Summary prepared by Kirk Scott Chapter 23 Strategy Summary prepared by Kirk Scott

Design Patterns in Java Chapter 23 Strategy Summary prepared by Kirk Scott

The Introduction Before the Introduction In general, a single strategy might be thought of as an algorithm or an operation In the context of the Strategy design pattern, the idea is that there are multiple approaches to doing something, depending on certain conditions or context The Strategy design pattern, then, depends on picking the approach or picking the strategy

When there is more than one way to go about doing something, complexity can result There are the implementations of the different strategies There is also the code for choosing which strategy to use

The purpose of the Strategy design pattern is to separate the implementations of different strategies from each other It also separates the code for picking the strategy from the strategy implementations

The pattern defines a single interface for all strategies The separate strategies are implemented with a method of the same name in each of the classes implementing the interface

Which strategy is used will depend on what kind of object the method implementing the strategy is called on The intent of the pattern is realized through an interface and depends on polymorphism and dynamic binding

Book Definition of Pattern The intent of Strategy is to encapsulate alternative approaches, or strategies, in separate classes that each implement a common operation.

Modeling Strategies Like with the previous chapters, and others, the book illustrates the Strategy design pattern in the following way: It develops an example with multiple strategies that doesn’t use the Strategy design pattern It then refactors the example using the Strategy design pattern

Example Scenario When a potential customer calls in, interested in buying fireworks, there is software which will make a recommendation or suggestion There are several different ways a recommendation can be made

Ways of recommending a purchase: Recommend a particular firework that is being promoted Use a piece of software, Rel8 Use another piece of software, LikeMyStuff Use a default recommendation option

Promoted Firework There is nothing special about this option If the company is promoting a firework, recommend it

Rel8 Rel8 relies on a customer’s already being registered During registration the customer specifies preferences in entertainment and fireworks Rel8 makes a suggestion based on the similarity of the customer to other customers (presumably suggesting something that similar customers have tended to buy) If the customer isn’t registered, Rel8 can’t be used

LikeMyStuff LikeMyStuff doesn’t rely on pre-registration, but it does rely on customer information The idea is that it will make a recommendation based on a profile of recent purchases by the customer If not enough data can be obtained to form the profile, then LikeMyStuff can’t be used

The Default Option This is the default: If none of the previous options applies, then a firework is suggested at random

UML for the Scenario The UML diagram on the following overhead shows the classes involved in the design as described so far Appendix D on UML clarifies the notation: “Use a dashed arrow between classes to show a dependency that does not use an object reference. For example, the Customer class relies on a static method from the LikeMyStuff recommendation engine.”

The getRecommended() Method Viewing the scenario from the top down, what you have is this: The Customer class has a getRecommended() method in it This method consists of if/else code which chooses one of the strategies, whether to do a promotion, or to use Rel8, LikeMyStuff, or the default

Doing a Promotion If there is a promotion underway, the first part of the logic of getRecommended() deals with that case The logic for doing a promotion consists of looking up the contents of a file named strategy.dat in a directory named config If there is such a file, its contents should look something like this: promote=JSquirrel

The basic idea is that if the data file is not empty, the firework it contains is returned If its contents come up null you go on to the next option Also, if the file read doesn’t work, you don’t do anything in the catch block, you just continue on to the other options

Using Rel8 The Rel8 class has a method advise() getRecommended() wraps a call to advise() if the Rel8 strategy is selected The call looks like this: if(isRegistered()) return (Firework) Rel8.advise(this); “this” is the customer, and Rel8 relies entirely on the information contained in the registered customer object

Using LikeMyStuff The LikeMyStuff class has a suggest() method getRecommended() wraps a call to suggest() if the LikeMyStuff strategy is selected The call looks like this: if(spendingSince(cal.getTime()) > 1000) return (Firework) LikeMyStuff.suggest(this);

spendingSince() is called on the implicit parameter, customer cal in the parameter refers to an instance of Calendar getTime() specifies a recent period of time “this” is the customer, which is sent as a parameter to suggest()

suggest() relies on a database of recent purchases by that customer The idea is that if the customer has recently spent $1,000, those purchases provide the basis for a recommendation

Doing the Default The Firework class has a getRandom() method, so if all else fails, getRecommended() wraps a call to that method

The Code for getRecommended() in the Customer Class The code for getRecommended() is shown on the following overheads It is a collection of if statements. It is unfortunate that it is not organized as a sequence of if/else if’s.

The Code for getRecommended() public Firework getRecommended() { // if promoting a particular firework, return it try Properties p = new Properties(); p.load(ClassLoader.getSystemResourceAsStream(“config/strategy.dat”)); String promotedName = p.getProperty(“promote”); if(promotedName != null) Firework f = Firework.lookup(promotedName); if(f != null) return f; }

catch(Exception ignored) { // If resource is missing or it failed to load, // fall through to the next approach. } // if registered, compare to other customers if(isRegistered()) return (Firework) Rel8.advise(this);

// check spending over the last year Calendar cal = Calendar.getInstance(); cal.add(Calendar.YEAR, -1); if(spendingSince(cal.getTime()) > 1000) return (Firework) LikeMyStuff.suggest(this); // oh well! return Firework.getRandom(); }

What’s Wrong with the Initial Design The book identifies two basic problems with the getRecommended() method as given: It’s too long It combines both selecting a strategy and executing it

This is actually one of the high points of the book It explains that you know that the method is too long because you need to put comments in it “Short methods are easy to understand, seldom need explanation…”

Comments Are Bad… Finally, what every student always knew: Comments are bad… More accurately, you might facetiously say that code which requires comments is bad. The book doesn’t say that putting a comment at the beginning for the whole method is bad. A useful observation might be that a method should be short and sweet enough that it doesn’t need internal commenting.

Refactoring to the Strategy Pattern Applying the Strategy design pattern involves three things: 1. Creating an interface that defines the strategic operation 2. Writing classes that implement the interface and embody each of the different strategies 3. Refactoring the code to select and use an instance of the right strategy class

The Interface 1. The interface for this example will be named Advisor The interface requires the implementation of a recommend() method The recommend() method will take a customer as a parameter It will return a firework A UML diagram of the interface is given on the next overhead

The Implementing Classes 2. The next step is to write the classes that implement the interface and embody each of the different strategies These classes will have to implement the recommend() method

The book does the refactoring in part with challenges As usual, it’s easiest to just look at the solutions The UML diagram on the following overhead shows: A new Customer class making use of an Advisor interface 4 classes which implement the interface and embody the 4 strategies

Solution 23.1

The Implementing Classes The PromotionAdvisor and RandomAdvisor class names should be self-explanatory GroupAdvisor refers to the use of Rel8 ItemAdvisor refers to the use of LikeMyStuff The implementations of the recommend() method for these classes will wrap a call to the static methods of Rel8 and LikeMyStuff An expanded UML diagram for these two classes is given on the next overhead

Making Instances of the Implementing Classes An interface can’t define static methods An interface defines what the book calls “object methods”—methods that are called on objects That means that client code will have to make instances of GroupAdvisor and ItemAdvisor The recommend() method will be called on these objects

Only one instance each of GroupAdvisor and ItemAdvisor are needed In the refactored design, these instances will be static objects in the Customer class So the advisor objects will be “singleton like” There won’t be an instance of each kind of advisor for each customer

Even though the recommend() method isn’t a static method, it more or less acts like one If there is only one advisor object, then there is the one recommend() method that can be called on that object The recommend() method does something for customers But it does so by taking the customer as an explicit parameter rather than being called on the customer

Code for the recommend() Method in the GroupAdvisor Class This is the recommend() method in the GroupAdvisor class: public Firework recommend(Customer c) { return (Firework) Rel8.advise(c); } It wraps a call to the advise() method of Rel8 In essence, the call is adapted to the recommend() interface of Advisor

Code for the recommend() Method in the ItemAdvisor Class The code for the recommend() method in the ItemAdvisor class is analogous. The book doesn’t give it and it doesn’t even bother to give it as a challenge. It should be straightforward to write that method.

Challenge 23.2 “In addition to Strategy, what pattern appears in the GroupAdvisor and ItemAdvisor classes?” [The answer to this was given away in the last remark about the recommend() code in GroupAdvisor.]

Solution 23.2 “The GroupAdvisor and ItemAdvisor classes are instances of Adapter, providing the interface a client expects, using the services of a class with a different interface.”

Code for the recommend() Method in the PromotionAdvisor Class A PromotionAdvisor class is also needed, with a recommend() method On the one hand, promotion should be a simple case On the other hand, the book puts a lot of detail into the implementation

Most of the logic of the original code is moved into the constructor for the new class If a promotion is on, then the promoted instance variable of the class is initialized In addition to the recommend() method, there is a hasItem() method which can be called to see whether a promoted item is available

The book’s implementation makes use of class loading logic This requires try/catch blocks The details of this technique will not be covered since they are extraneous to the design pattern The code is shown on the following overheads

public class PromotionAdvisor implements Advisor { private Firework promoted; public PromotionAdvisor() try Properties p = new Properties(); p.load(ClassLoader.getSystemResourceAsStream("config/strategy.dat")); String promotedFireworkName = p.getProperty("promote"); if (promotedFireworkName != null) promoted = Firework.lookup(promotedFireworkName); } catch (Exception ignored) // Resource not found or failed to load promoted = null;

public boolean hasItem() { return promoted != null; } public Firework recommend(Customer c) return promoted;

Code for the recommend() Method in the RandomAdvisor Class The RandomAdvisor class is simple Its code is shown on the following overhead

public class RandomAdvisor implements Advisor { public Firework recommend(Customer c) return Firework.getRandom(); }

Refactoring the Customer Class to Use the Interface

Creating the Advisor Objects A single instance of each kind of advisor is created in the new Customer2 class For any given customer, recommend() may be called on one of these advisor objects The code for the new customer class, Customer2, begins by creating these objects This code is shown on the following overhead

private static PromotionAdvisor promotionAdvisor = new PromotionAdvisor(); private static GroupAdvisor groupAdvisor = new GroupAdvisor(); private static ItemAdvisor itemAdvisor = new ItemAdvisor(); private static RandomAdvisor randomAdvisor = new RandomAdvisor();

Implementing the getAdvisor() Method Customer2 contains a method named getAdvisor() for picking which kind of advisor to use The original design had if/else statements in which different strategies were called In the new design you don’t eliminate if/else statements, but they appear in a different place

In the new design the if/else logic is implemented in the getAdvisor() method getAdvisor() returns an advisor object Polymorphism and dynamic binding determine which strategy is used when recommend() is called on the advisor object

How the getAdvisor() Method Works Recall that getAdvisor() is in the Customer2 class Its implementation includes a call to hasItem(), mentioned earlier to check for a promotion Its implementation also includes calls to isRegistered() and isBigSpender() on the implicit customer parameter to determine if the advisor should be Rel8 or LikeMyStuff

getAdvisor() does lazy initialization of the advisor The value of the advisor instance variable is only set at the time the getAdvisor() method is called It’s not lazy construction because one instance of each kind of advisor has already been created at the top of Customer2 The code is shown on the following overhead

private Advisor getAdvisor() { if (advisor == null) if (promotionAdvisor.hasItem()) advisor = promotionAdvisor; else if (isRegistered()) advisor = groupAdvisor; else if (isBigSpender()) advisor = itemAdvisor; else advisor = randomAdvisor; } return advisor;

The getRecommended() Method in Customer2 Challenge 23.3 “Write the new code for Customer.getRecommended().” Comment mode on: It may be helpful to step back and review where we’re at in the redesign The UML for the redesign is repeated on the following overhead

We’ve got the 4 advisor classes We’ve got the getAdvisor() method The final ingredient of the new design is the getRecommended() method in the Customer2 class This code should rely on the advisor for the customer, and a call to the recommend() method on that advisor The code is shown on the following overhead

Solution 23.3 Your code should look something like: public Firework getRecommended() { return getAdvisor().recommend(this); }

The Benefit of the Design Pattern It is apparent that some of the code in the application is simplified by the application of the design pattern Polymorphism and dynamic binding determines which version of recommend() will be used depending on the type of advisor that was initialized earlier.

Comparing Strategy and State In the book State and Strategy are presented one after the other These patterns have some structural similarities and the book compares them The similarities can be summarized by looking at the UML diagram for state again It is shown on the following overhead

In both patterns there is a client that makes use of an interface/abstract class Each implementing class/subclass contains a method of the same name In the client, the method is called on an object and polymorphism/dynamic binding determines which version of the method is called

Just like the adapter, this structure will appear again in other patterns Using the book’s terminology, the design patterns under consideration which share this structure are different patterns because their intents differ In other words, what you accomplish by using the structure is not the same

It is quite apparent at the detailed level that getting a recommended firework in no way causes a customer to transition into a different state However, touching the control button on the carousel door caused it to enter a different state This important kind of difference isn’t apparent from looking at the overall structure at a macro level in a UML diagram

Comparing Strategy and Template Method (A Sideshow?) This section is not critically important There may be some worthwhile ideas presented, but it doesn’t help too much in focusing on what the strategy design pattern is It’s really more focused on showing that the template design pattern is not the same as the strategy design pattern

The book compares the Strategy and Template methods In this case the similarities that the book suggests might exist are not illustrated by a UML diagram The authors are directly addressing the question of whether Strategy and Template can be viewed as having the same intent

Recall the first presentation of the Template Method design pattern The template was a sort() method in the Arrays or Collections class What was stubbed in was the pairwise comparison operation for elements of the data structure

The authors state that you might argue that changing the pairwise comparison step changes the strategy of sorting They illustrate this idea by suggesting that sorting rockets by price rather than by thrust would support a different marketing strategy Observe that the claim about strategy in this argument is about something in the problem domain, marketing, not in the code writing

Challenge 23.4 “Provide an argument that the Arrays.sort() method provides an example of Template Method or that it is an example of Strategy.” Comment mode on: It’s clear that the book thinks sorting is an example of template, not strategy, since sorting was covered in the template unit. I agree with its reasoning, as presented in the second of the following arguments

Solution 23.4 “Is a reusable sort routine an example of Template Method or of Strategy?” [See the following overheads.]

An Argument that Sorting is an Example of the Strategy Design Pattern “Template Method, according to the original Design Patterns book, lets “subclasses” redefine certain steps of an algorithm. But the Collections.sort() method doesn’t work with subclasses; it uses a Comparator instance. Each instance of Comparator provides a new method and thus a new algorithm and a new strategy. The sort() method is a good example of Strategy.

An Argument that Sorting is an Example of the Template Design Pattern There are many sorting algorithms, but Collections.sort() uses only one (QuickSort). Changing the algorithm would mean changing to, say, a heap sort or a bubble sort. The intent of Strategy is to let you plug in different algorithms. That doesn’t happen here. The intent of Template Method is to let you plug a step into an algorithm. That is precisely how the sort() method works.”

The book’s argument is good—but maybe not entirely clear. The point is that at the macro level, a choice among strategies would mean a choice between different sorting algorithms At the micro level, choosing between pairwise comparisons doesn’t change the overall strategy It is an example of applying the template pattern

Another Example The other example is based on cups and seeds in Togiz Kumalak One side of a Togiz Kumalak board, consisting of 9 cups, is initialized with random seed counts and presented on the screen This half of the board belongs to player whose turn it is The player has to choose which cup to play

The application implements 3 different strategies: Play a cup at random Play the cup with the minimum number of seeds Play the cup with the maximum number of seeds

The player is prompted to pick a strategy The application then tells which cup to play based on that strategy The application is illustrated in the screenshot on the following overhead

The code for this example application is not presented in the overheads It is available on the course Web page in case you’re interested You will find it is an extremely simple implementation of the strategy design pattern as presented in the book

UML for the Pattern The UML diagram given earlier is repeated on the following overhead It is specific to the fireworks problem, but otherwise it shows what the pattern is about It might be possible to make this and other patterns clearer by including sequence diagrams It would show how the getRecommended() method makes a call to the recommend() method

A UML Diagram for the Pattern from Lasater Lasater’s UML diagram is given on the following overhead It is useful, as usual, because it uses generic names for the classes in the pattern This one is also useful because it shows that you could use an abstract class just as easily as you could use an interface when applying the pattern

Summary Multiple strategies may appear in a single method This can be an unwieldy design A solution is to implement each separate strategy as a method of the same name in different classes The individual implementations will be simpler

The original design would have logic to pick one of the strategies The new design would also contain such logic somewhere, but at least it is separated from the implementations of the strategies The new logic will be based on which of the strategy classes is constructed It would be reasonable to hold the strategy object as a reference in a client class

Then at the appropriate time, this object has the strategy method called on it In other words, once the object choice is made, the method choice is accomplished by polymorphism and dynamic binding The result of this kind of design is individual pieces of code that are smaller, simpler, and unified in purpose

Overall, the design should be clearer, with different strategies implemented in different classes The cleaner the design, the less commenting may be needed…

The End