When to use the Bridge pattern and how is it different from the Adapter pattern?

The Bridge pattern is an application of the old advice, "prefer composition over inheritance". It becomes handy when you must subclass different times in ways that are orthogonal with one another. Say you must implement a hierarchy of colored shapes. You wouldn't subclass Shape with Rectangle and Circle and then subclass Rectangle with RedRectangle, BlueRectangle and GreenRectangle and the same for Circle, would you? You would prefer to say that each Shape has a Color and to implement a hierarchy of colors, and that is the Bridge Pattern. Well, I wouldn't implement a "hierarchy of colors", but you get the idea...

A classic example of the Bridge pattern is used in the definition of shapes in an UI environment (see the Bridge pattern Wikipedia entry). The Bridge pattern is a composite of the Template and Strategy patterns.

It is a common view some aspects of the Adapter pattern in the Bridge pattern. However, to quote from this article:

At first sight, the Bridge pattern looks a lot like the Adapter pattern in that a class is used to convert one kind of interface to another. However, the intent of the Adapter pattern is to make one or more classes' interfaces look the same as that of a particular class. The Bridge pattern is designed to separate a class's interface from its implementation so you can vary or replace the implementation without changing the client code.

Adapter and Bridge are certainly related, and the distinction is subtle. It's likely that some people who think they are using one of these patterns are actually using the other pattern.

The explanation I've seen is that Adapter is used when you're trying to unify the interfaces of some incompatible classes that already exist. The Adapter functions as a kind of translator to implementations that could be considered legacy.

Whereas the Bridge pattern is used for code that is more likely to be greenfield. You're designing the Bridge to provide an abstract interface for an implementation that needs to vary, but you also define the interface of those implementation classes.

Device drivers is an often-cited example of Bridge, but I'd say it's a Bridge if you're defining the interface spec for device vendors, but it's an Adapter if you're taking existing device drivers and making a wrapper-class to provide a unified interface.

So code-wise, the two patterns are very similar. Business-wise, they're different.

See also http://c2.com/cgi/wiki?BridgePattern

I have used the bridge pattern at work. I program in C++, where it is often called the PIMPL idiom (pointer to implementation). It looks like this:

class A
{
public: 
  void foo()
  {
    pImpl->foo();
  }
private:
  Aimpl *pImpl;
};

class Aimpl
{
public:
  void foo();
  void bar();
};  

In this example class A contains the interface, and class Aimpl contains the implementation.

One use for this pattern is to expose only some of the public members of the implementation class, but not others. In the example only Aimpl::foo() can be called through the public interface of A, but not Aimpl::bar()

Another advantage is that you can define Aimpl in a separate header file that need not be included by the users of A. All you have to do is use a forward declaration of Aimpl before A is defined, and move the definitions of all the member functions referencing pImpl into the .cpp file. This gives you the ability to keep the Aimpl header private, and reduce the compile time.

You're working for an insurance company where you develop a workflow application that manages different kind of tasks: accounting, contract, claims. This is the abstraction. On the implementation side, you must be able to create tasks from different sources: email, fax, e-messaging.

You begin your design with these classes:

public class Task {...}
public class AccountingTask : Task {...}
public class ContractTask : Task {...}
public class ClaimTask : Task {...}

Now, since each sources must be handled in a specific way, you decide to specialize each task type:

public class EmailAccountingTask : AccountingTask {...}
public class FaxAccountingTask : AccountingTask {...}
public class EmessagingAccountingTask : AccountingTask {...}

public class EmailContractTask : ContractTask {...}
public class FaxContractTask : ContractTask {...}
public class EmessagingContractTask : ContractTask {...}

public class EmailClaimTask : ClaimTask {...}
public class FaxClaimTask : ClaimTask {...}
public class EmessagingClaimTask : ClaimTask {...}

You end up with 13 classes. Adding a task type or a source type becomes challenging. Using the bridge pattern produces something easier to maintain by decoupling the task (the abstraction) from the source (which is an implementation concern):

// Source
public class Source {
   public string GetSender();
   public string GetMessage();
   public string GetContractReference();
   (...)
}

public class EmailSource : Source {...}
public class FaxSource : Source {...}
public class EmessagingSource : Source {...}

// Task
public class Task {
   public Task(Source source);
   (...)
}
public class AccountingTask : Task {...}
public class ContractTask : Task {...}
public class ClaimTask : Task {...}

Adding a task type or a source is now much more easier.

Note: Most developers would not create the 13 classes hierarchy upfront to handle this problem. However, in real life, you might not know in advance the number of source and task types ; if you have only one source and two task types, you would probably not decouple Task from Source. Then, the overall complexity grows as new sources and task types are added. At some point, you will refactor and, most often, end up with a bridge-like solution.

The key difference between the Adapter and Bridge design patterns lies in their intents. From Design Patterns, chapter 4, section ‘Bridge’, paragraph ‘Related Patterns’ (Gamma et al. 1994):

The Adapter (139) pattern is geared toward making unrelated classes work together. It is usually applied to systems after they’re designed. Bridge, on the other hand, is used up-front in a design to let abstractions and implementations vary independently.

1. The word ‘independently’ means that the Bridge design pattern applies in this situation because shapes and colours are independent:

             ------Shape-----                           Shape       Colour
            /                \              Bridge       / \         / \
       Circle                Square         ----->  Circle Square  Red Blue
        / \                   / \
RedCircle BlueCircle  RedSquare BlueSquare

2. But it does not apply in this situation because colours depend on shapes:

             ------Shape-----
            /                \
       Circle                Square
        / \                   / \
RedCircle BlueCircle  RedSquare GreenSquare

3. And it is useless in this situation because there are a single shape and a single colour:

Shape
  |
Circle
  |
RedCircle

Tabular representation of situation 1:

| Shape  | Colour |          | Shape  |  | Colour |
| ------ | ------ |          | ------ |  | ------ |
| circle | red    |  Bridge  | circle |  | red    |
| circle | blue   |  ----->  | square |  | blue   |
| square | red    |
| square | blue   |

Tabular representation of situation 2:

| Shape  | Colour |
| ------ | ------ |
| circle | red    |
| circle | blue   |
| square | red    |
| square | green  |

Tabular representation of situation 3:

| Shape  | Colour |
| ------ | ------ |
| circle | red    |

Thus the Bridge design pattern in object-oriented programming is equivalent to normalisation to projection–join normal form, denoted PJ/NF (Fagin 1979), in relational databases.

In situation 1, the relation schema R(Shape, Colour) has the multivalued dependencies ∅ ↠ {Shape} (independent shapes) and ∅ ↠ {Colour} (independent colours) which are not implied by the set of key dependencies {KEY({Shape, Colour})}, so it is not in PJ/NF. Its projections are in PJ/NF because R₁(Shape) has the trivial functional dependency {Shape} → {Shape} which is implied by the set of key dependencies {KEY({Shape})} and R₂(Colour) has the trivial functional dependency {Colour} → {Colour} which is implied by the set of key dependencies {KEY({Colour})}.

In situation 2, the relation schema R(Shape, Colour) has the trivial multivalued dependency {Shape} ↠ {Colour} which is implied by the set of key dependencies {KEY({Shape, Colour})}, so it is already in PJ/NF.

In situations 3, the relation schema R(Shape, Colour) has the functional dependencies ∅ → {Shape} (single shape) and ∅ → {Colour} (single colour) which are implied by the set of key dependencies {KEY({Shape, Colour}), KEY(∅)}, so it is already in PJ/NF.

I’ll give you one new example for bridge pattern if you get board of the same old Shape and Color example.

Let’s say you’ve different way to make payment like Card payment and net banking. And there are different payments gateways like CITI bank and HSBC bank.

Then you can just add the payment gateway member to the payment modes. And at runtime pass this information to the payment mode object. And make the payment.

So for example it will make the Card payment on CITI bank payment gateway.