory method, instead of creating a concrete object, he/she may further delegate the responsibility to an abstract-factory object, provided the tool-builder foresees variations of the concrete objects for future extensions. Sample Code: //Part of framework-code BoardGame { Board createBoard() //factory method. Default implementation can be provided as well Piece createPiece() //factory method startGame(){ //template method Board borad = createBoard() Piece piece = createPiece() initState(board, piece) } } //Part of Tool-builder code Ludo inherits BoardGame { Board createBoard(){ //overriding of factory method //Option A: return new LudoBoard() //Lodu knows object creation //Option B: return LudoFactory.createBoard() //Lodu asks AbstractFacory } …. } //Part of Tool-builder code Chess inherits BoardGame { Board createBoard(){ //overriding of factory method //return a Chess board } …. }

据我估计,@TomDalling给出的答案确实是正确的(不管它有什么价值),但是评论中似乎仍然有很多困惑。

我在这里所做的是为这两种模式创建一些略显非典型的示例,并试图使它们乍一看非常相似。这将有助于查明将它们分开的关键差异。

如果您对这些模式完全不熟悉,那么这些示例可能不是最好的开始。

工厂方法

Client.javaish

Client(Creator creator) {
    ProductA a = creator.createProductA();
}

Creator.javaish

Creator() {}

void creatorStuff() {
    ProductA a = createProductA();
    a.doSomething();
    ProductB b = createProductB();
    b.doStuff();
}

abstract ProductA createProductA();

ProductB createProductB() {
    return new ProductB1();
}

为什么会有创造者和客户?

为什么不呢?FactoryMethod可以与两者一起使用,但它将是决定所创建的特定产品的Creator类型。

为什么createProductB在Creator中不是抽象的?

可以提供默认实现,子类仍然可以覆盖该方法以提供自己的实现。

我以为工厂方法只生产一种产品?

每个方法只返回一个产品,但创建者可以使用多个工厂方法,只是它们不一定以任何特定的方式相关。

抽象工厂

Client.javaish

AbstractFactory factory;

Client() {
    if (MONDAY) {
        factory = new Factory2();
    } else {
        factory = new AbstractFactory();
    }
}

void clientStuff() {
    ProductA a = factory.createProductA();
    a.doSomething();
    ProductB b = factory.createProductB();
    b.doStuff();
}

等等!你的AbstractFactory不是,嗯……er文摘

没关系,我们仍然在提供接口。create方法的返回类型是我们想要生成的产品的超类型。

圣烟蝙蝠侠!Factory2没有覆盖createProductA(),“产品族”发生了什么?

模式中并没有说一个对象不能属于一个以上的家族(尽管您的用例可能禁止这样做)。每个混凝土工厂负责决定哪些产品可以一起生产。

这是不对的,客户端没有使用依赖注入

您必须决定某个地方的具体类是什么,客户机仍然被写入AbstractFactory接口。

这里的混淆在于人们将组合与依赖注入混为一谈。客户端拥有一个AbstractFactory,而不管它是如何得到它的。与IS-A关系相比,Client和AbstractFactory之间没有继承关系。

关键的不同点

抽象工厂总是关于对象的家族 工厂方法只是一个允许子类指定具体对象类型的方法 抽象工厂为客户端提供了一个接口,它与产品的使用位置是分开的,工厂方法可以由创建者自己使用,也可以暴露给客户端。

总结

工厂的目的是为客户端或工厂本身提供对象。

创建者有自己的职责,可能需要使用对象或将对象传递给客户端

定义一个用于创建对象的接口,但是让子类来决定实例化哪个类。工厂方法允许类延迟实例化到子类。——GoF

抽象工厂:

提供一个接口来创建相关或依赖的对象族,而不指定它们的具体类。——GoF


PlantUML代码,如果你想玩图:

@startuml FactoryMethod
abstract class Creator {
    creatorStuff()
    {abstract} createProductA(): ProductA
    createProductB(): ProductB
}
class Creator1 {
    createProductA(): ProductA
}
class Creator2 {
    createProductA(): ProductA
    createProductB(): ProductB
}

together {
    interface ProductA {
        doSomething()
    }
    class ProductA1
    ' class Product1B
}
together {
    interface ProductB {
        doStuff()
    }
    class ProductB1
    class ProductB2
}
Client --> Creator

Creator <|-- Creator1
Creator <|-- Creator2

Creator --> ProductB1
ProductA1 <-- Creator1
ProductA1 <-- Creator2
ProductB2 <-- Creator2

ProductA <|.. ProductA1
ProductB <|.. ProductB1
ProductB <|.. ProductB2

ProductA <- Creator

@enduml
@startuml AbstractFactory

together {
    interface ProductA {
        doSomething()
    }
    class ProductA1
}

together {
    interface ProductB {
        doStuff()
    }
    class ProductB1
    class ProductB2
}

class AbstractFactory {
    createProductA(): ProductA
    createProductB(): ProductB
    --
    -
}

class Factory2 {
    createProductB(): ProductB
}

Client --> AbstractFactory
AbstractFactory <|-- Factory2

ProductA <|.. ProductA1
ProductB <|.. ProductB1
ProductB <|.. ProductB2

AbstractFactory --> ProductA1
AbstractFactory --> ProductB1
ProductB2 <-- Factory2

@enduml

之前的很多回答都没有提供抽象工厂和工厂方法模式之间的代码比较。下面是我试图用Java来解释它。我希望它能帮助那些需要简单解释的人。

正如GoF所言:抽象工厂提供了一个接口,无需指定就可以创建相关或依赖的对象族 具体的阶级。

public class Client {
    public static void main(String[] args) {
        ZooFactory zooFactory = new HerbivoreZooFactory();
        Animal animal1 = zooFactory.animal1();
        Animal animal2 = zooFactory.animal2();
        animal1.sound();
        animal2.sound();

        System.out.println();

        AnimalFactory animalFactory = new CowAnimalFactory();
        Animal animal = animalFactory.createAnimal();
        animal.sound();
    }
}

public interface Animal {
    public void sound();
}

public class Cow implements Animal {

    @Override
    public void sound() {
        System.out.println("Cow moos");
    }
}

public class Deer implements Animal {

    @Override
    public void sound() {
        System.out.println("Deer grunts");
    }

}

public class Hyena implements Animal {

    @Override
    public void sound() {
        System.out.println("Hyena.java");
    }

}

public class Lion implements Animal {

    @Override
    public void sound() {
        System.out.println("Lion roars");
    }

}

public interface ZooFactory {
    Animal animal1();

    Animal animal2();
}

public class CarnivoreZooFactory implements ZooFactory {

    @Override
    public Animal animal1() {
        return new Lion();
    }

    @Override
    public Animal animal2() {
        return new Hyena();
    }

}

public class HerbivoreZooFactory implements ZooFactory {

    @Override
    public Animal animal1() {
        return new Cow();
    }

    @Override
    public Animal animal2() {
        return new Deer();
    }

}

public interface AnimalFactory {
    public Animal createAnimal();
}

public class CowAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Cow();
    }

}

public class DeerAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Deer();
    }

}

public class HyenaAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Hyena();
    }

}

public class LionAnimalFactory implements AnimalFactory {

    @Override
    public Animal createAnimal() {
        return new Lion();
    }

}

抽象工厂创建了一个基类,其抽象方法定义了应该创建的对象的方法。派生基类的每个工厂类都可以创建每种对象类型的自己的实现。

工厂方法只是一个用于在类中创建对象的简单方法。它通常添加在聚合根中(Order类有一个名为CreateOrderLine的方法)

抽象工厂

在下面的示例中,我们设计了一个接口,这样我们就可以将队列创建与消息传递系统分离,因此可以为不同的队列系统创建实现,而不必更改代码库。

interface IMessageQueueFactory
{
  IMessageQueue CreateOutboundQueue(string name);
  IMessageQueue CreateReplyQueue(string name);
}

public class AzureServiceBusQueueFactory : IMessageQueueFactory
{
      IMessageQueue CreateOutboundQueue(string name)
      {
           //init queue
           return new AzureMessageQueue(/*....*/);
      }

      IMessageQueue CreateReplyQueue(string name)
      {
           //init response queue
           return new AzureResponseMessageQueue(/*....*/);
      }

}

public class MsmqFactory : IMessageQueueFactory
{
      IMessageQueue CreateOutboundQueue(string name)
      {
           //init queue
           return new MsmqMessageQueue(/*....*/);
      }

      IMessageQueue CreateReplyQueue(string name)
      {
           //init response queue
           return new MsmqResponseMessageQueue(/*....*/);
      }
}

工厂方法

HTTP服务器的问题在于,我们总是需要为每个请求提供响应。

public interface IHttpRequest
{
    // .. all other methods ..

    IHttpResponse CreateResponse(int httpStatusCode);
}

如果没有工厂方法,HTTP服务器用户(即程序员)将被迫使用特定于实现的类,这就违背了IHttpRequest接口的目的。

因此,我们引入工厂方法,以便响应类的创建也被抽象出来。

总结

区别在于,包含工厂方法的类的预期目的不是创建对象,而抽象工厂应该只用于创建对象。

使用工厂方法时应该小心,因为在创建对象时很容易破坏LSP(里斯科夫替换原则)。