Structural Patterns: Adapter and Façade
This chapter is excerpted from C# 3.0 Design Patterns: Use the Power of C# 3.0 to Solve Real-World Problems by Judith Bishop, published by O'Reilly Media
.gif "Cover")
.png "Logo")
.png "Buy Now")
The main pattern we will consider in this chapter is the Adapter pattern. It is a versatile pattern that joins together types that were not designed to work with each other. It is one of those patterns that comes in useful when dealing with legacy code-i.e., code that was written a while ago and to which one might not have access. There are different kinds of adapters, including class, object, two-way, and pluggable. We'll explore the differences here. The second pattern we will look at in this chapter-the Façade pattern-is a simple one that rounds out the structural group. The aim of this pattern is to provide a simplified interface to a set of complex systems.
Adapter Pattern
Role
The Adapter pattern enables a system to use classes whose interfaces don't quite match its requirements. It is especially useful for off-the-shelf code, for toolkits, and for libraries. Many examples of the Adapter pattern involve input/output because that is one domain that is constantly changing. For example, programs written in the 1980s will have very different user interfaces from those written in the 2000s. Being able to adapt those parts of the system to new hardware facilities would be much more cost effective than rewriting them.
Toolkits also need adapters. Although they are designed for reuse, not all applications will want to use the interfaces that toolkits provide; some might prefer to stick to a well-known, domain-specific interface. In such cases, the adapter can accept calls from the application and transform them into calls on toolkit methods.
Illustration
Our illustration of the Adapter pattern is a very real one-it involves hardware instruction sets, not input/output. From 1996 to 2006, Apple Macintosh computers ran on the PowerPC processor. The operating system was Mac OS X. But in April 2006, Apple started releasing all new Apple computers-iMacs, Minis, and MacBooks-with Intel Core Duo processors. Mac OS X was rewritten to target the new processor, and users of the new computers mostly accessed existing Intel-based software via other operating systems, such as Linux and Windows. Figure 4.1, "Adapter pattern illustration-migration of Mac OS X from a 1998 PowerPC-based iMac to a 2007 Intel-based iMac" shows iMacs made in 1998 and 2006.
Figure 4.1. Adapter pattern illustration-migration of Mac OS X from a 1998 PowerPC-based iMac to a 2007 Intel-based iMac
.png "Adapter pattern illustration-migration of Mac OS X from a 1998 PowerPC-based iMac to a 2007 Intel-based iMac")
Mac OS X was originally designed to take advantage of the AltiVec floating-point and integer SIMD instruction set that is part of the PowerPC processor. When the Intel processor replaced this processor, calls to AltiVec instructions from Mac OS X had to be retargeted to the Intel x86 SSE extensions, which provide similar functionality to AltiVec.
For something as important as an operating system, the code could be rewritten to replace the calls to AltiVec with calls to SSE. However, Apple recognized that application developers might not want to do this, or might not have access to the source of old AltiVec-based code, so they recommended the use of the Accelerate framework. The Accelerate framework is a set of high-performance vector-accelerated libraries. It provides a layer of abstraction for accessing vector-based code without needing to use vector instructions or knowing the architecture of the target machine. (This is the important point for us here.) The framework automatically invokes the appropriate instruction set, be it PowerPC or Intel (in these processors' various versions).
Thus, the Accelerate framework is an example of the Adapter pattern. It takes an existing situation and adapts it to a new one, thus solving the problem of migrating existing code to a new environment. No alterations to the original code are required.[3]
Design
The Adapter pattern's important contribution is that it promotes programming to interfaces. The Client works to a domain-specific standard, which is specified in the ITarget interface. An Adaptee class provides the required functionality, but with a different interface. The Adapter implements the ITarget interface and routes calls from the Client through to the Adaptee, making whatever changes to parameters and return types are necessary to meet the requirements. A Target class that implements the ITarget interface directly could exist, but this is not a necessary part of the pattern. In any case, the Client is aware only of the ITarget interface, and it relies on that for its correct operation.
The adapter shown in Figure 4.2, "Adapter pattern UML diagram" is a class adapter because it implements an interface and inherits a class. The alternative to inheriting a class is to aggregate the Adaptee. This creates an object adapter. The design differences are primarily that overridingAdaptee behavior can be done more easily with a class adapter, whereas adding behavior to Adaptees can be done more easily with an object adapter. As we go along, I will point out different instances.
Figure 4.2. Adapter pattern UML diagram
.png "Adapter pattern UML diagram")
The purpose of the ITarget interface is to enable objects of adaptee types to be interchangeable with any other objects that might implement the same interface. However, the adaptees might not conform to the operation names and signatures of ITarget, so an interface alone is not a sufficiently powerful mechanism. That is why we need the Adapter pattern. An Adaptee offers similar functionality to Request, but under a different name and with possibly different parameters. The Adaptee is completely independent of the other classes and is oblivious to any naming conventions or signatures that they have. Now, let's consider the roles in the pattern:
ITarget
The interface that the Client wants to useAdaptee
An implementation that needs adaptingAdapter
The class that implements the ITarget interface in terms of the AdapteeRequest
An operation that the Client wantsSpecificRequest
The implementation of Request's functionality in the Adaptee
Tip
The pattern applies to a single computer, which would only have either the PowerPC or the Intel chip. In this case, it has the Intel chip-hence the need for the adapter. There is no Target class present, just the ITarget interface.
Quiz: Match the Adapter Pattern Players with the Mac OS X Migration Illustration
To test whether you understand the Adapter pattern, cover the lefthand column of the table below and see if you can identify its players among the items from the illustrative example (Figure 4.1, "Adapter pattern illustration-migration of Mac OS X from a 1998 PowerPC-based iMac to a 2007 Intel-based iMac"), as shown in the righthand column. Then check your answers against the lefthand column.
| Client | Mac OS X (or any Mac application) |
| ITarget | The specification of the AltiVec instruction set |
| Request | A call to an AltiVec instruction |
| Adapter | The Accelerate framework |
| Adaptee | An Intel processor with an SSE instruction set |
| SpecificRequest | A Call to an SSE instruction |
Implementation
It is best to illustrate the structure of the adapter with a small example, even at the theory code level. Suppose that technical readings are being collected and reported at a high level of precision, but the client can only make use of rough estimates. The signatures for the interface would be couched in terms of integers, and for the actual implementation in terms of double-precision numbers. Thus, an adapter is needed, as shown in Example 4.1, "Adapter pattern theory code".
Example 4.1. Adapter pattern theory code
1 using System;
2
3 // Adapter Pattern - Simple Judith Bishop Oct 2007
4 // Simplest adapter using interfaces and inheritance
5
6 // Existing way requests are implemented
7 class Adaptee {
8 // Provide full precision
9 public double SpecificRequest (double a, double b) {
10 return a/b;
11 }
12 }
13
14 // Required standard for requests
15 interface ITarget {
16 // Rough estimate required
17 string Request (int i);
18 }
19
20 // Implementing the required standard via Adaptee
21 class Adapter : Adaptee, ITarget {
22 public string Request (int i) {
23 return "Rough estimate is " + (int) Math.Round(SpecificRequest (i,3));
24 }
25 }
26
27 class Client {
28
29 static void Main ( ) {
30 // Showing the Adapteee in standalone mode
31 Adaptee first = new Adaptee( );
32 Console.Write("Before the new standard\nPrecise reading: ");
33 Console.WriteLine(first.SpecificRequest(5,3));
34
35 // What the client really wants
36 ITarget second = new Adapter( );
37 Console.WriteLine("\nMoving to the new standard");
38 Console.WriteLine(second.Request(5));
39 }
40 }
41/* Output
42 Before the new standard
43 Precise reading: 1.66666666666667
44
45 Moving to the new standard
46 Rough estimate is 2
47 */
The main program in the client shows two scenarios. First, there is an example of how the Adaptee could be called directly (line 33)-its output is shown in line 43. However, the client wants to work to a different interface for requests (lines 17 and 38). The Adapter implements the ITarget interface and inherits the Adaptee (line 21). Therefore, it can accept Request calls with a string-int signature and route them to the Adaptee with a double-double-double signature (line 23). The new output is shown on line 46.
A feature of adapters is that they can insert additional behavior between the ITarget interface and the Adaptee. In other words, they do not have to be invisible to the Client. In this case, the Adapter adds the words "Rough estimate is" to indicate that the Request has been adapted before it calls the SpecificRequest (line 23).
Adapters can put in varying amounts of work to adapt an Adaptee's implementation to the Target's interface. The simplest adaptation is just to reroute a method call to one of a different name, as in the preceding example. However, it may be necessary to support a completely different set of operations. For example, the Accelerate framework mentioned in the "the section called "Illustration" section will need to do considerable work to convert AltiVec instructions to those of the Intel Core Duo processor. To summarize, we have the following options when matching adapter and adaptee interfaces:
Adapter interface and adaptee interface have same signature
This is the trivial case, with not much work to do.Tip
Many examples of the Adapter pattern operate at this level and are not illustrative or helpful in explaining its real power. Beware of them.
Adapter interface has fewer parameters than adaptee interface
The adapter calls the adaptee with some dummy input.Tip
This case is shown in Example 4.1, "Adapter pattern theory code", where the second parameter is defaulted to 3.
Adapter interface has more parameters than adaptee interface
The adapter adds the missing functionality, making it half an adapter and half a component in its own right.Adapter interface has other types than adaptee interface
The adapter performs some type conversion (casting).Tip
This case is shown in Example 4.1, "Adapter pattern theory code", where the first double parameter is created from an integer and the double return type is cast back to a string.
Of course, combinations of these basic cases are also possible.
Two-Way Adapters
Adapters provide access to some behavior in the Adaptee (the behavior required in the ITarget interface), but Adapter objects are not interchangeable with Adaptee objects. They cannot be used where Adaptee objects can because they work on the implementation of the Adaptee, not its interface. Sometimes we need to have objects that can be transparently ITarget or Adaptee objects. This could be easily achieved if the Adapter inherited both from both classes; however, such multiple inheritance is not possible in C#, so we must look at other solutions.
The two-way adapter addresses the problem of two systems where the characteristics of one system have to be used in the other, and vice versa. An Adapter class is set up to absorb the important common methods of both and to provide adaptations to both. The resulting adapter objects will be acceptable to both sides. Theoretically, this idea can be extended to more than two systems, so we can have multiway adapters, but there are some implementation limitations: without multiple inheritance, we have to insert an interface between each original class and the adapter.
Our Macintosh example has a follow-up that illustrates this point nicely. With an Intel processor on board, a Mac can run the Windows operating system.[4] Windows, of course, is targeted directly for the Intel processor and will make use of its SSE instructions where necessary. In such a situation, we can view Windows and Mac OS X as two clients wanting to get access to the Adaptee (the Intel processor). The Adapter catches both types of instruction calls and translates them if required. For an instruction issued by an application, the situation on the different operating systems running on a Mac with an Intel processor can be summed up using pseudocode as follows:
Mac OS X
ExecuteAltiVec(instruction);Windows
ExecuteSEE(instruction);Adapter
void ExecuteAltiVec(instruction) { ExecuteSSE(ConvertToSSE(instruction)); } void ExecuteSSE(instruction) { Intel.ExecuteSSE(instruction); }
A key point with a two-way adapter is that it can be used in place of both ITarget and the Adaptee. When called to execute AltiVec instructions, the adapter behaves as a PowerPC processor (the Target), and when called to execute SSE instructions, it behaves as an Intel processor (the Adaptee).
Example: The Seabird
We have already looked at some theory code and discussed an interesting application of the Adapter pattern concept. It is now time for an example. That illustrates a two-way adapter but sticks closely to the structure of Example 4.1, "Adapter pattern theory code".
Suppose an inventor has an embedded control system for a revolutionary water plane called the Seabird. The plane derives from both aircraft and seacraft design: specifically, the Seabird has the body of a plane but the controls and engine of a ship. Both parts are being assembled as-is. The inventor of the Seabird is adapting as much as he can so that it is possible to control the craft via the interfaces provided by both parts.
In pattern terms, the Seabird will be a two-way adapter between the Aircraft and Seacraft classes. When running the experiments on the Seabird, the inventor will use an adapter and will be able to access methods and data in both classes. In other words, Seabird will behave as both an Aircraft and a Seacraft. We could get a simple adapter to behave as an Aircraft, say, and use the features of a Seacraft, but we could not do this the other way as well. With a two-way adapter, however, this is possible.
The ITarget interface, IAircraft, has two properties, Airborne and Height, and one method, TakeOff. The Aircraft class implements the interface in the manner of an aircraft. The IAdaptee interface, ISeacraft (new in this version of the pattern), has two methods-Speed and IncreaseRevs-that are implemented by the Seacraft class. The Adapter inherits from the Adaptee (Seacraft) and implements the ITarget (IAircraft) in the normal way. The adapter then has to do some work to match these two different interfaces to each other. Table 4.1, "Adapter pattern Seabird example-methods and properties" makes it easier to see how one would approach such an adapter.
Table 4.1. Adapter pattern Seabird example-methods and properties
Aircraft (Target) |
Seacraft (Adaptee) |
Seabird (Adapter) |
Experiment (Client) |
|---|---|---|---|
Inherits Seabird, implements Aircraft |
Instantiates seabird |
||
Methods |
|||
TakeOff-sets Airborne and Height to 200 |
TakeOff-involves Seacraft, IsFlying, and IncreaseRevs |
seabird.TakeOff-goes to Seabird |
|
IncreaseRevs-changes speed by 10 |
IncreaseRevs-calls Seacraft IncreaseRevs and SeacraftIsFlying and sets the height |
(seabird as ISeacraft) IncreaseRevs-goes to Seabird |
|
Variables |
|||
Airborne-is true after takeoff |
Airborne-is true if Height > 50 |
seabird.Airborne-goes to Seabird |
|
Speed-returns the speed |
(seabird as Seacraft) Speed-goes to Seacraft |
||
Height-returns the height |
Height-returns the stored height |
seabird.Height-goes to Seabird |
The classes representing each part of the invention offer different methods: TakeOff for an aircraft and IncreaseRevs for a seacraft. In the simple adapter, only TakeOff would work. In the two-way adapter, we also capture the method from the Adaptee (IncreaseRevs) and adapt it to include information that otherwise would be supplied by the Target (the height, here).
Two-way adapters also handle variables-in this case, Airborne, Speed, and Height. Those from the Aircraft (the Target) are trapped and adapted to return locally held information. The one in the Seacraft (Adaptee) is routed through.
The result of all of the above, when translated into C# classes, is that the Client can conduct experiments on the Seabird as follows:
1 Console.WriteLine("\nExperiment 3: Increase the speed of the Seabird:");
2 (seabird as ISeacraft).IncreaseRevs( );
3 (seabird as ISeacraft).IncreaseRevs( );
4 if (seabird.Airborne)
5 Console.WriteLine("Seabird flying at height "
6 + seabird.Height +
7 " meters and speed "+(seabird as ISeacraft).Speed + " knots");
8 Console.WriteLine("Experiments successful; the Seabird flies!");
The calls to seabird.Airborne and seabird.Height (lines 4 and 6) are regular adapter methods that adapt as described in Table 4.1, "Adapter pattern Seabird example-methods and properties". However, the ability to treat the Seabird as a Seacraft as well (lines 2, 3, and 7) is peculiar to the two-way adapter. The full program is given in Example 4.2, "Two-way Adapter pattern example code-Seabird".
Example 4.2. Two-way Adapter pattern example code-Seabird
using System;
// Two-Way Adapter Pattern Pierre-Henri Kuate and Judith Bishop Aug 2007
// Embedded system for a Seabird flying plane
// ITarget interface
public interface IAircraft {
bool Airborne {get;}
void TakeOff( );
int Height {get;}
}
// Target
public sealed class Aircraft : IAircraft {
int height;
bool airborne;
public Aircraft( ) {
height = 0;
airborne = false;
}
public void TakeOff ( ) {
Console.WriteLine("Aircraft engine takeoff");
airborne = true;
height = 200; // Meters
}
public bool Airborne {
get {return airborne;}
}
public int Height {
get {return height;}
}
}
// Adaptee interface
public interface ISeacraft {
int Speed {get;}
void IncreaseRevs( );
}
// Adaptee implementation
public class Seacraft : ISeacraft {
int speed = 0;
public virtual void IncreaseRevs( ) {
speed += 10;
Console.WriteLine("Seacraft engine increases revs to " + speed + " knots");
}
public int Speed {
get {return speed;}
}
}
// Adapter
public class Seabird : Seacraft, IAircraft {
int height = 0;
// A two-way adapter hides and routes the Target's methods
// Use Seacraft instructions to implement this one
public void TakeOff( ) {
while (!Airborne)
IncreaseRevs( );
}
// Routes this straight back to the Aircraft
public int Height {
get {return height;}
}
// This method is common to both Target and Adaptee
public override void IncreaseRevs( ) {
base.IncreaseRevs( );
if(Speed > 40)
height += 100;
}
public bool Airborne {
get {return height > 50;}
}
}
class Experiment_MakeSeaBirdFly {
static void Main ( ) {
// No adapter
Console.WriteLine("Experiment 1: test the aircraft engine");
IAircraft aircraft = new Aircraft( );
aircraft.TakeOff( );
if (aircraft.Airborne) Console.WriteLine(
"The aircraft engine is fine, flying at "
+aircraft.Height+"meters");
// Classic usage of an adapter
Console.WriteLine("\nExperiment 2: Use the engine in the Seabird");
IAircraft seabird = new Seabird( );
seabird.TakeOff( ); // And automatically increases speed
Console.WriteLine("The Seabird took off");
// Two-way adapter: using seacraft instructions on an IAircraft object
// (where they are not in the IAircraft interface)
Console.WriteLine("\nExperiment 3: Increase the speed of the Seabird:");
(seabird as ISeacraft).IncreaseRevs( );
(seabird as ISeacraft).IncreaseRevs( );
if (seabird.Airborne)
Console.WriteLine("Seabird flying at height "+ seabird.Height +
" meters and speed "+(seabird as ISeacraft).Speed + " knots");
Console.WriteLine("Experiments successful; the Seabird flies!");
}
}/* Output
Experiment 1: test the aircraft engine
Aircraft engine takeoff
The aircraft engine is fine, flying at 200 meters
Experiment 2: Use the engine in the Seabird
Seacraft engine increases revs to 10 knots
Seacraft engine increases revs to 20 knots
Seacraft engine increases revs to 30 knots
Seacraft engine increases revs to 40 knots
Seacraft engine increases revs to 50 knots
The Seabird took off
Experiment 3: Increase the speed of the Seabird:
Seacraft engine increases revs to 60 knots
Seacraft engine increases revs to 70 knots
Seabird flying at height 300 meters and speed 70 knots
Experiments successful; the Seabird flies!
*/
Pluggable Adapters
Developers who recognize that their systems will need to work with other components can increase their chances of adaptation. Identifying in advance the parts of the system that might change makes it easier to plug in adapters for a variety of new situations. Keeping down the size of an interface also increases the opportunities for new systems to be plugged in. Although not technically different from ordinary adapters, this feature of small interfaces gives them the name pluggable adapters.
A distinguishing feature of pluggable adapters is that the name of a method called by the client and that existing in the ITarget interface can be different. The adapter must be able to handle the name change. In the previous adapter variations, this was true for all Adaptee methods, but the client had to use the names in the ITarget interface. Suppose the client wants to use its own names, or that there is more than one client and they have different terminologies. To achieve these name changes in a very dynamic way, we can use delegates (see later sidebar).
Now, consider Example 4.3, "Pluggable Adapter pattern theory code", which shows how to write pluggable adapters with delegates.
Example 4.3. Pluggable Adapter pattern theory code
1 using System;
2
3 // Adapter Pattern - Pluggable Judith Bishop Oct 2007
4 // Adapter can accept any number of pluggable adaptees and targets
5 // and route the requests via a delegate, in some cases using the
6 // anonymous delegate construct
7
8 // Existing way requests are implemented
9 class Adaptee {
10 public double Precise (double a, double b) {
11 return a/b;
12 }
13 }
14
15 // New standard for requests
16 class Target {
17 public string Estimate (int i) {
18 return "Estimate is " + (int) Math.Round(i/3.0);
19 }
20 }
21
22 // Implementing new requests via old
23 class Adapter : Adaptee {
24 public Func <int,string> Request;
25
26 // Different constructors for the expected targets/adaptees
27
28 // Adapter-Adaptee
29 public Adapter (Adaptee adaptee) {
30 // Set the delegate to the new standard
31 Request = delegate(int i) {
32 return "Estimate based on precision is " +
33 (int) Math.Round(Precise (i,3));
34 };
35 }
36
37 // Adapter-Target
38 public Adapter (Target target) {
39 // Set the delegate to the existing standard
40 Request = target.Estimate;
41 }
42 }
43
44 class Client {
45
46 static void Main ( ) {
47
48 Adapter adapter1 = new Adapter (new Adaptee( ));
49 Console.WriteLine(adapter1.Request(5));
50
51 Adapter adapter2 = new Adapter (new Target( ));
52 Console.WriteLine(adapter2.Request(5));
53
54 }
55 }
56/* Output
57 Estimate based on precision is 2
58 Estimate is 2
59 */
The delegate is contained in the Adapter and is instantiated on line 24, from one of the standard generic delegates. On lines 33 and 40, it is assigned to the methods Precise and Estimate, which are in the Adaptee and Target, respectively. Lines 31-34 show the use of an anonymous function to augment the results from the Adaptee. Notice that the Client (the Main method) refers only to its chosen method name, Request (see sidebar).
The pluggable adapter sorts out which object is being plugged in at the time. Once a service has been plugged in and its methods have been assigned to the delegate objects, the association lasts until another set of methods is assigned. What characterizes a pluggable adapter is that it will have constructors for each of the types that it adapts. In each of them, it does the delegate assignments (one, or more than one if there are further methods for rerouting).
C# Feature-Delegates
A method that is specified in an interface is implemented with the same name in the base class. However, such close coupling is not always appropriate. The delegate construct can be used to break the coupling for this purpose. A delegate is a type that defines a method signature. A delegate instance is then able to accept a method of that signature, regardless of its method name or the type that encapsulates it.
The delegate syntax in C# evolved considerably from Versions 1.0 to 2.0 to 3.0. We shall concentrate on the 3.0 version, which is the simplest to code. The language has predefined standard generic delegate types, as follows:
delegate R Func<R>( );
delegate R Func<A1, R>(A1 a1);
delegate R Func<A1, A2, R>(A1 a1, A2 a2);
// ... and up to many arguments
where R is the return type and the As and as represent the argument types and names, respectively. There is also another set of generic delegates introduced with the name Action that represent methods that do not return a value. Thus, declaring a delegate instance is now straightforward. For example, we can define a Request delegate that takes an integer parameter and returns a string:
| public Func <int,string> Request; |
Next, we can assign an actual method to Request, as in:
| Request = Target.Estimate; |
The delegate can then be invoked just as any other method would be:
| string s = Request(5); |
This statement would invoke the Estimate method in the Target class, returning a string.
cf. C# Language Specification Version 3.0, September 2007, Section 10.8
Example: CoolBook
Our last Adapter pattern example picks up on an earlier example that we explored with the Proxy and Bridge patterns: SpaceBook. Recall that Example 2.4, "Proxy pattern example code-SpaceBook" introduced the SpaceBook class and its authentication frontend, MySpaceBook. Then, Example 2.6, "Bridge pattern example code-OpenBook" showed how we could create a Bridge to an alternative version of MySpaceBook called MyOpenBook, which did not have authentication. Now, we are going to consider going GUI. The input and output of SpaceBook (wall writing, pokes, etc.) will be done via Windows forms. There will be a separate form for each user, and users will be able to write on each other's pages as before. However, now the input will be interactive, as well as being simulated by method calls in the program. Thus, we will have a prototype of a much more realistic system.
C# Feature-Anonymous Functions
Anonymous functions simplify the creation of one-time behavior for delegates. They are useful when additional behavior is to be added before or after a method is invoked. For example:
Request = delegate(int i) {
return "Estimate based on precision is " +
(int) Math.Round(Precise (i,3));
};
Here, the method to be invoked is Precise. The parameters are different from the ones in the delegate, as is the return value. The anonymous function can wrap up the changes and assign a "complete solution" to the delegate for later invocation.
cf. C# Language Specification Version 3.0, September 2007, Section 6.5
Creating a GUI and handling its events is a specialized function, and it is best to isolate it as much as possible from the ordinary logic of the system. In setting up CoolBook, we wrote a minimal GUI system called Interact. All Interact does is set up a window with a TextBox and a Button called "Poke," and pass any click events on the Button to a delegate (see sidebar). Separately from this, we wrote MyCoolBook, which mimics the functionality of MyOpenBook and, for reasons of simplicity at this stage, maintains its own community of users. Given the following client program, the output will be as shown in Figure 4.3, "Adapter pattern-output from CoolBook".
static void Main( ) {
MyCoolBook judith = new MyCoolBook("Judith");
judith.Add("Hello world");
MyCoolBook tom = new MyCoolBook("Tom");
tom.Poke("Judith");
tom.Add("Hey, We are on CoolBook");
judith.Poke("Tom");
Console.ReadLine( );
}
The second "Tom : Poked you" was created interactively by Tom typing in "Judith" on his wall, selecting it, and clicking the Poke button. Judith then wrote on her own wall, and was getting ready to poke Tom when the snapshot was taken.
MyCoolBook builds on top of Interact and acts as the adapter class. As can be seen in the Client, MyOpenBook and MySpaceBook have been completely plugged out and replaced by MyCoolBook. We can just change the instantiations back, and everything will revert to the old system. This is what a pluggable adapter achieves. Consider the insides of the adapter in Example 4.4, "Pluggable Adapter pattern example code-MyCoolBook". It inherits from MyOpenBook and, through inheritance, it makes use of the MySpaceBook object stored there, as well as the Name property. It reimplements the three important methods-Poke and the two Add methods-and has two methods that connect it to Interact via a form object called visuals.
Figure 4.3. Adapter pattern-output from CoolBook
.png "Adapter pattern-output from CoolBook")
Example 4.4. Pluggable Adapter pattern example code-MyCoolBook
// Adapter
public class MyCoolBook : MyOpenBook {
static SortedList<string, MyCoolBook> community =
new SortedList<string, MyCoolBook>(100);
Interact visuals;
// Constructor starts the GUI
public MyCoolBook(string name) : base(name) {
// Create the Interact GUI on the relevant thread, and start it
new Thread(delegate( ) {
visuals = new Interact("CoolBook Beta");
visuals.InputEvent += new InputEventHandler(OnInput);
visuals.FormClosed += new FormClosedEventHandler(OnFormClosed);
Application.Run(visuals);
}).Start( );
community[name] = this;
while (visuals == null) {
Application.DoEvents( );
Thread.Sleep(100);
}
Add("Welcome to CoolBook " + Name);
}
// Closing the GUI
private void OnFormClosed(object sender, FormClosedEventArgs e) {
community.Remove(Name);
}
// A handler for input events, which then calls Add to
// write on the GUI
private void OnInput(object sender, EventArgs e, string s) {
Add("\r\n");
Add(s, "Poked you");
}
// This method can be called directly or from the other
// Add and Poke methods. It adapts the calls by routing them
// to the Interact GUI.
public new void Add(string message) {
visuals.Output(message);
}
public new void Poke(string who) {
Add("\r\n");
if (community.ContainsKey(who))
community[who].Add(Name, "Poked you");
else
Add("Friend " + who + " is not part of the community");
}
public new void Add(string friend, string message) {
if (community.ContainsKey(friend))
community[friend].Add(Name + " : " + message);
else
Add("Friend " + friend + " is not part of the community");
}
}
Of the three reimplemented methods, only the behavior of the first Add is specific to MyCoolBook. The other two methods are very much like those in MyOpenBook. However, the problem is that given the closed nature of SpaceBook, MyCoolBook cannot access the dictionary there and has to keep its own community. This sometimes happens with adapters. If it were possible to rewrite parts of the Target in a more collaborative way, the adapter could do less work. This idea is addressed in the upcoming "the section called "Exercises" section. The code for the full CoolBook system is shown in the Appendix.
One point to note is the anonymous function that is passed to the Thread class in the CoolBook constructor. This is a very quick way of creating an object in a new thread. The last statement is Application.Run( ), which starts drawing the form and opens up a message "pump" for the interactive input/output on it. Finally, Start is called on the thread.
C# Feature-Events
Delegates are used extensively in Windows GUI event-driven programming, where they reflect the need to call back into the user's code when some event happens. Mostly, existing code of this type will use an older syntax. Also, because the new Func delegates must have return types, void delegates must use the original syntax too. Consider a simple example of wanting to inform one object that input has occurred in another object (this is part of Example 4.4, "Pluggable Adapter pattern example code-MyCoolBook"). We first declare a delegate visible to both classes:
public delegate void InputEventHandler(object sender,
EventArgs e, string s);
Then, in the class where the event is handled, we create an instance of the delegate and add it to the event object of the class that will receive the event. When creating the delegate, we indicate the method that it will call (in this case, OnInput):
visuals.InputEvent += new InputEventHandler(OnInput);
void OnInput(object sender, EventArgs e, string s) {
// Do something
}
The signature of OnInput must match that of InputEventHandler, which it does. Now, in the class where event occurs, we declare the event:
public event InputEventHandler InputEvent;
and in some method we invoke it:
public void Input(object source, EventArgs e) {
InputEvent(this, EventArgs.Empty, who);
}
The action of invoking the InputEvent delegate causes the method currently assigned to it (here, OnInput) to be invoked. Thus, the callback from one class to the other is achieved.
More than one method can be associated with a delegate; when such a delegate is invoked, all its methods are called. Thus, if another object needed to know about input in the preceding example, it could add its own handler method on to InputEvent using +=. Event handlers can be removed using -=.
cf. C# Language Specification Version 3.0, September 2007, Section 10.8
Use
The Adapter pattern is found wherever there is code to be wrapped up and redirected to a different implementation. In 2002, Nigel Horspool and I developed a system called Views that enabled an XML specification of a Windows GUI to run on the cross-platform version of Windows called Rotor. The Views engine ran with the GUI program and adapted Views method calls to Windows calls. That benefited the clients (students) because they could use a simpler interface to GUI programming. A subsequent advantage was that Vista, the successor to Windows XP, used the same approach.
At the time, it was a long way around to get Windows forms, but the adaptation paid off later. In 2004, the backend of the Views engine was ported by Basil Worrall to QT4, a graphics toolkit running on Linux. Immediately, all applications that were using Views for GUI programming became independent of Windows and could run with the Mono .NET Framework on Linux. The Views engine was therefore a pluggable adapter. (Our paper describing the approach is referenced in the Bibliography at the end of the book.)
Use the Adapter pattern when... |
|---|
You have:
|
You want to:
|
Choose the Adapter you need...
|
Exercises
Consider the Seabird program. Would it be possible to instantiate an Aircraft object instead of a Seacraft object and change the methods inside Seabird accordingly? If so, make the changes. If not, explain how the present program would need to be altered to enable this and then make the changes.
Add a "SuperPoke" button to CoolBook, enabling one user to send a message to another.
Having two different communities for SpaceBook and CoolBook is clearly a disadvantage. Assume you can make minor changes to the SpaceBook, MySpaceBook, and MyOpenBook classes, and see whether you can remove the community collection from MyCoolBook, routing all accesses back through MyOpenBook to SpaceBook.
Façade Pattern
Role
The role of the Façade pattern is to provide different high-level views of subsystems whose details are hidden from users. In general, the operations that might be desirable from a user's perspective could be made up of different selections of parts of the subsystems.
Illustration
Simple interfaces to complex subsystems abound in real life. They can be created to make frequent use of a system faster, or to differentiate between novices and power users. A good illustration is Amazon.com's 1-Click® system (Figure 4.4, "Façade pattern illustration-1-Click®"), which simplifies the process of ordering items for well-known customers. Normally, after selecting an item to purchase, an Amazon customer has to enter delivery and bank account details before the order is accepted. If these details are stored and the customer verifies her identity in some way, 1-Click takes that customer straight to the checkout. The customer's stored bank account details and selected delivery address are used for the purchase, thus considerably speeding up (and simplifying) the ordering process. Thus, the 1-Click option forms a façade to the fuller system underneath.
Figure 4.4. Façade pattern illustration-1-Click®
.png "Façade pattern illustration-1-Click®")
Design
Hiding detail is a key programming concept. What makes the Façade pattern different from, say, the Decorator or Adapter patterns is that the interface it builds up can be entirely new. It is not coupled to existing requirements, nor must it conform to existing interfaces. There can also be several façades built up around an existing set of subsystems. The term "subsystem" is used here deliberately; we are talking at a higher level than classes. See the UML diagram in Figure 4.5, "Façade pattern UML diagram"; it considers the subsystems to be grouped together, so they can interact in agreed ways to form the top-level operations.
Figure 4.5. Façade pattern UML diagram
.png "Façade pattern UML diagram")
The roles are:
Namespace 1
A library of subsystemsSubsystem
A class offering detailed operationsFaçade
A class offering a few high-level operations as selections from the subsystemsNamespace 2
Where the client residesClient
Calls the high-level operations in the Façade in Namespace 1
As shown in the UML diagram, the client's code does not make reference to the classes of the names of the subsystems; it only gets access to their operations via the Façade.
Implementation
The Façade pattern is simple to implement. It uses the C# concept of namespaces. Classes in namespaces have the facility to define accessibility as internal or public. If accessibility is defined as internal, the member is visible only in the assembly in which the namespace is compiled. In a very large system, the client's GUI will be in a different namespace from the library, so we can enforce the Façade. (Alternative implementations of the Façade pattern will be discussed shortly.)
QUIZ: Match the Façade Pattern Players with the 1-Click Illustration
To test whether you understand the Façade pattern, cover the lefthand column of the table below and see if you can identify its players among the items from the illustrative example (Figure 4.4, "Façade pattern illustration-1-Click®"), as shown in the righthand column. Then check your answers against the lefthand column.
| Namespace 1 | Amazon.com server |
| Subsystems | Registering, checkout, address entering, account details verification, etc. |
| Facade | Window offering choices for purchasing |
| Operation | 1-Click |
| Namespace 2 | Client GUI |
| Client | Customer |
In Example 4.5, "Façade pattern theory code", the theory code comes from two files: Façade-Main.cs and Façade-Library.cs. Both have to be compiled with special directives so that library in Façade-Library.cs is recognized as a lib file and the client in Façade-Main.cs can reference it. These commands are:
// Compile the library
csc /t:library /out:FacadeLib.dll Facade-Library.cs
// Compile the main program
csc /r:FacadeLib.dll Facade-Main.cs
Tip
This process of compiling and using libraries is facilitated in environments such as Visual Studio.
The example mirrors the diagram in Figure 4.5, "Façade pattern UML diagram". Three subsystems, implemented as classes, are inserted into the library. Facade is a static class that instantiates the three subsystems under the façade as objects called a, b, and c. Operations 1 and 2 then select combinations of methods from a, b, and c. For example, Operation1 will call two methods in a, one in b, and none in c. Thus, the Facade is a means of providing an interface to operations that should, on their own, remain hidden.
The client starts with a using statement that indicates it wants access to the public members in the FacadeLib namespace. Then it calls the Facade's two high-level operations. The output is shown below in Example 4.5, "Façade pattern theory code".
Example 4.5. Façade pattern theory code
using System;
// Facade Pattern Judith Bishop Dec 2006
// Sets up a library of three systems, accessed through a
// Facade of two operations
// Compile with
// csc /t:library /out:FacadeLib.dll Facade-Library.cs
namespace Library {
internal class SubsystemA {
internal string A1( ) {
return "Subsystem A, Method A1\n";
}
internal string A2( ) {
return "Subsystem A, Method A2\n";
}
}
internal class SubsystemB {
internal string B1( ) {
return "Subsystem B, Method B1\n";
}
}
internal class SubsystemC {
internal string C1( ) {
return "Subsystem C, Method C1\n";
}
}
}
public static class Facade {
static SubsystemA a = new SubsystemA( );
static SubsystemB b = new SubsystemB( );
static SubsystemC c = new SubsystemC( );
public static void Operation1( ) {
Console.WriteLine("Operation 1\n" +
a.A1( ) +
a.A2( ) +
b.B1( ));
}
public static void Operation2( ) {
Console.WriteLine("Operation 2\n" +
b.B1( ) +
c.C1( ));
}
}
// ============= Different compilation
using System;
using FacadeLib;
// Compile with csc /r:FacadeLib.dll Facade-Main.cs
class Client {
static void Main ( ) {
Facade.Operation1( );
Facade.Operation2( );
}
}/* Output
Operation 1
Subsystem A, Method A1
Subsystem A, Method A2
Subsystem B, Method B1
Operation 2
Subsystem B, Method B1
Subsystem C, Method C1
*/
Everything in the façade has to be public so that the Client, which is compiled into a different assembly, can access it. The classes all have the default internal visibility, limiting access to them to the assembly in which they were compiled (excluding the Client). As a test, if we try to let the Client instantiate any of the subsystems directly, we will get an error like the following:
SubsystemC x = new SubsystemC( );
x.C1( );Facade2Main.cs(12,3): error CS0122: 'FacadeLib.SubsystemC' is inaccessible
due to its protection level
Façade Alternatives
Some alternative implementations of the Façade pattern are:
Transparent façades
The façade described in the preceding example is opaque, in that the subsystems cannot be accessed except via the Facade object. This requirement might be too stringent. Suppose some users want to get at the individual operations of particular subsystems. We can change all the internal modifiers to public, which will make the façade optional, or transparent. That is, as well as being able to go through the Facade, the client will be able to instantiate SubsystemA directly, for example, and then call A1.Static façades
In most cases, there will only be one instance of a façade in the client for a set of subsystems, so its operations could more appropriately be called on the user's side as members of the class itself, as in:public void ClientMain ( ) { Facade.Operation1( ); Facade.Operation2( ); }This implies that Facade is a static class. No instantiation is necessary; the user interfaces with the Facade class directly. In fact, the Singleton pattern (Chapter 5, Creational Patterns: Prototype, Factory Method, and Singleton) would be the preferred way of achieving this effect.
Example: Novice Photo Library
Consider the Composite pattern example in Chapter 3, Structural Patterns: Composite and Flyweight that showed how photos could be loaded into directories of arbitrary configurations. The instructions for using the six commands relied on the current place ("where I am"), which was a powerful, but perhaps confusing, concept. For novices, it might be a good idea to abstract from the power of the Photo Library and just let them load sets of photos, all at the same level, and immediately display them (as Flickr does). The commands could simply be:
Upload setname
photoname1 photoname2 ...
ending with a blank line or some other indicator. These instructions would translate into the following existing ones:
Find album
AddSet setname
AddPhoto photoname1
AddPhoto photoname2
...
Display
Instead of going in and altering the code to have a new command, we can have a completely separate Façade that makes the calls as described above. The more complex and rich commands might be available to expert users, but not to novices. (See the preceding discussion on opaque and transparent façades.)
Use
Façades can be useful in different circumstances. There are many instances where a computer system is built up out of a set of largely independent subsystems. One well-known case is a compiler: it consists of clearly identifiable subsystems called a lexical analyzer, a syntax analyzer, semantic analyzers, a code generator, and several optimizers. In modern compilers, each subsystem has many subtasks. The different tasks and subtasks are called in a sequence, but sometimes they are not all needed. For example, if an error occurs in one of the analysis tasks, the compiler might not go onto a later phase. (The .NET compilers follow this approach.) Hiding this detail behind a façade enables a program to call tasks within subsystems in a logical order, passing the necessary data structures between them.
Use the Façade pattern when... |
|---|
A system has several identifiable subsystems and:
|
But consider using instead:
|
Choose the Façade you need...
|
Exercises
Program the suggested extension for novices to the Photo Library program.
Consider large systems that you use on the Internet, and come up with more examples of Façades.
If you have access to source code for a compiler, find the part where the subsystems are called and examine how the data structures are passed between them.
Pattern Comparison
The Adapter pattern has much in common with the patterns discussed in Chapter 2, Structural Patterns: Decorator, Proxy, and Bridge. The differences are in the intents of the patterns. A bridge, for example, separates an interface and its implementation so that they can vary independently, whereas an adapter changes the interface of an existing object. The Adapter pattern is more useful for one-off changes, whereas the Bridge pattern anticipates that change might happen continuously.
A decorator enhances an object without changing its interface and should be transparent to the application. An adapter is not transparent, as it is the named implementation of the interface the client sees. The Proxy pattern does not change any interfaces; it defines substitute objects for other objects.
From a certain point of view, the Façade pattern is also adapting requests: it transforms high-level requests into a sequence of lower-level requests. The Façade's intent is to hide complexity, and the Façade subsystems are not intended to be accessible by the client.
To complete the picture, we can classify the Adapter and Façade patterns according to the mechanisms shown in Table 4.2, "Comparison of Adapter and Façade patterns".
Table 4.2. Comparison of Adapter and Façade patterns
Mechanism |
Adapter |
Façade |
|---|---|---|
Original |
Adaptee |
SubsystemA, B, and C |
Interface |
ITarget |
Façade |
New |
Adapter |
Operation1 and 2 |
Client |
Aggregates ITarget |
Accesses Façade |
Client activates |
New |
New |
Original changed by |
No change |
No change |
New classes/subsystems |
Adapter provides adaptations to their methods |
Façade supplies high-level operations |
Operation routed |
From new to original |
From new to original |
[3] * For more about this migration, read the "Introduction to AltiVec/SSE Migration Guide" at http://developer.apple.com.
[4] * Windows runs on a Mac with the help of the Parallels or BootCamp virtual machines.