Presentation is loading. Please wait.

Presentation is loading. Please wait.

Summary prepared by Kirk Scott

Similar presentations


Presentation on theme: "Summary prepared by Kirk Scott"— Presentation transcript:

1 Summary prepared by Kirk Scott
Unit 21 Abstract Factory Summary prepared by Kirk Scott

2 Design Patterns in Java Chapter 17 Abstract Factory
Summary prepared by Kirk Scott

3 The Introduction Before the Introduction
The book provides several examples of abstract factory with context Sometimes the context seems to obscure the basic pattern The overheads will try to distinguish between the background information and the pattern

4 The Book’s Definition of the Pattern
Book definition: The intent of the Abstract Factory, or Kit, is to allow creation of families of related or dependent objects.

5 Comment mode on, 1: Even though the pattern name includes the word abstract, the book’s examples don’t include an abstract class Instead, the examples show a concrete superclass and a subclass of it It would be possible to do the same thing with an abstract class and two concrete subclasses

6 Comment mode on, 2: The book examples show using an abstract factory to help develop a user interface Parts of this chapter are related to some of the ideas that came up with facades The example abstract factory can viewed as sort of a façade into the Java API, making its GUI elements easier to use It is a façade that is limited to creating objects

7 Background for the Book’s Example
The book starts the example this way: In a base UI class you define methods that will return instances of graphical user interface components that have certain features

8 For example, an application may require a button
Rather than having the application construct an instance of JButton directly, it acquires a JButton object by calling a UI method The method that constructs the JButton takes care of the construction parameters that will be used for the button

9 This grows into the Abstract Factory design pattern in this way:
What if you would like it to be possible for the application to seamlessly use JButtons of a different form? Then you extend the UI class, overriding those methods that create objects so that the visual aspects of the objects are changed

10 If the application is typed to use references of the UI superclass, it can also use objects of the UI subclass What kind of JButton the client gets depends on which UI class it uses It will get the kind of JButton returned by the method in that UI class

11 Factory Method and Abstract Factory
A very large scale way of seeing what’s going on is this: With a factory method, you wanted something that implemented an interface. A number of classes satisfied this requirement, and the logic for deciding what to give the client was in the service code.

12 With abstract factory, there is a hierarchy of classes containing sets of overridden methods that return various objects Each of the methods is basically a factory method Each method will return an object that meets the requirement of the client.

13 The client may request which member of the family of classes it would like to make use of
With this choice, a set of methods becomes a available Which kind of object comes back from a method call depends on the implementation of the method in the class that was chosen

14 Builder, Factory Method, and Abstract Factory
Abstract factory falls in a spectrum with the builder and factory method patterns. With builder, a method call returned one kind of desired, specified object. With factory method, a method call returned a reference to one of a set of kinds of objects back, with no control over which one.

15 With abstract factory it is possible to specify which one of a set of methods for generating objects is desired. Those methods may be simple methods where the return type is a simple class reference. Those methods may also be factory methods which return objects which agree with a required supertype, but which may actually be instances of a subtype.

16 A Classic Example: GUI Kits
A GUI kit is an implementation of the Abstract Factory design pattern The GUI kit supplies components to client applications These components are things like buttons or other graphical user interface components The kit constructs the component objects, passing in parameters so that together the objects have a consistent look and feel

17 A simple illustration of the abstract factory idea is given in the UML diagram on the next overhead
The UI class implements create() methods that have to do with the visual appearance of an application The diagram includes a subclass, BetaUI The same set of methods would be provided in the subclass, but they would return objects with different visual characteristics

18

19 It might be a fuller illustration of the pattern overall if there were an abstract superclass AbstractUI Then this could have multiple subclasses, including NormalUI and BetaUI However, the book has chosen to do it with a simple superclass and subclass, so the example will be pursued in that way

20 BetaUI inherits from NormalUI, overriding methods to create a different user interface
The UML diagram on the following overhead illustrates the idea that a client application has a reference to an instance of the UI class If the Visualization application had a reference to a BetaUI object instead of a UI object, the visualization would get a different look and feel

21

22 Making the Example Concrete
The screen shot on the next overhead shows the graphical user interface for a Visualization application The application interface supports: Adding icons representing machines, Clicking and dragging them around the work area, Undoing a recent action, like adding a machine

23

24 The screen shot on the next overhead shows the application with slight changes in the user interface
The add and undo button images have been changed from rockets to cherries (cherry bombs) The text under the images has been made italic These changes are accomplished by creating a subclass of UI which has these different characteristics coded into it

25

26 UI Class Code The UML diagram for the UI class showed these methods:
createButton(), getFont(), createPaddedPanel(), getIcon()

27 This is only a subset of the methods the UI class would contain
In continuing the example, the book considers these methods: createButton(), createButtonOk(), and createButtonCancel() Code for these methods is given on the following overheads

28 public JButton createButton()
{ JButton button = new JButton(); button.setSize(128, 128); button.setFont(getFont()); button.setVerticalTextPosition(AbstractButton.BOTTOM); button.setHorizontalTextPosition(AbstractButton.CENTER); return button; }

29 public JButton createButtonOk()
{ JButton button = createButton(); button.setIcon(getIcon(“images/rocket-large.gif”)); button.setText(“Ok!”); return button; }

30 public JButton createButtonCancel()
{ JButton button = createButton(); button.setIcon(getIcon(“images/rocket-large-down.gif”)); button.setText(“Cancel!”); return button; }

31 The creation of the OK and the Cancel buttons make use of the creation of a simple button
The use of the rocket icon/image is hardcoded into the characteristics of the buttons In the visualization application, these button creation methods are used to create visual components

32 Take the createButtonCancel() method for example
You will notice that it seems to be somewhat like the “undo” button from the GUI, but with different text Shortly we will see how the cancel button code is related to the undo button

33 Client Code The constructor for the Visualization client takes an instance of the UI class as a parameter Also, the visualization client has a (protected) instance variable named undoButton of the type JButton

34 The following overhead shows code that belongs to the visualization client
The client contains a method named undoButton() which returns an instance of a JButton This client method generates a button by making this call on the ui object to one of the factory methods in the UI class: ui.createButtonCancel()

35 /* Notice the protected method. This is creepy and
disappointing, but not surprising. */ protected JButton undoButton() { if(undoButton == null) undoButton = ui.createButtonCancel(); undoButton.setText(“Undo”); undoButton.setEnabled(false); undoButton.addActionListener(mediator.undAction()); } return undoButton;

36 ui is the instance variable which refers to the UI object belonging to the client
The look and feel of the visualization can be changed by passing the visualization an instance of a different UI when it’s constructed The difference in the createButtonCancel() method would propagate as a difference in the undoButton in the client

37 This means the appearance of the visualization client can be changed without making any changes in its code All you have to do is make the new UI class a subclass of the old one You can always pass in a subclass object when the parameter is typed to a superclass

38 Code for the BetaUI Subclass
BetaUI is the subclass in the example The requirements for BetaUI are these: The image should be a cherry instead of a rocket The font should be italic On the next overhead, the constructor for BetaUI is given with some commentary

39 public class BetaUI extends UI
{ public BetaUI() Font oldFont = getFont(); font = new Font( oldFont.getName(), oldFont.getStyle() | Font.ITALIC, oldFont.getSize()); } /* Notice that the plan is to make use of the superclass as much as possible. In addition to relying on inherited methods, something new is being done here. The font instance variable is inherited in BetaUI. Its value in the subclass depends on the inherited value, but one of its characteristics is being changed. */

40 The rest of the code for BetaUI would be the createButton…() methods
The book does this as a challenge As usual, it makes sense to just look at the solution provided, shown on the following overhead The calls to super do most of the work This rest of the code makes the modifications for the new BetaUI

41 Solution 17.1 public JButton createButtonOk() {
JButton b = super.createButtonOk(); b.setIcon(getIcon(“images/cherry-large.gif”)); return b; } public JButton creatButtonCancel() Jbutton b = super.createButtonCancel(); b.setIcon(getIcon(“images/cherry-large-down.gif”));

42 Passing a client an instance of the BetaUI causes it to take on those display characteristics
public class ShowBetaVisualization { public static void main(String[] args) Jpanel panel = new Visualization(new BetaUI()); SwingFacade.launch(panel, “Operational Model”); }

43 Shortcomings of the Design
The book states that this example illustrates the usefulness of the design pattern However, it also states that the example has some shortcomings The shortcomings will be given on the following overheads After that the book will suggest an alternative design

44 One shortcoming as stated in the book:
As written, the BetaUI class depends on the ability to override the creation methods It certainly seems up to this point that this was done intentionally Based on this observation alone, it’s not clear how this is a shortcoming It remains to be seen whether or how an alternative would avoid this

45 Another shortcoming, as stated in the book:
The subclass methods need access to protected instance variables, like font, from the superclass First of all, I’m not a fan of the protected modifier Whether you use it or not, you still should have access with get and set methods It will not be clear exactly what the book means until it presents its alternative

46 Challenge 17.2 “Suggest a design change that would still allow for the development of a variety of GUI control factories but that would reduce the reliance of subclasses on method [access] modifiers in the UI class.”

47 Comment mode on: I have included the book’s challenge statement because it mentions method modifiers It is worth noting that they are not just concerned with protected instance variables This goes back to the point in the code where they declared a method protected

48 Solution 17.2 “One solution for producing a more resilient design would be to specify the expected creation methods and standard GUI properties in an interface, as Figure B.20 shows.” [See the next overhead.]

49

50 Defining a UI Interface is a Good Idea
Comment mode on: I agree that defining an interface is a good idea

51 What the Book Code’s Problem Really is, and Whether the Interface Solves Them
The real problem with the book’s code is the use of the protected access modifier In the new design the UI class has to implement the interface The implementing class has to implement the methods The implementing class has to provide any needed instance variables

52 Interface Methods are Public
Interface methods are public by default If protected methods are an issue, this is solved by using the interface The question then becomes, why did you use protected methods to begin with?

53 Instance Variables Should Be Private
Instance variables provided by an implementing class don’t have a required access modifier It’s the programmer’s choice, package, public, protected, or private Private is the best choice The question then becomes, why did you use protected instance variables to begin with?

54 The New Design Still Has Inheritance
In the new design, BetaUI is still a subclass of UI It meets the requirements for the interface by inheritance but would not have direct access to inherited private instance variables If BetaUI implemented the interface directly BetaUI would have to provide the instance variables itself

55 A better alternative might be to define an abstract class
Both NormalUI and BetaUI could extend that class Some things could be implemented in the superclass and inherited Other things could be implemented in the subclasses or overridden

56 Abstract Factory and Factory Method
The book observes that in a sense an abstract factory class is like a collection of factory methods In the factory method pattern there was a client that used something typed to an interface There was an interface There were classes that implemented the interface

57 There was a factory class that contained a factory method
That method contained the logic for deciding which kind of object to give the client when it requested one The client called the method and received back a reference to an object that implemented the interface The client didn’t have to know exactly what kind of object was returned to it

58 The goal now is to show how the abstract factory pattern functions as a collection of factory methods Each UI class contains a collection, or family, of methods that return objects It’s true that the client doesn’t “know” exactly what kind of object it’s getting back

59 For example, it may get back a JButton reference when the object could be an instance of a subclass of JButton On the other hand, there is some choice in what comes back depending on which UI class is used The logic for deciding which UI class to use can be on the client side rather than the service side

60 Abstract Factory and Façade
Abstract Factory has something in common with the façade pattern It’s like a façade that is restricted to methods that return objects The abstract factory is kind of like a utility Its methods aren’t literally static, but it’s a situation where you are accomplishing something like static methods with a single instance of the UI class

61 The Next Examples from the Book
The book has several examples and challenges The remaining overheads for this chapter may be in book order in some places In other places, they may be rearranged and the explanations adjusted to try and make things clearer

62 As part of the example, the book starts separating classes into different packages
Mixed in with the information about abstract factory there is reasoning about what classes belong together and why This is a worthwhile topic It’s a topic that isn’t covered in CS 202

63 Developing an Abstract Factory Example from the Credit Check Factory Method Example
The next example is built on the factory method example of the previous unit Here is a thumbnail review of the factory method example The CreditCheckFactory has a method createCreditCheck() that returns a reference to an object that implements the CreditCheck interface Two classes, CreditCheckOnline and CreditCheckOffline implement the interface

64 The code of the createCreditCheck() method knows which kind of object to create based on “server” code conditions Client code will receive a reference to one of the two different kinds of objects Client code doesn’t have to to specify or care which kind of object comes back The UML diagram for this is given on the next overhead

65

66 Expanding the Example to the Abstract Factory Pattern
You may realize that you would like to apply the abstract factory pattern when several related factory methods arise in an application design The idea is that there is a group of related objects that need to be created and you put the methods for creating them together in a common class

67 The book now adds things called billing checks and shipping checks to the credit check example
It doesn’t really explain what these things are, but it’s apparent that as checks, they are classified with credit checks

68 An intermediate UML diagram is given on the next overhead
It introduces the idea that you can start putting related things together in packages The diagram is intermediate because it doesn’t have enough information in it to show how it illustrates the abstract factory design pattern Specific comments on that will follow the diagram

69

70 In the preceding diagram there were two new classes, ShippingCheck and BillingCheck
Although they are checks of some sort, they don’t seem to be new types of credit checks They don’t implement the CreditCheck interface

71 Not surprisingly then, the CreditCheckFactory doesn’t have a method which constructs objects of these two new classes So the question remains, what are they exactly, and what role will they play in the abstract factory design pattern?

72 Another UML diagram is given on the following overhead
This is the book’s next step in explaining the design pattern—but it still doesn’t give the whole picture In this diagram, the checks have become interfaces Also, the CreditCheckFactory includes methods that return values that are of the types billing check, shipping check, and credit check

73

74 Review of the Contents of the Diagram
The preceding overhead was a UML diagram of the com.oozinoz.credit package, which contained these elements: A CreditCheck interface A CreditCheckOffline class that implements the interface Two more interfaces, BillingCheck and ShippingCheck A CreditCheckFactory class containing methods to create instances of each kind of check

75 Questions about the Design so Far
The diagram is clearly still not complete. Only the CreditCheck interface has an implementing class. There are no classes that implement the BillingCheck or ShippingCheck interfaces The create() methods in the CreditCheckFactory have no classes which they could construct instances of

76 Also, there is just the one CreditCheckFactory
If there are to be families of methods, then there should be more than one factory class It also turns out that the CreditCheckFactory is now misnamed The factory class has methods for all 3 kinds of checks, not just credit checks

77 The issue also remains whether the CreditCheckFactory is a plain concrete class with subclasses
Or would it be better to make it an interface or an abstract class

78 Bringing the Example to Completion—Adding Packages
The book brings the example to completion in more or less the following way: There will be no direct instances of the CreditCheckFactory class in the credit package There will be other packages, one for each country, such as the U.S. or Canada

79 In these packages there will be classes that implement the interfaces in the credit package
There will also be a factory class which extends the CreditCheckFactory class in the credit package The UML diagram on the following overhead illustrates the relationship between the credit package and the Canada package

80

81 The Canada package contains live classes BillingCheckCanada, ShippingCheckCanada, and CreditCheckCanadaOnline which implement the check interfaces The package contains the CheckFactoryCanada class which extends the CreditCheckFactory class The contents of the CheckFactoryCanada class aren’t shown in detail in this diagram However, it will either inherit or override the methods which return either a billing check, a shipping check, or a credit check

82 Comment mode on: I think the example would be better if the CreditCheckFactory class were abstract I also think it would be better if that class were renamed simply CheckFactory Billing and shipping checks don’t seem to be kinds of credit checks, but methods to create them are included in the factory

83 CreditCheckFactory only has the name it does because it was named before the example was expanded
Notice that the subclass is named CheckFactoryCanada, not CreditCheckFactoryCanada In other words, implicitly the authors recognize that at the implementation level this is a check factory, not a credit check factory

84 Also, if you read the text closely, you’ll discover that at one point the authors refer to the CheckFactoryCanada class as a concrete factory, not an abstract factory This suggests that they recognize that the plain CreditCheckFactory doesn’t have to be a concrete class

85 Continuing to Outline the Example
No separate package is given for the U.S., but conceptually the example is set up as if there would also be a concrete package for the U.S., analogous to the Canada package There could also be separate packages for other countries, like Mexico, etc.

86 The original diagram is shown again on the next overhead.
Take a moment to consider what it would look like if a package for the U.S. were added to it

87

88 If set up as I would set it up, the plain credit package would contain an abstract CheckFactory class and the packages for specific countries would each contain a concrete check factory class that extended it

89 There is a reason that the CreditCheckOffline class in the credit package is already a concrete class that implements the CreditCheck interface The CreditCheckOffline is going to be the same for both Canada and the U.S. and can be shared by the code for both That can be taken care of with one implementing class in the parent package The original factory method pattern is still at work because there are two kinds of credit check, online and offline

90 The abstract factory pattern is like a collection of factory methods
It is after the U.S. package is added to the picture that this becomes apparent createBillingCheck() illustrates the idea There are two kinds of billing check, Canadian and American There are different classes for each, but they both implement the common BillingCheck interface

91 createBillingCheck() in the Canada package implementation will return something of type BillingCheck
There would also be a createBillingCheck() method in the U.S. package It would also return an object of type BillingCheck This creation method along with the others together form a family of methods for each country

92 There could be an additional layer of code on the server side of the example
That layer would contain if/else logic to determine whether a customer that a billing check was being run on was Canadian or American

93 In other words, there might be a method whatKindOfCustomer() in this example
That would be analogous to isAgencyUp() in the factory method example of the previous chapter The actual kind of billing check object returned to a client would depend on the outcome of a call to whatKindOfCustomer()

94 The ultimate client code would simply be handed a reference to an object that implemented the BillingCheck interface The client wouldn’t have to be aware or specifically request a billing check for one country or the other Such a solution would make these abstract factory methods quite analogous to plain old factory methods, with the decision about what to return embedded on the server side.

95 However, the application overall could also be structured more like the GUI example
On the client side would be handed a reference to a customer of a certain nationality, or would itself create one Then what kind of billing check came back would simply be determined by polymorphism when createBillingCheck() was called The server side wouldn’t explicitly be in charge of determining exactly what was returned

96 The Implementation of the CheckFactoryCanada Class
I’m skipping this and everything else in the book up until its comments on packages. The overheads for the intervening stuff have been saved after the “The End” overhead, but there is no need to look at them. Enough is enough…

97 Packages and Abstract Factories
The book now returns to the question of packages What more they actually have to say about packages is contained in the next challenge

98 Challenge 17.5 Write down an argument supporting the decision to place each factory and its related classes in a separate package. Or, argue that another approach is superior.

99 Solution 17.5 An example justification is: Placing country-specific classes in separate packages helps our Oozinoz developers to organize our software and our development efforts. By placing the classes for each country in a separate package, we keep country-specific packages independent of one another.

100 We can be confident, for example, that U. S
We can be confident, for example, that U.S.-specific classes have no impact on Canada-specific classes. We can also easily add support for new countries. For example, when we start doing business with Mexico, we can create a new package that provides the check services we need in a way that makes sense in that country.

101 This has the further advantage of letting us assign the credit
This has the further advantage of letting us assign the credit.mx package to a developer who has expertise in working with services and data from Mexico.

102 An argument against: Although this separation is nice in theory, it’s overwrought in practice. I’d rather have only one package with all the classes in it, at least until we expand to nine or ten countries. Spreading these classes over multiple packages winds up causing me three or more times the configuration-management work when I need to implement a change that cuts across all these packages.

103 Comment mode on: Eventually, as software grows, you may start to use packages If so, you’ll have to decide how to divide them up. When the time comes, you’ll have to make your own decision about the right approach

104 The book’s argument “against” summarizes my feelings about packages at this stage
Up through CS 202 and CS 304 there are no programming projects so complex that using packages is justified Trying to use them just introduces an artificial level of complexity Why, for example, Eclipse generates packages for a simple program consisting of a couple of classes is beyond me…

105 Another Example The “other” example for the Abstract Factory design pattern builds on the “other” example for the Factory Method design pattern. Because this set of overheads is already so long, no code will be given. Screen shots will be given, along with a UML diagram.

106 On the following overhead is the initial dialog box for the application.
The client is allowed to choose what interface is wanted—i.e., what family of classes is desired, the fruit classes or the mushroom classes. After that two overheads are shown, one for each of the two interfaces.

107

108

109

110 UML for the Pattern Although not really complete, the book’s last diagram is repeated on the following overhead. What’s missing is some indication that the factory makes use of the check classes What’s also missing is another set of classes representing another family, for example, a set of classes for the United States, Mexico, etc.

111

112 UML for the Other Example
On the following overhead I show a subset of the UML diagram for the fruit and mushroom example. This is an attempt to show factories that are responsible for creating families of fruits and mushrooms.

113

114 Lasater’s UML Diagram Lasater’s diagram closely follows my diagram in structure, reinforcing the idea that that is a reasonable way to view the pattern He does use an abstract class for the abstract factory, which may be a plus He also uses generic names that aren’t tied to a specific example.

115

116 Summary What follows is a fairly faithful summary of what the book includes in the summary, interspersed with commentary The Abstract Factory pattern lets you arrange for a client to create objects that are part of a family of related dependent objects Comment mode on: It is clear from the examples that the objects are related The book hasn’t explained specifically what it means when it now says that they’re dependent

117 One example of related objects is groups of objects for different GUIs
Another example of related objects is groups of objects for different countries

118 As with Factory Method, Abstract Factory isolates clients from knowing which class to instantiate.
Comment mode on: This can be true, but the examples in this chapter didn’t grow to the point of illustrating it specifically For example, there was no server side code that checked which country a check request was being made for

119 Comment mode, continued:
Going back to the earlier example there was just one example of using an abstract factory It contained this call: JPanel panel = new Visualization(new BetaUI()); Whatever else you might say, it’s clear that this code had to specify which UI to use The only question is whether code of this form would appear on the client side or on the server side of the application overall

120 Comment mode, continued:
No example client code was given for the Canadian credit check example However, the situation is analogous to the previous one At some point a CheckFactoryCanada object would have to be created so that the unique implementations of the methods in it could be called The question is whether the creation of the object would be in client code or in server code

121 The End

122 The Implementation of the CheckFactoryCanada Class
Examining the code clarifies to a degree what is going on The only concrete method that would be necessary in the CreditCheckFactory class of the credit package would be isAgencyUp() The implementations of the methods in the CreateCheckCanada class that create checks would be country specific and override methods of the same name in the CreditCheckFactory class This is further evidence that the superclass should be abstract…

123 The book presents writing the code for CreateCheckCanada as challenge 17.4
The complete code is given on the following overheads

124 package com.oozinoz.ca;
import com.oozinoz.credit.*; public class CheckFactoryCanada extends CheckFactory { public BillingCheck createBillingCheck() return new BillingCheckCanada(); } public CreditCheck createCreditCheck() if(isAgencyUp()) return new CreditCheckCanadaOnline(); return new CreditCheckOffline(); public ShippingCheck createShippingCheck() return new ShippingCheckCanada();

125 Solution 17.4, book stuff continued
Your solution should Implement create- methods for methods inherited from the abstract CreditCheckFactory class Comment mode on: This indicates again what has gone wrong with the book example: They didn’t show that class and its methods as abstract in the UML diagram

126 Solution 17.4, book stuff continued
[Your solution should] Have the proper interface for the return type of each create- method Return a CreditCheckOffline object if the agency is down Comment mode on: This is obvious from the code already shown…

127 The UI Example vs. the Credit Check Example
After all this, if you’re still following my attempt to explain why the book’s presentation is muddled, consider these points In the UI example they had a concrete UI superclass for a given “standard” interface Then they added a concrete BetaUI subclass for an alternative interface

128 There was no literally abstract factory class in that example
It might have been better if they had had an abstract superclass with two concrete subclasses, UI and BetaUI

129 In the current, credit check example, they claim that there would be three packages: the credit package, the Canadian package, and a U.S. package They only show things for the credit and Canadian package

130 The credit package contains a concrete class, not an abstract class
This is what I think they did, probably unintentionally They started with a package that would have worked for the U.S. alone It would have contained concrete classes for shipping, billing, and credit checks It also would have contained a concrete factory class

131 In other words, the credit/U. S
In other words, the credit/U.S. factory class would have been analogous to the UI class in the previous example Then the Canadian factory class would have been analogous to the BetaUI class in the previous example They started changing over and only got halfway

132 They changed the checks to interfaces, but they left the concrete check factory (with its old name) in place If this explanation is correct, it’s understandable how the mix-up could occur Unfortunately, it makes things overall harder to sort out

133 Final note on this topic:
At one point in the previous materials the book referred to the CreateCheckCanada class as a concrete factory By the end of the chapter it’s back to referring to the pattern overall as abstract factory And as a consequence, even though the class in the Canadian package was concrete, it is referred to as an abstract factory class

134


Download ppt "Summary prepared by Kirk Scott"

Similar presentations


Ads by Google