Note: This document requires the installation of the fonts Georgia, Verdana and Andale Mono (code font) for proper viewing. These can be found at: http://sourceforge.net/project/showfiles.php?group_id=34153&release_id=105355
Modifications in Revision 0.9:
∑ Prose has still had little/no work. My current goal is to get the structure and examples worked out so that the seminar works well. Once it has been proven in the seminars, then I will spend time on the prose.
∑ Added proxy:PoolManager.java to make a more generic/customizeable Pool Manager, and modified proxy:ConnectionPoolProxyDemo.java accordingly [[ Still need to decide what to return when you run out of objects in the pool ]]
∑ Changed PoolManager.java to use an ArrayList (and thus does not require a fixed size at initialization)
∑ Added KissingPrincess.java to State description, as a motivational example for the pattern
∑ Added a simple Flyweight example
∑ Simplified the enumeration in PaperScissorsRock.java
Modifications in Revision 0.8:
∑ Changed Bridge example to improve clarity.
∑ Removed superscripts for better viewing with IE (see note above)
Modifications in Revision 0.7:
∑ NOTE primary changes have been made to structure of book and code examples, but not to prose. Prose can be considered to be mostly a mess in this revision.
∑ Complete reorganization under headings that attempt to describe the problems you are trying to solve with a pattern.
∑ Addition of placeholders for the remainder of the GoF patterns
∑ Addition of ďSimplifying IdiomsĒ section and examples
∑ Addition of Builder section and examples
∑ Removed unit-testing chapter; replaced with reference to ďnewĒ JUnit (which uses reflection)
∑ (4-30-2003) Added Ant build.xml files, and support files from TIJ necessary to do a full standalone build. You should be able to type ďantĒ from the code root directory and get a successful build.
∑ Dramatically simplified chainofresponsibility:FindMinima.java
∑ Added object pool/connection pool examples
∑ Refactored small things in many examples
∑ Some exercises may have been left behind when patterns were moved.
∑ For simplicity, saved from Word into a single HTML document, using ďfilteredĒ version to remove Office stuff. Seems to work pretty well; checked it with both IE and Mozilla (actually seems to work better on Mozilla than on IE!).
∑ Reconfigure for new backtalk system
∑ Replace references to TIJ2 with TIJ3
Problem-Solving† Techniques using Java
President, MindView, Inc.
The material in this book has been developed in conjunction with a seminar that I have given for several years, mostly with Bill Venners. Bill and I have given many iterations of this seminar and weíve changed it many times over the years as we both have learned more about patterns and about giving the seminar.
In the process weíve both produced more than enough information for us each to have our own seminars, an urge that weíve both strongly resisted because we have so much fun giving the seminar together. Weíve given the seminar in numerous places in the US, as well as in Prague (where we try to have a mini-conference every Spring together with a number of other seminars). Weíve also given it as an on-site seminar.
A great deal of appreciation goes to the people who have participated in these seminars over the years, as they have helped me work through these ideas and to refine them. I hope to be able to continue to form and develop these kinds of ideas through this book and seminar for many years to come.
This book will not stop here, either. After much pondering, Iíve realized that I want Thinking in Python to be, initially, a translation of this book rather than an introduction to Python (there are already plenty of fine introductions to that superb language). I find this prospect to be much more exciting than the idea of struggling through another language tutorial (my apologies to those who were hoping for that).
This is a book about design that I have been working on for years, basically ever since I first started trying to read Design Patterns (Gamma, Helm, Johnson & Vlissides, Addison-Wesley, 1995), commonly referred to as the Gang of Four or just GoF).
There is a chapter on design patterns in the first edition of Thinking in C++, which has evolved in Volume 2 of the second edition of Thinking in C++, and youíll also find a chapter on patterns in the first edition of Thinking in Java (I took it out of the second edition because that book was getting too big, and also because I had decided to write this book).
This is not an introductory book. I am assuming that you have worked your way through Thinking in Java or an equivalent text before coming to this book.
In addition, I assume you have more than just a grasp of the syntax of Java. You should have a good understanding of objects and what theyíre about, including polymorphism. Again, these are topics covered in Thinking in Java.
On the other hand, by going through this book youíre going to learn a lot about object-oriented programming by seeing objects used in many different situations. If your knowledge of objects is rudimentary, it will get much stronger in the process of understanding the designs in this book.
In a book that has ďproblem-solving techniquesĒ in its subtitle, itís worth mentioning one of the biggest pitfalls in programming: premature optimization. Every time I bring this concept forward, virtually everyone agrees to it. Also, everyone seems to reserve in their own mind a special case ďexcept for this thing that I happen to know is a particular problem.Ē
The reason I call this the Y2K syndrome has to do with that special knowledge. Computers are a mystery to most people, so when someone announced that those silly computer programmers had forgotten to put in enough digits to hold dates past the year 1999, then suddenly everyone became a computer expert Ė ďthese things arenít so difficult after all, if I can see such an obvious problem.Ē For example, my background was originally in computer engineering, and I started out by programming embedded systems. As a result, I know that many embedded systems have no idea what the date or time is, and even if they do that data often isnít used in any important calculations. And yet I was told in no uncertain terms that all the embedded systems were going to crash on January 1, 2000. As far as I can tell the only memory that was lost on that particular date was that of the people who were predicting doom Ė itís as if they had never said any of that stuff.
The point is that itís very easy to fall into a habit of thinking that the particular algorithm or piece of code that you happen to partly or thoroughly understand is naturally going to be the bottleneck in your system, simply because you can imagine whatís going on in that piece of code and so you think that it must somehow be much less efficient than all the other pieces of code that you donít know about. But unless youíve run actual tests, typically with a profiler, you canít really know whatís going on. And even if you are right, that a piece of code is very inefficient, remember that most programs spend something like 90% of their time in less than 10% of the code in the program, so unless the piece of code youíre thinking about happens to fall into that 10% it isnít going to be important.
ďWe should forget about small efficiencies, say about 97% of the time: Premature optimization is the root of all evil.ĒóDonald Knuth
One of the terms you will see used over and over in design patterns literature is context. In fact, one common definition of a design pattern is ďa solution to a problem in a context.Ē The GoF patterns often have a ďcontext objectĒ that the client programmer interacts with. At one point it occurred to me that such objects seemed to dominate the landscape of many patterns, and so I began asking what they were about.
The context object often acts as a little faÁade to hide the complexity of the rest of the pattern, and in addition it will often be the controller that manages the operation of the pattern. Initially, it seemed to me that these were not really essential to the implementation, use and understanding of the pattern. However, I remembered one of the more dramatic statements made in the GoF: ďprefer composition to inheritance.Ē The context object allows you to use the pattern in a composition, and that may be itís primary value.
1) The great value of exceptions is the unification of error reporting: a standard mechanism by which to report errors, rather than the popourri of ignorable approaches that we had in C (and thus, C++, which only adds exceptions to the mix, and doesn't make it the exclusive approach). The big advantage Java has over C++ is that exceptions are the only way to report errors.
2) "Ignorable" in the previous paragraph is the other issue. The theory is that if the compiler forces the programmer to either handle the exception or pass it on in an exception specification, then the programmer's attention will always be brought back to the possibility of errors and they will thus properly take care of them. I think the problem is that this is an untested assumption we're making as language designers that falls into the field of psychology. My theory is that when someone is trying to do something and you are constantly prodding them with annoyances, they will use the quickest device available to make those annoyances go away so they can get their thing done, perhaps assuming they'll go back and take out the device later. I discovered I had done this in the first edition of Thinking in Java:
And then more or less forgot it until the rewrite. How many people thought this was a good example and followed it? Martin Fowler began seeing the same kind of code, and realized people were stubbing out exceptions and then they were disappearing. The overhead of checked exceptions was having the opposite effect of what was intended, something that can happen when you experiment (and I now believe that checked exceptions were an experiment based on what someone thought was a good idea, and which I believed was a good idea until recently).
When I started using Python, all the exceptions appeared, none were accidentally "disappeared." If you *want* to catch an exception, you can, but you aren't forced to write reams of code all the time just to be passing the exceptions around. They go up to where you want to catch them, or they go all the way out if you forget (and thus they remind you) but they don't vanish, which is the worst of all possible cases. I now believe that checked exceptions encourage people to make them vanish. Plus they make much less readable code.
In the end, I think we must realize the experimental nature of exceptions and look at them carefully before assuming that everything about exceptions in Java are good. I believe that having a single mechanism for handling errors is excellent, and I believe that using a separate channel (the exception handling mechanism) for moving the exceptions around is good. But I do remember one of the early arguments for exception handling in C++ was that it would allow the programmer to separate the sections of code where you just wanted to get work done from the sections where you handled errors, and it seems to me that checked exceptions do not do this; instead, they tend to intrude (a lot) into your "normal working code" and thus are a step backwards. My experience with Python exceptions supports this, and unless I get turned around on this issue I intend to put a lot more RuntimeExceptions into my Java code.
ďDesign patterns help you learn from others' successes instead of your own failures.Ē
. That book shows 23 different solutions to particular classes of problems. In this book, the basic concepts of design patterns will be introduced along with examples. This should whet your appetite to read Design Patterns by Gamma, et. al., a source of what has now become an essential, almost mandatory, vocabulary for OOP programmers.Probably the most important step forward in object-oriented design is the ďdesign patternsĒ movement, chronicled in Design Patterns (ibid)
The latter part of this book contains an example of the design evolution process, starting with an initial solution and moving through the logic and process of evolving the solution to more appropriate designs. The program shown (a trash sorting simulation) has evolved over time, and you can look at that evolution as a prototype for the way your own design can start as an adequate solution to a particular problem and evolve into a flexible approach to a class of problems.
Initially, you can think of a pattern as an especially clever and insightful way of solving a particular class of problems. That is, it looks like a lot of people have worked out all the angles of a problem and have come up with the most general, flexible solution for it. The problem could be one you have seen and solved before, but your solution probably didnít have the kind of completeness youíll see embodied in a pattern.
Although theyíre called ďdesign patterns,Ē they really arenít tied to the realm of design. A pattern seems to stand apart from the traditional way of thinking about analysis, design, and implementation. Instead, a pattern embodies a complete idea within a program, and thus it can sometimes appear at the analysis phase or high-level design phase. This is interesting because a pattern has a direct implementation in code and so you might not expect it to show up before low-level design or implementation (and in fact you might not realize that you need a particular pattern until you get to those phases).
The basic concept of a pattern can also be seen as the basic concept of program design: adding a layer of abstraction. Whenever you abstract something youíre isolating particular details, and one of the most compelling motivations behind this is to separate things that change from things that stay the same. Another way to put this is that once you find some part of your program thatís likely to change for one reason or another, youíll want to keep those changes from propagating other changes throughout your code. Not only does this make the code much cheaper to maintain, but it also turns out that it is usually simpler to understand (which results in lowered costs).
Often, the most difficult part of developing an elegant and cheap-to-maintain design is in discovering what I call ďthe vector of change.Ē (Here, ďvectorĒ refers to the maximum gradient and not a container class.) This means finding the most important thing that changes in your system, or put another way, discovering where your greatest cost is. Once you discover the vector of change, you have the focal point around which to structure your design.
So the goal of design patterns is to isolate changes in your code. If you look at it this way, youíve been seeing some design patterns already in this book. For example, inheritance can be thought of as a design pattern (albeit one implemented by the compiler). It allows you to express differences in behavior (thatís the thing that changes) in objects that all have the same interface (thatís what stays the same). Composition can also be considered a pattern, since it allows you to changeódynamically or staticallyóthe objects that implement your class, and thus the way that class works.
Youíve also already seen another pattern that appears in Design Patterns: the iterator (Java 1.0†and 1.1 capriciously calls it the Enumeration; Java 2†containers use ďiteratorĒ). This hides the particular implementation of the container as youíre stepping through and selecting the elements one by one. The iterator allows you to write generic code that performs an operation on all of the elements in a sequence without regard to the way that sequence is built. Thus your generic code can be used with any container that can produce an iterator.
One of the events thatís occurred with the rise of design patterns is what could be thought of as the ďpollutionĒ of the term Ė people have begun to use the term to mean just about anything synonymous with ďgood.Ē After some pondering, Iíve come up with a sort of hierarchy describing a succession of different types of categories:
1. Idiom: how we write code in a particular language to do this particular type of thing. This could be something as common as the way that you code the process of stepping through an array in C (and not running off the end).
2. Specific Design: the solution that we came up with to solve this particular problem. This might be a clever design, but it makes no attempt to be general.
3. Standard Design: a way to solve this kind of problem. A design that has become more general, typically through reuse.
4. Design Pattern: how to solve an entire class of similar problem. This usually only appears after applying a standard design a number of times, and then seeing a common pattern throughout these applications.
I feel this helps put things in perspective, and to show where something might fit. However, it doesnít say that one is better than another. It doesnít make sense to try to take every problem solution and generalize it to a design pattern Ė itís not a good use of your time, and you canít force the discovery of patterns that way; they tend to be subtle and appear over time.
One could also argue for the inclusion of Analysis Pattern and Architectural Pattern in this taxonomy.
(Update from slides to here)
When I put out a call for ideas in my newsletter, a number of suggestions came back which turned out to be very useful, but different than the above classification, and I realized that a list of design principles is at least as important as design structures, but for a different reason: these allow you to ask questions about your proposed design, to apply tests for quality.
∑ Principle of least astonishment (donít be astonishing).
∑ Make common things easy, and rare things possible
∑ Consistency. One thing has become very clear to me, especially because of Python: the more random rules you pile onto the programmer, rules that have nothing to do with solving the problem at hand, the slower the programmer can produce. And this does not appear to be a linear factor, but an exponential one.
∑ Law of Demeter: a.k.a. ďDonít talk to strangers.Ē An object should only reference itself, its attributes, and the arguments of its methods.
∑ Subtraction: a design is finished when you cannot take anything else away.
∑ Simplicity before generality. (A variation of Occamís Razor, which says ďthe simplest solution is the bestĒ). A common problem we find in frameworks is that they are designed to be general purpose without reference to actual systems. This leads to a dizzying array of options that are often unused, misused or just not useful. However, most developers work on specific systems, and the quest for generality does not always serve them well. The best route to generality is through understanding well-defined specific examples. So, this principle acts as the tie breaker between otherwise equally viable design alternatives. Of course, it is entirely possible that the simpler solution is the more general one.
∑ Reflexivity (my suggested term). One abstraction per class, one class per abstraction. Might also be called Isomorphism.
∑ Independence or Orthogonality. Express independent ideas independently. This complements Separation, Encapsulation and Variation, and is part of the Low-Coupling-High-Cohesion message.
∑ Once and once only: Avoid duplication of logic and structure where the duplication is not accidental, ie where both pieces of code express the same intent for the same reason.
In the process of brainstorming this idea, I hope to come up with a small handful of fundamental ideas that can be held in your head while you analyze a problem. However, other ideas that come from this list may end up being useful as a checklist while walking through and analyzing your design.
The Design Patterns book discusses 23 different patterns, classified under three purposes (all of which revolve around the particular aspect that can vary). The three purposes are:
1. Creational: how an object can be created. This often involves isolating the details of object creation so your code isnít dependent on what types of objects there are and thus doesnít have to be changed when you add a new type of object. The aforementioned Singleton is classified as a creational pattern, and later in this book youíll see examples of Factory Method and Prototype.
2. Structural: designing objects to satisfy particular project constraints. These work with the way objects are connected with other objects to ensure that changes in the system donít require changes to those connections.
3. Behavioral: objects that handle particular types of actions within a program. These encapsulate processes that you want to perform, such as interpreting a language, fulfilling a request, moving through a sequence (as in an iterator), or implementing an algorithm. This book contains examples of the Observer and the Visitor patterns.
The Design Patterns book has a section on each of its 23 patterns along with one or more examples for each, typically in C++ (rather restricted C++, at that) but sometimes in Smalltalk. (Youíll find that this doesnít matter too much since you can easily translate the concepts from either language into Java.) This book will revisit many of the patterns shown in Design Patterns but with a Java orientation, since the language changes the expression and understanding of the patterns. However, the GoF examples will not be repeated here, since I believe that itís possible to produce more illuminating examples given some effort. The goal is to provide you with a decent feel for what patterns are about and why they are so important.
After years of looking at these things, it began to occur to me that the patterns themselves use basic principles of organization, other than (and more fundamental than) those described in Design Patterns. These principles are based on the structure of the implementations, which is where I have seen great similarities between patterns (more than those expressed in Design Patterns). Although we generally try to avoid implementation in favor of interface, for awhile I thought that it was easier to understand the patterns in terms of these structural principles, and tried reorganizing the book around the patterns based on their structure instead of the categories presented in Design Patterns.
However, a later insight made me realize that itís more useful to organize the patterns in terms of the problems they solve. I believe this is a subtle but important distinction from the way Metsker organizes the patterns by intent in Design Patterns Java Workshop (Addison-Wesley 2002), because I hope that you will then be able to recognize your problem and search for a solution, if the patterns are organized this way.
In the process of doing all this ďbook refactoringĒ I realized that if I changed it once, I would probably change it again (thereís definitely a design maxim in there), so I removed all references to chapter numbers in order to facilitate this change (the little-known ďnumberless chapterĒ pattern J).HowHH
Issues of development, the UML process, Extreme Programming.
Is evaluation valuable? The Capability Immaturity Model:
Pair programming research:
In an earlier version of this book I decided that unit testing was essential (for all of my books) and that JUnit was too verbose and clunky to consider. At that time I wrote my own unit testing framework using Java reflection to simplify the syntax necessary to achieve unit testing. For the third edition of Thinking in Java, we developed another unit testing framework for that book which would test the output of examples.
In the meantime, JUnit has changed to add a syntax remarkably similar to the one that I used in an earlier version of this book. I donít know how much influence I may have had on that change, but Iím simply happy that it has happened, because I no longer feel the need to support my own system (which you can still find <some URL here>) and can simply recommend the defacto standard.
I have introduced and described the style of JUnit coding that I consider a ďbest practiceĒ (primarily because of simplicity), in Thinking in Java, 3rd edition, chapter 15. That section provides an adequate introduction to any of the unit testing you will see associated with this book (however, the unit testing code will not normally be included in the text of this book). When you download the code for this book, you will find (4/9/2003: Eventually, not yet) unit tests along with the code examples whenever possible.
Public: in test subdirectory; different package (donít include in jar).
Package access: same package, subdirectory path underneath library code (donít include in jar)
Private access: (white box testing). Nested class, strip out, or Junit addons.
Before getting into more complex techniques, itís helpful to look at some basic ways to keep code simple and straightforward.
The most trivial of these is the messenger, which simply packages information into an object to be passed around, instead of passing all the pieces around separately. Note that without the messenger, the code for translate() would be much more confusing to read:
Since the goal of a messenger is only to carry data, that data is made public for easy access. However, you may also have reasons to make the fields private.
Messengerís big brother is the collecting parameter, whose job is to capture information from the method to which it is passed. Generally, this is used when the collecting parameter is passed to multiple methods, so itís like a bee collecting pollen.
A container makes an especially useful collecting parameter, since it is already set up to dynamically add objects:
The collecting parameter must have some way to set or insert values. Note that by this definition, a messenger could be used as a collecting parameter. The key is that a collecting parameter is passed about and modified by the methods it is passed to.
The two patterns described here are solely used to control the quantity of objects.
Singleton could actually be thought of as a special case of Object Pool, but the applications of the Object Pool tend to be uniqe enough from Singleton that itís worth treating the two separately.
Possibly the simplest design pattern is the singleton, which is a way to provide one and only one object of a particular type.† An important aspect of Singleton is that you provide a global access point, so singletons are often a solution for what you would have used a global variable for in C. In addition, a singleton often has the characteristics of a registry or lookup service Ė itís a place you go to find references to other objects.
Singletons can be found in the Java libraries, but hereís a more direct example:
The key to creating a singleton is to prevent the client programmer from having any way to create an object except the ways you provide. You must make all constructors private, and you must create at least one constructor to prevent the compiler from synthesizing a default constructor for you (which it will create using package access).
At this point, you decide how youíre going to create your object. Here, itís created statically, but you can also wait until the client programmer asks for one and create it on demand. In any case, the object should be stored privately. You provide access through public methods. Here, getReference( ) produces the reference to the Singleton object. The rest of the interface (getValue( ) and setValue( )) is the regular class interface.
Java also allows the creation of objects through cloning. In this example, making the class final prevents cloning. Since Singleton is inherited directly from Object, the clone( ) method remains protected so it cannot be used (doing so produces a compile-time error). However, if youíre inheriting from a class hierarchy that has already overridden clone( ) as public and implemented Cloneable, the way to prevent cloning is to override clone( ) and throw a CloneNotSupportedException as described in Appendix A of Thinking in Java, 2nd edition. (You could also override clone( ) and simply return this, but that would be deceiving since the client programmer would think they were cloning the object, but would instead still be dealing with the original.) Actually, this isnít precisely true, because even in the above situation someone could still use reflection to invoke clone( ) [[is this true? clone( ) is still protected so Iím not so sure. If it is true, youíd have to throw CloneNotSupportedException as the only way to guarantee un-cloneability ]]
1. SingletonPattern.java always creates an object, even if itís never used. Modify this program to use lazy initialization, so the singleton object is only created the first time that it is needed.
2. Create a registry/lookup service that accepts a Java interface and produces a reference to an object that implements that interface.
Note that you arenít restricted to creating only one object. This is also a technique to create a limited pool of objects. In that situation, however, you can be confronted with the problem of sharing objects in the pool. If this is an issue, you can create a solution involving a check-out and check-in of the shared objects.
As an example, consider a database. Commercial databases often restrict the number of connections that you can use at any one time. Here is an implementation that uses an object pool to manage the connections. First, the basic concept of managing a pool of objects is implemented as a separate class:
1. Add unit tests to ConnectionPoolDemo.java to demonstrate the problem that the client may release the connection but still continue to use it.
Both Proxy and State provide a surrogate class that you use in your code; the real class that does the work is hidden behind this surrogate class. When you call a method in the surrogate, it simply turns around and calls the method in the implementing class. These two patterns are so similar that the Proxy is simply a special case of State. One is tempted to just lump the two together into a pattern called Surrogate, but the term ďproxyĒ has a long-standing and specialized meaning, which probably explains the reason for the two different patterns.
The basic idea is simple: from a base class, the surrogate is derived along with the class or classes that provide the actual implementation:
When a surrogate object is created, it is given an implementation to which to send all of the method calls.
Structurally, the difference between Proxy and State is simple: a Proxy has only one implementation, while State has more than one. The application of the patterns is considered (in Design Patterns) to be distinct: Proxy is used to control access to its implementation, while State allows you to change the implementation dynamically. However, if you expand your notion of ďcontrolling access to implementationĒ then the two fit neatly together.
If we implement Proxy by following the above diagram, it looks like this:
Of course, it isnít necessary that Implementation have the same interface as Proxy; as long as Proxy is somehow ďspeaking forĒ the class that it is referring method calls to then the basic idea is satisfied (note that this statement is at odds with the definition for Proxy in GoF). However, it is convenient to have a common interface so that Implementation is forced to fulfill all the methods that Proxy needs to call.
In JDK 1.3, the Dynamic Proxy was introduced. Although a little complex at first, this is an intruiging tool.
Here's an interesting little starting example, which works and proves that yes, indeed, the invocation handler is being called so the proxying etc. is actually working. So it's pretty cool, and it's in my head now, but I still have to figure out something reasonable to do with the invocation handler to come up with a useful example...
Exercise: Use the Java dynamic proxy to create an object that acts as a front end for a simple configuration file. For example, in good_stuff.txt you can have entries like this:
A client programmer of this NeatPropertyBundle could then write:
The contents of the configuration file can contain anything, with any variable names. The dynamic proxy will either map to the name or tell you it doesnít exist somehow (probably by returning null). If you set a property and it doesnít already exist, the dynamic proxy will create the new entry. The toString( ) method should display all the current entries.
Exercise: similar to the previous exercise, use the Java dynamic proxy to make a connection to the DOS Autoexec.bat file.
Exercise: Accept an SQL query which returns data, then read the DB metadata.† Now, for each record, provide an object which has attributes corresponding to the column names and of appropriate data types.†
Exercise: Create a simple server and client that uses XML-RPC. Each object the client returns should use the dynamic proxy concept to exercise the remote methods.
I'm not sure about the exercises you suggest, except for the last one. The thing is that I like to think at invocation handler as somthing providing features that are orthogonal to the ones provided by the object being "proxied".
In other words: the implementation of the invocation handler is completely independent from the interface(s) of the object that the dynamically-generated proxy represent. Which means that once you have implemented an invocation handler, you can use for any class that exposes interfaces, even for classes and interfaces that were not present when the handler was implemented.
That's why I say the the handler provides services that are orthogonal to the ones provided by the proxied object. Rickard has a few handlers in his SmartWorld example, and they one I like the best is a call-retry handler. It basically makes a call into the actual object, and if the call generates an exception if waits for a while, then makes the same call again for a total of three times. If all three calls fails, it returns an exception. And you can use such a handler on _any_ class.
The implementation is way too complex for what you are trying to demonstrate. I'm using this example just to explain what I mean by orthogonal services.
In your list of exercises, the only one that, in my opinion, makes sense to implement using dynamic proxies is the last one, the one using XML-RPC to communicate with an object. And that's because the mechanism you use to dispatch the message (XML-RPC) is orthogonal to the services provided by the object you want to reach.
An object that appears to change its class.
Indications: conditional code in most or all methods.
The State pattern switches from one implementation to another during the lifetime of the surrogate, in order to produce different behavior from the same method call(s). Itís a way to improve the implementation of your code when you seem to be doing a lot of testing inside each of your methods before deciding what to do for that method. For example, the fairy tale of the frog-prince contains an object (the creature) that behaves differently depending on what state itís in. You could implement this using a boolean that you test:
However, the greet() method, and any other methods that must test isFrog before they perform their operations, ends up with awkward code. By delegating the operations to a State object that can be changed, this code is simplified.
In addition, changes to the State are automatically propagated throughout, rather than requiring an edit across the class methods in order to effect changes.
Hereís the basic structure of State:[[ [[
In main( ), you can see that the first implementation is used for a bit, then the second implementation is swapped in and that is used.
There are a number of details that are choices that you must make according to the needs of your own implementation, such as whether the fact that you are using State is exposed to the client, and how the changes to State are made. Sometimes (as in the Swing LayoutManager) the client may pass in the object directly, but in KissingPrincess2.java the fact that State is used is invisible to the client. In addition, the mechanism for changing state may be simple or complex Ė in State Machine, described later in this book, larger sets of states and different mechanisms for changing are explored.
The Swing LayoutManager example mentioned above is an interesting example because it show behavior of both Strategy and State.
The difference between Proxy and State is in the problems that are solved. The common uses for Proxy as described in Design Patterns are:
1. Remote proxy. This proxies for an object in a different address space. A remote proxy is created for you automatically by the RMI compiler rmic as it creates stubs and skeletons.
2. Virtual proxy. This provides ďlazy initializationĒ to create expensive objects on demand.
3. Protection proxy. Used when you donít want the client programmer to have full access to the proxied object.
4. Smart reference. To add additional actions when the proxied object is accessed. For example, or to keep track of the number of references that are held for a particular object, in order to implement the copy-on-write idiom and prevent object aliasing. A simpler example is keeping track of the number of calls to a particular method.
You could look at a Java reference as a kind of protection proxy, since it controls access to the actual object on the heap (and ensures, for example, that you donít use a null reference).
[[ Rewrite this: In Design Patterns, Proxy and State are not seen as related to each other because the two are given (what I consider arbitrarily) different structures. State, in particular, uses a separate implementation hierarchy but this seems to me to be unnecessary unless you have decided that the implementation is not under your control (certainly a possibility, but if you own all the code there seems to be no reason not to benefit from the elegance and helpfulness of the single base class). In addition, Proxy need not use the same base class for its implementation, as long as the proxy object is controlling access to the object it ďfrontingĒ for. Regardless of the specifics, in both Proxy and State a surrogate is passing method calls through to an implementation object.]]]
State can be found everywhere because itís such a fundamental idea. For example, in Builder, the ďDirectorĒ uses a backend Builder object to produce different behaviors.
Alexander Stepanov thought for years about the problem of generic programming techniques before creating the STL (along with Dave Musser). He came to the conclusion that all algorithms are defined on algebraic structures Ė what we would call containers.
In the process, he realized that iterators are central to the use of algorithms, because they decouple the algorithms from the specific type of container that the algorithm might currently be working with. This means that you can describe the algorithm without worrying about the particular sequence it is operating on. More generally, any code that you write using iterators is decoupled from the data structure that the code is manipulating, and thus your code is more general and reusable.
The use of iterators also extends your code into the realm of functional programming, whose objective is to describe what a program is doing at every step rather than how it is doing it. That is, you say ďsortĒ rather than describing the sort. The objective of the C++ STL was to provide this generic programming approach for C++ (how successful this approach will actually be remains to be seen).
If youíve used containers in Java (and itís hard to write code without using them), youíve used iterators Ė in the form of the Enumeration in Java 1.0/1.1 and the Iterator in Java 2. So you should already be familiar with their general use. If not, see Chapter 9, Holding Your Objects, under Iterators in Thinking in Java, 2nd edition (freely downloadable from www.BruceEckel.com).
Because the Java 2 containers rely heavily on iterators they become excellent candidates for generic/functional programming techniques. This chapter will explore these techniques by converting the STL algorithms to Java, for use with the Java 2 container library.
In Thinking in Java, 2nd edition, I show the creation of a type-safe container that will only accept a particular type of object. A reader, Linda Pazzaglia, asked for the other obvious type-safe component, an iterator that would work with the basic java.util containers, but impose the constraint that the type of objects that it iterates over be of a particular type.
If Java ever includes a template mechanism, this kind of iterator will have the added advantage of being able to return a specific type of object, but without templates you are forced to return generic Objects, or to require a bit of hand-coding for every type that you want to iterate through. I will take the former approach.
A second design decision involves the time that the type of object is determined. One approach is to take the type of the first object that the iterator encounters, but this is problematic because the containers may rearrange the objects according to an internal ordering mechanism (such as a hash table) and thus you may get different results from one iteration to the next. The safe approach is to require the user to establish the type during construction of the iterator.
Lastly, how do we build the iterator? We cannot rewrite the existing Java library classes that already produce Enumerations and Iterators. However, we can use the Decorator design pattern, and create a class that simply wraps the Enumeration or Iterator that is produced, generating a new object that has the iteration behavior that we want (which is, in this case, to throw a RuntimeException if an incorrect type is encountered) but with the same interface as the original Enumeration or Iterator, so that it can be used in the same places (you may argue that this is actually a Proxy pattern, but itís more likely Decorator because of its intent). Here is the code:
1. Create an example of the ďvirtual proxy.Ē
2. Create an example of the ďSmart referenceĒ proxy where you keep count of the number of method calls to a particular object.
3. Create a program similar to certain DBMS systems that only allow a certain number of connections at any time. To implement this, use a singleton-like system that controls the number of ďconnectionĒ objects that it creates. When a user is finished with a connection, the system must be informed so that it can check that connection back in to be reused. To guarantee this, provide a proxy object instead of a reference to the actual connection, and design the proxy so that it will cause the connection to be released back to the system.
4. Using the State pattern, make a class called UnpredictablePerson which changes the kind of response to its hello( ) method depending on what kind of Mood itís in. Add an additional kind of Mood called Prozac.
5. Create a simple copy-on write implementation.
6. The java.util.Map has no way to automatically load a two-dimensional array of objects into a Map as key-value pairs. Create an adapter class that does this.
7. Create an Adapter Factory that dynamically finds and produces the adapter that you need to connect a given object to a desired interface.
8. Solve the above exercise using the dynamic proxy thatís part of the Java standard library.
9. Modify the Object Pool solution so that the objects are returned to the pool automatically after a certain amount of time.
10. Modify the above solution to use ďleasingĒ so that the client can renew the lease on the object to prevent it from being automatically released by the timer.
11. Modify the Object Pool system to take threading issues into account.
Applying the ďonce and only onceĒ principle produces the most basic pattern of putting code that changes into a method.
This can be expressed two ways:
Strategy also adds a ďContextĒ which can be a surrogate class that controls the selection and use of the particular strategy objectójust like State! Hereís what it looks like:
Note similarity with template method Ė TM claims distinction that it has more than one method to call, does things piecewise. However, itís not unlikely that strategy object would have more than one method call; consider Shallowayís order fulfullment system with country information in each strategy.
Strategy example from JDK: comparator objects.
Although GoF says that Policy is just another name for strategy, their use of Strategy implicitly assumes a single method in the strategy object Ė that youíve broken out your changing algorithm as a single piece of code.
Others use Policy to mean an object that has multiple methods that may vary independently from class to class. This gives more flexibility than being restricted to a single method.
For example, a shipping policy for a product can be used to describe shipping issues for sending a package to various different countries. This may include the available methods of shipping, how to calculate postage or shipping cost, customs requirements and fees, and special handling costs. All these things may vary independently of each other, and more importantly you may need the information from each at different points in the shipping process.
It also seems generally useful to distinguish Strategies with single methods from Policies with multiple methods.
An application framework allows you to inherit from a class or set of classes and create a new application, reusing most of the code in the existing classes and overriding one or more methods in order to customize the application to your needs. A fundamental concept in the application framework is the Template Method which is typically hidden beneath the covers and drives the application by calling the various methods in the base class (some of which you have overridden in order to create the application).
For example, whenever you create an applet youíre using an application framework: you inherit from JApplet and then override init( ). The applet mechanism (which is a Template Method) does the rest by drawing the screen, handling the event loop, resizing, etc.
An important characteristic of the Template Method is that it is defined in the base class and cannot be changed. Itís sometimes a private method but itís virtually always final. It calls other base-class methods (the ones you override) in order to do its job, but it is usually called only as part of an initialization process (and thus the client programmer isnít necessarily able to call it directly).
The base-class constructor is responsible for performing the necessary initialization and then starting the ďengineĒ (the template method) that runs the application (in a GUI application, this ďengineĒ would be the main event loop). The client programmer simply provides definitions for customize1( ) and customize2( ) and the ďapplicationĒ is ready to run.
Create a framework that takes a list of file names on the command
line. It opens each file except the last for reading, and the last for writing.
The framework will process each input file using an undetermined policy and
write the output to the last file. Inherit to customize this framework to
create two separate applications:
1) Converts all the letters in each file to uppercase.
2) Searches the files for words given in the first file.
When you discover that you need to add new types to a system, the most sensible first step is to use polymorphism to create a common interface to those new types. This separates the rest of the code in your system from the knowledge of the specific types that you are adding. New types may be added without disturbing existing code Ö or so it seems. At first it would appear that the only place you need to change the code in such a design is the place where you inherit a new type, but this is not quite true. You must still create an object of your new type, and at the point of creation you must specify the exact constructor to use. Thus, if the code that creates objects is distributed throughout your application, you have the same problem when adding new typesóyou must still chase down all the points of your code where type matters. It happens to be the creation of the type that matters in this case rather than the use of the type (which is taken care of by polymorphism), but the effect is the same: adding a new type can cause problems.
The solution is to force the creation of objects to occur through a common factory rather than to allow the creational code to be spread throughout your system. If all the code in your program must go through this factory whenever it needs to create one of your objects, then all you must do when you add a new object is to modify the factory.
Since every object-oriented program creates objects, and since itís very likely you will extend your program by adding new types, I suspect that factories may be the most universally useful kinds of design patterns.
Although only the Simple Factory Method is a true singleton, youíll find that each specify factory class in the more general types of factories will only have a single instance.
As an example, letís revisit the Shape system.†
One approach is to make the factory a static method of the base class:
The factory( ) takes an argument that allows it to determine what type of† Shape to create; it happens to be a String in this case but it could be any set of data. The factory( ) is now the only other code in the system that needs to be changed when a new type of Shape is added (the initialization data for the objects will presumably come from somewhere outside the system, and not be a hard-coded array as in the above example).
To encourage creation to only happen in the factory( ), the constructors for the specific types of Shape are give package access, so factory( ) has access to the constructors but they are not available outside the package.
The static factory( ) method in the previous example forces all the creation operations to be focused in one spot, so thatís the only place you need to change the code. This is certainly a reasonable solution, as it throws a box around the process of creating objects. However, the Design Patterns book emphasizes that the reason for the Factory Method pattern is so that different types of factories can be subclassed from the basic factory (the above design is mentioned as a special case). However, the book does not provide an example, but instead just repeats the example used for the Abstract Factory (youíll see an example of this in the next section). Here is ShapeFactory1.java modified so the factory methods are in a separate class as virtual functions. Notice also that the specific Shape classes are dynamically loaded on demand:
Now the factory method appears in its own class, ShapeFactory, as the create( ) method. This is a protected method which means it cannot be called directly, but it can be overridden. The subclasses of Shape must each create their own subclasses of ShapeFactory and override the create( ) method to create an object of their own type. The actual creation of shapes is performed by calling ShapeFactory.createShape( ), which is a static method that uses the Map in ShapeFactory to find the appropriate factory object based on an identifier that you pass it. The factory is immediately used to create the shape object, but you could imagine a more complex problem where the appropriate factory object is returned and then used by the caller to create an object in a more sophisticated way. However, it seems that much of the time you donít need the intricacies of the polymorphic factory method, and a single static method in the base class (as shown in ShapeFactory1.java) will work fine.
Notice that the ShapeFactory must be initialized by loading its Map with factory objects, which takes place in the static initialization clause of each of the Shape implementations. So to add a new type to this design you must inherit the type, create a factory, and add the static initialization clause to load the Map. This extra complexity again suggests the use of a static factory method if you donít need to create individual factory objects.
The Abstract Factory pattern looks like the factory objects weíve seen previously, with not one but several factory methods. Each of the factory methods creates a different kind of object. The idea is that at the point of creation of the factory object, you decide how all the objects created by that factory will be used. The example given in Design Patterns implements portability across various graphical user interfaces (GUIs): you create a factory object appropriate to the GUI that youíre working with, and from then on when you ask it for a menu, button, slider, etc. it will automatically create the appropriate version of that item for the GUI. Thus youíre able to isolate, in one place, the effect of changing from one GUI to another.
As another example suppose you are creating a general-purpose gaming environment and you want to be able to support different types of games. Hereís how it might look using an abstract factory:
In this environment, Player objects interact with Obstacle objects, but there are different types of players and obstacles depending on what kind of game youíre playing. You determine the kind of game by choosing a particular GameElementFactory, and then the GameEnvironment controls the setup and play of the game. In this example, the setup and play is very simple, but those activities (the initial conditions and the state change) can determine much of the gameís outcome. Here, GameEnvironment is not designed to be inherited, although it could very possibly make sense to do that.
This also contains examples of Double Dispatching and the Factory Method, both of which will be explained later.
1. Add a class Triangle to ShapeFactory1.java
2. Add a class Triangle to ShapeFactory2.java
3. Add a new type of GameEnvironment called GnomesAndFairies to Games.java
4. Modify ShapeFactory2.java so that it uses an Abstract Factory to create different sets of shapes (for example, one particular type of factory object creates ďthick shapes,Ē another creates ďthin shapes,Ē but each factory object can create all the shapes: circles, squares, triangles etc.).
Objects are created by cloning a prototypical instance. An example of this appears in the ďPattern RefactoringĒ chapter.
The goal of builder is to separate the construction from the ďrepresentation,Ē to allow multiple different representations. The construction process stays the same, but the resulting object has different possible representations. GoF points out that the main difference with Abstract Factory is that a Builder creates the object step-by-step, so the fact that the creation process is spread out in time seems to be important. In addition, it seems that the ďdirectorĒ gets a stream of pieces that it passes to the Builder, and each piece is used to perform one of the steps in the build process.
One example given in GoF is that of a text format converter. The incoming format is RTF, and once it is parsed the directives are passed to the text converter, which may be implemented in different ways depending on whether the resulting format is ASCII, TeX, or a ďGUI Text Widget.Ē Although the resulting ďobjectĒ (the entire converted text file) is created over time, if you consider the conversion of each RTF directive to be an object, this feels to me a little more like Bridge, because the specific types of converters extend the interface of the base class. Also, the general solution to the problem would allow multiple readers on the ďfront endĒ and multiple converters on the ďback end,Ē which is a primary characteristic of Bridge.
To me, the fact that Builder has multiple steps in creating an object, and those steps are accessed externally to the Builder object, is the essence of what distinguishes it (structurally, anyway) from a regular factory.† However, GoF emphasizes that youíre able to create different representations using the same process. They never define exactly what they mean by representation. (Does the ďrepresentationĒ involve an object that is too large? Would the need for Builder vanish if the representation was broken into smaller objects?)
The other example in GoF creates a maze object and adds rooms within the maze and doors within the rooms. Thus it is a multistep process, but alas, the different ďrepresentationsĒ are the ďStandardĒ and ďComplexĒ mazes Ė not really different kinds of mazes, but instead different complexity. I think I would have tried to create one maze builder that could handle arbitrarily complex mazes. The final variation of the maze builder is something that doesnít create mazes at all, but instead counts the rooms in an existing maze.
Neither the RTF converter nor the Mazebuilder example makes an overwhelmingly compelling case for Builder. Readers have suggested that the output of the Sax XML parser, and standard compiler parsers, might naturally be fed into a Builder.
Hereís an example that may be a little more compelling, or at least give more of an idea of what Builder is trying to do. Media may be constructed into different representations, in this case books, magazines and web sites. The example argues that the steps involved are the same, and so can be abstracted into the director class.
Note that in some ways this could be seen as a more complicated State pattern, since the behavior of the director depends on what type of builder you use. Instead of simply forwarding the requests through to the underlying State object, however, the director has a sequence of operations to perform, and it uses the State object as a Policy to fulfill its job. Thus, Builder could be described as using a Policy to create objects.
1. Break a text file up into an input stream of words (consider using regular expressions for this). Create one Builder that puts the words into a java.util.TreeSet, and another that produces a java.util.HashMap containing words and occurrences of those words (that is, it does a word count).
The odd thing about flyweight, in the company of the other design patterns, is that itís a performance hack. Itís generally ideal to simply make an object for every item in your system, but some problems generate a prohibitive number of objects, which may result in excessive slowness or running out of memory.
Flyweight solves this problem by reducing the number of objects. To do this, you externalize some of the data in an object, so that you can pretend that you have more objects than you really do. However, this adds complexity to the interface for using such objects, because you must pass in additional information to method calls in order to tell the method how to find the externalized information.
As a very simple example, consider a DataPoint object that holds an int, a float, and an id that carries the object number. Suppose you need to create a million of these objects, and then manipulate them, like so:
Depending on your computer, this program may take several seconds to run. More complex objects and more involved operations may cause the overhead to become untenable. To solve the problem the DataPoint can be reduced from a million objects to one object by externalizing the data held in the DataPoint:
Since all the data is now in ExternalizedData, each call to a FlyPoint method must include the index into ExternalizedData. For consistency, and to remind the reader of the similarity with the implicit this pointer in method calls, the ďthis indexĒ is passed in as the first argument.
Naturally, itís worth repeating admonishments against premature optimization. ďFirst make it work, then make it fast Ė if you have to.Ē Also, a profiler is the tool to use for discovering performance bottlenecks, not guesswork.
The use of layered objects to dynamically and transparently add responsibilities to individual objects is referred to as the decorator pattern.
Used when subclassing creates too many (& inflexible) classes
All decorators that wrap around the original object must have the same basic interface
This accounts for the odd inheritance structure
Tradeoff: coding is more complicated when using decorators
Consider going down to the local coffee shop, BeanMeUp, for a coffee.† There are typically many different drinks on offer -- espressos, lattes, teas, iced coffees, hot chocolate to name a few, as well as a number of extras (which cost extra too) such as whipped cream or an extra shot of espresso. You can also make certain changes to your drink at no extra cost, such as asking for decaf coffee instead of regular coffee.
Quite clearly if we are going to model all these drinks and combinations, there will be sizeable class diagrams. So for clarity we will only consider a subset of the coffees: Espresso, Espresso Con Panna, Cafť Late, Cappuccino and Cafť Mocha. We'll include 2 extras - whipped cream ("whipped") and an extra shot of espresso; and three changes - decaf, steamed milk ("wet") and foamed milk ("dry").
One solution is to create an individual class for every combination. Each class describes the drink and is responsible for the cost etc. The resulting menu is huge, and a part of the class diagram would look something like this:
Here is one of the combinations, a simple implementation of a Cappuccino:
The key to using this method is to find the particular combination you want.† So, once you've found the drink you would like, here is how you would use it, as shown in the CoffeeShop class in the following code:
And here is the corresponding output:
You can see that creating the particular combination you want is easy, since you are just creating an instance of a class. However, there are a number of problems with this approach. Firstly, the combinations are fixed statically so that any combination a customer may wish to order needs to be created up front. Secondly, the resulting menu is so huge that finding your particular combination is difficult and time consuming.
Another approach would be to break the drinks down into the various components such as espresso and foamed milk, and then let the customer combine the components to describe a particular coffee.
In order to do this programmatically, we use the Decorator pattern.† A Decorator adds responsibility to a component by wrapping it, but the Decorator conforms to the interface of the component it encloses, so the wrapping is transparent. Decorators can also be nested without the loss of this transparency.
Methods invoked on the Decorator can in turn invoke methods in the component, and can of course perform processing before or after the invocation.
So if we added getTotalCost() and getDescription() methods to the DrinkComponent interface, an Espresso looks like this:
You combine the components to create a drink as follows, as shown in the code below:
This approach would certainly provide the most flexibility and the smallest menu. You have a small number of components to choose from, but assembling the description of the coffee then becomes rather arduous.
If you want to describe a plain cappuccino, you create it with
Creating a decaf Cafť Mocha with whipped cream requires an even longer description.
The previous approach takes too long to describe a coffee. There will also be certain combinations that you will describe regularly, and it would be convenient to have a quick way of describing them.
The 3rd approach is a mixture of the first 2 approaches, and combines flexibility with ease of use. This compromise is achieved by creating a reasonably sized menu of basic selections, which would often work exactly as they are, but if you wanted to decorate them (whipped cream, decaf etc.) then you would use decorators to make the modifications. This is the type of menu you are presented with in most coffee shops.
Here is how to create a basic selection, as well as a decorated selection:
You can see that creating a basic selection is quick and easy, which makes sense since they will be described regularly.† Describing a decorated drink is more work than when using a class per combination, but clearly less work than when only using decorators.
The final result is not too many classes, but not too many decorators either. Most of the time it's possible to get away without using any decorators at all, so we have the benefits of both approaches.
What happens if we decide to change the menu at a later stage, such as by adding a new type of drink? If we had used the class per combination approach, the effect of adding an extra such as syrup would be an exponential growth in the number of classes. However, the implications to the all decorator or compromise approaches are the same - one extra class is created.
How about the effect of changing the cost of steamed milk and foamed milk, when the price of milk goes up? Having a class for each combination means that you need to change a method in each class, and thus maintain many classes. By using decorators, maintenance is reduced by defining the logic in one place.
1. Add a Syrup class to the decorator approach described above. Then create a Cafť Latte (you'll need to use steamed milk with an espresso) with syrup.
2. Repeat Exercise 1 for the compromise approach.
3. Create a simple decorator system that models the fact that some birds fly and some donít, some swim and some donít, and some do both.
4. Implement the decorator pattern to create a Pizza restaurant, which has a set menu of choices as well as the option to design your own pizza.† Follow the compromise approach to create a menu consisting of a Margherita, Hawaiian, Regina, and Vegetarian pizzas, with toppings (decorators) of Garlic, Olives, Spinach, Avocado, Feta and Pepperdews. Create a Hawaiian pizza, as well as a Margherita decorated with Spinach, Feta, Pepperdews and Olives.
5. About Decorator, the Design Patterns book states: ďWith decorators, responsibilities can be added and removed at run-time simply by attaching and detaching them.Ē Implement the coffee decoration system to allow this ďsimpleĒ detaching of a responsibility from the middle of the list of decorators of a complex coffee beverage.
Adapter takes one type and produces an interface to some other type.When youíve got this, and you need that, Adapter solves the problem. The only requirement is to produce a that, and there are a number of ways you can accomplish this adaptation.
Iím taking liberties with the term ďproxyĒ here, because in Design Patterns they assert that a proxy must have an identical interface with the object that it is a surrogate for. However, if you have the two words together: ďproxy adapter,Ē it is perhaps more reasonable.
While researching Bridge, I discovered that it appears to be the most poorly-described pattern in the GoF. I began to come to this conclusion when reading Alan Shallowayís chapter on Bridge in his book Design Patterns Explained Ė he begins by pointing out that the description in GoF left him quite unenlightened.
At a conference, I talked to two people who had written about and were giving talks on design patterns, including Bridge. In two separate discussions I got completely different perspectives on the structure of Bridge.
Armed with misinformation, I delved back into the GoF and realized that neither of the above perspectives agreed with the book. I also found that the book did a miserable job of describing Bridge, except in one place Ė not the general structure chart describing the pattern, which wasnít helpful, but in the structure chart describing their specific example. Only if you stare at that for a bit does Bridge begin to make sense.
An important feature to understand when looking at Bridge is that it is often a construct that is used to help you write code. You may choose the objects you use for a particular situation at compile-time or runtime, but the goal of Bridge is to allow you to structure your code so that you can easily add new kinds of front-end objects which are implemented with functionality in new kinds of back-end objects. Thus, both front-end and back-end can vary independently of each other.
The front-end classes can have completely different interfaces from each other, and typically do. What they have in common is that they can implement their functionality using facilities from any number of different back-end objects. The back-end objects also donít have the same interface. The only thing the back-end objects must have in common is that they implement the same kind of functionality Ė for example, a group of different ways to implement a graphics library or a set of different data-storage solutions.
Bridge is really a code-organization tool that allows you to add in any number of new front-end services that implement their operations by delegating to any number of back-end options. Using Bridge, you can accomplish this without the normal combinatorial explosion of possibilities that would otherwise occur. But keep in mind that the vector of change with Bridge is typically happening at coding time: it keeps your code organized when you are dealing with an increasing number of options for implementing functionality.
Hereís an example whose sole purpose is to demonstrate the structure of Bridge (it implements the above diagram):
The front-end base class provides the operations used to fulfill the front-end derived classes, in terms of the methods of the back-end base class. Thus, any back-end derived class can be used to perform the operations needed by the front-end class. Notice that the bridging happens in this sequence of steps, each of which is providing a layer of abstraction. Here, Implementation is defined as an interface to emphasize that all the functionality is implemented in the back-end derived classes, and none in the back-end base.
The back-end derived classes perform the operations defined in the base class by delegating to objects (of class Library1 and Library2, in this case) that typically have radically different interfaces, but somehow provide the same functionality (this is one of the important objectives of Bridge). Effectively, each of the back-end implementations is an adapter to a different library or tool used to implement the desired functionality in a different way.
1. Modify BridgeStructure.java so that the implementations are chosen using a factory.
2. Modify BridgeStructure.java to use delegation instead of inheritance on the front end. What benefits and drawbacks do you see by using delegation rather than inheritance?
3. Create an example of Bridge with an abstraction that is an associative array. This allows you to fetch elements by passing in a key Object. The constructor provides an initial set of key-value pairs which are placed in an array. As long as you only fetch elements, the array is used, but as soon as you set a new key-value pair, the implementation is switched to a map.
4. Use a bridge along with the collections in java.util.collections to create stack and queue classes using an ArrayList. After you get the system working, add a double-ended queue class. Now add a LinkedList as an implementation. These steps will demonstrate how Bridge allows you to add new front-end classes and new back-end classes in your code, with minimal impact.
5. Create a Bridge that provides a connection between various kinds of bookkeeping programs (along with their interfaces and data formats) and different banks (which provide different kinds of services and interfaces).
The important thing here is that all elements in the part-whole have operations, and that performing an operation on a node/composite also performs that operation on any children of that node/composite. GoF includes implementation details of containment and visitation of children in the interface of the base class, but this doesnít seem necessary. In the following example, the Composite class simply inherits ArrayList in order to gain its containment abilities.
While this approach seems to be ďthe simplest thing that could possibly work,Ē itís possible that in a larger system problems could arise. However, itís probably best to start with the simplest approach and change it only if the situation demands.
Like the other forms of callback, this contains a hook point where you can change code. The difference is in the observerís completely dynamic nature. It is often used for the specific case of changes based on other objectís change of state, but is also the basis of event management. Anytime you want to decouple the source of the call from the called code in a completely dynamic way.
The observer pattern solves a fairly common problem: What if a group of objects needs to update themselves when some object changes state? This can be seen in the ďmodel-viewĒ aspect of Smalltalkís MVC (model-view-controller), or the almost-equivalent ďDocument-View Architecture.Ē Suppose that you have some data (the ďdocumentĒ) and more than one view, say a plot and a textual view. When you change the data, the two views must know to update themselves, and thatís what the observer facilitates. Itís a common enough problem that its solution has been made a part of the standard java.util library.
There are two types of objects used to implement the observer pattern in Java. The Observable class keeps track of everybody who wants to be informed when a change happens, whether the ďstateĒ has changed or not. When someone says ďOK, everybody should check and potentially update themselves,Ē the Observable class performs this task by calling the notifyObservers( ) method for each one on the list. The notifyObservers( ) method is part of the base class Observable.
There are actually two ďthings that changeĒ in the observer pattern: the quantity of observing objects and the way an update occurs. That is, the observer pattern allows you to modify both of these without affecting the surrounding code.
Observer is an ďinterfaceĒ class that only has one member function, update( ). This function is called by the object thatís being observed, when that object decides its time to update all its observers. The arguments are optional; you could have an update( ) with no arguments and that would still fit the observer pattern; however this is more generalóit allows the observed object to pass the object that caused the update (since an Observer may be registered with more than one observed object) and any extra information if thatís helpful, rather than forcing the Observer object to hunt around to see who is updating and to fetch any other information it needs.
The ďobserved objectĒ that decides when and how to do the updating will be called the Observable.
Observable has a flag to indicate whether itís been changed. In a simpler design, there would be no flag; if something happened, everyone would be notified. The flag allows you to wait, and only notify the Observers when you decide the time is right. Notice, however, that the control of the flagís state is protected, so that only an inheritor can decide what constitutes a change, and not the end user of the resulting derived Observer class.
Most of the work is done in notifyObservers( ). If the changed flag has not been set, this does nothing. Otherwise, it first clears the changed flag so repeated calls to notifyObservers( ) wonít waste time. This is done before notifying the observers in case the calls to update( ) do anything that causes a change back to this Observable object. Then it moves through the set and calls back to the update( ) member function of each Observer.
At first it may appear that you can use an ordinary Observable object to manage the updates. But this doesnít work; to get an effect, you must inherit from Observable and somewhere in your derived-class code call setChanged( ). This is the member function that sets the ďchangedĒ flag, which means that when you call notifyObservers( ) all of the observers will, in fact, get notified. Where you call setChanged( ) depends on the logic of your program.
Here is an example of the observer pattern:
The events of interest are that a Flower can open or close. Because of the use of the inner class idiom, both these events can be separately observable phenomena. OpenNotifier and CloseNotifier both inherit Observable, so they have access to setChanged( ) and can be handed to anything that needs an Observable.
The inner class idiom also comes in handy to define more than one kind of Observer, in Bee and Hummingbird, since both those classes may want to independently observe Flower openings and closings. Notice how the inner class idiom provides something that has most of the benefits of inheritance (the ability to access the private data in the outer class, for example) without the same restrictions.
In main( ), you can see one of the prime benefits of the observer pattern: the ability to change behavior at run time by dynamically registering and un-registering Observers with Observables.
If you study the code above youíll see that OpenNotifier and CloseNotifier use the basic Observable interface. This means that you could inherit other completely different Observer classes; the only connection the Observers have with Flowers is the Observer interface.
The following example is similar to the ColorBoxes example from Chapter 14 in Thinking in Java, 2nd Edition. Boxes are placed in a grid on the screen and each one is initialized to a random color. In addition, each box implements the Observer interface and is registered with an Observable object. When you click on a box, all of the other boxes are notified that a change has been made because the Observable object automatically calls each Observer objectís update( ) method. Inside this method, the box checks to see if itís adjacent to the one that was clicked, and if so it changes its color to match the clicked box.
When you first look at the online documentation for Observable, itís a bit confusing because it appears that you can use an ordinary Observable object to manage the updates. But this doesnít work; try itóinside BoxObserver, create an Observable object instead of a BoxObservable object and see what happens: nothing. To get an effect, you must inherit from Observable and somewhere in your derived-class code call setChanged( ). This is the method that sets the ďchangedĒ flag, which means that when you call notifyObservers( ) all of the observers will, in fact, get notified. In the example above setChanged( ) is simply called within notifyObservers( ), but you could use any criterion you want to decide when to call setChanged( ).
BoxObserver contains a single Observable object called notifier, and every time an OCBox object is created, it is tied to notifier. In OCBox, whenever you click the mouse the notifyObservers( ) method is called, passing the clicked object in as an argument so that all the boxes receiving the message (in their update( ) method) know who was clicked and can decide whether to change themselves or not. Using a combination of code in notifyObservers( ) and update( ) you can work out some fairly complex schemes.
It might appear that the way the observers are notified must be frozen at compile time in the notifyObservers( ) method. However, if you look more closely at the code above youíll see that the only place in BoxObserver or OCBox where you're aware that youíre working with a BoxObservable is at the point of creation of the Observable objectófrom then on everything uses the basic Observable interface. This means that you could inherit other Observable classes and swap them at run time if you want to change notification behavior then.
Sweep coupling under the rug, how is this different from MVC?
MVC has distinct model and view; mediator could be anything. MVC a flavor of mediator
1. Create a minimal Observer-Observable design in two classes. Just create the bare minimum in the two classes, then demonstrate your design by creating one Observable and many Observers, and cause the Observable to update the Observers.
2. Create a minimal Observer system using java.util.Timer inside your Observable, to generate events that are reported to the Observers. Create several different Observers using anonymous inner classes, register these with the Observable, and show that they are called when the Timer events occur.
3. Modify BoxObserver.java to turn it into a simple game. If any of the squares surrounding the one you clicked is part of a contiguous patch of the same color, then all the squares in that patch are changed to the color you clicked on. You can configure the game for competition between players or to keep track of the number of clicks that a single player uses to turn the field into a single color. You may also want to restrict a player's color to the first one that was chosen.
Sometimes the problem that youíre solving is as simple as ďI donít have the interface that I want.Ē FaÁade creates an interface to a set of classes, simply to provide a more comfortable way to deal with a library or bundle of resources.
A general principle that I apply when Iím casting about trying to mold requirements into a first-cut object is ďIf something is ugly, hide it inside an object.Ē This is basically what FaÁade accomplishes. If you have a rather confusing collection of classes and interactions that the client programmer doesnít really need to see, then you can create an interface that is useful for the client programmer and that only presents whatís necessary.
FaÁade is often implemented as singleton abstract factory. Of course, you can easily get this effect by creating a class containing static factory methods:
The example given in Design Patterns is just a class that uses the other classes.
A tax adviser is a FaÁade between you and the tax code, and a mediator between you and the tax system.
To me, the FaÁade has a rather ďproceduralĒ (non-object-oriented) feel to it: you are just calling some functions to give you objects. And how different is it, really, from Abstract Factory? The point of FaÁade is to hide part of a library of classes (and their interactions) from the client programmer, to make the interface to that group of classes more digestible and easier to understand.
However, this is precisely what the packaging features in Java accomplish: outside of the library, you can only create and use public classes; all the non-public classes are only accessible within the package. Itís as if FaÁade is a built-in feature of Java.
To be fair, Design Patterns is written primarily for a C++ audience. Although C++ has namespaces to prevent clashes of globals and class names, this does not provide the class hiding mechanism that you get with non-public classes in Java. The majority of the time I think that Java packages will solve the FaÁade problem.
In Advanced C++:Programming Styles And Idioms (Addison-Wesley, 1992), Jim Coplien coins the term functor which is an object whose sole purpose is to encapsulate a function (since ďfunctorĒ has a meaning in mathematics, in this book I shall use the more explicit term function object). The point is to decouple the choice of function to be called from the site where that function is called.
This term is mentioned but not used in Design Patterns. However, the theme of the function object is repeated in a number of patterns in that book.
A Command is a function object in its purest sense: a method thatís an object. By wrapping a method in an object, you can pass it to other methods or objects as a parameter, to tell them to perform this particular operation in the process of fulfilling your request. You could say that a Command is a messenger (because its intent and use is very straightforward) that carries behavior, rather than data.
The primary point of Command is to allow you to hand a desired action to a method or object. In the above example, this provides a way to queue a set of actions to be performed collectively. In this case, it allows you to dynamically create new behavior, something you can normally only do by writing new code but in the above example could be done by interpreting a script (see the Interpreter pattern if what you need to do gets very complex).
Another example of Command is refactor:DirList.java [????]. The DirFilter class is the command object which contains its action in the method accept( ) that is passed to the list( ) method. The list( ) method determines what to include in its result by calling accept( ).
.Ē However, I think that the word ďbackĒ is an essential part of the concept of callbacks. That is, I think a callback actually reaches back to the creator of the callback. On the other hand, with a Command object you typically just create it and hand it to some method or object, and are not otherwise connected over time to the Command object. Thatís my take on it, anyway. Later in this book, I combine a group of design patterns under the heading of ďcallbacks.ĒDesign Patterns says that ďCommands are an object-oriented replacement for callbacks
Strategy appears to be a family of Command classes, all inherited from the same base. But if you look at Command, youíll see that it has the same structure: a hierarchy of function objects. The difference is in the way this hierarchy is used. As seen in refactor:DirList.java, you use Command to solve a particular problemóin that case, selecting files from a list. The ďthing that stays the sameĒ is the body of the method thatís being called, and the part that varies is isolated in the function object. I would hazard to say that Command provides flexibility while youíre writing the program, whereas Strategyís flexibility is at run time. Nonetheless, it seems a rather fragile distinction.
4. Use Command in Chapter 3, Exercise 1.
Example: translation service (local->global->Babelfish).
Chain of Responsibility might be thought of as a dynamic generalization of recursion using Strategy objects. You make a call, and each Strategy in a linked sequence tries to satisfy the call. The process ends when one of the strategies is successful or the chain ends. In recursion, one method calls itself over and over until a termination condition is reached; with Chain of Responsibility, a method calls itself, which (by moving down the chain of Strategies) calls a different implementation of the method, etc., until a termination condition is reached. The termination condition is either the bottom of the chain is reached (in which case a default object is returned; you may or may not be able to provide a default result so you must be able to determine the success or failure of the chain) or one of the Strategies is successful.
Instead of calling a single method to satisfy a request, multiple methods in the chain have a chance to satisfy the request, so it has the flavor of an expert system. Since the chain is effectively a linked list, it can be dynamically created, so you could also think of it as a more general, dynamically-built switch statement.
In the GoF, thereís a fair amount of Thidiscussion of how to create the chain of responsibility as a linked list. However, when you look at the pattern it really shouldnít matter how the chain is maintained; thatís an implementation detail. Since GoF was written before the Standard Template Library (STL) was incorporated into most C++ compilers, the reason for this is most likely (1) there was no list and thus they had to create one and (2) data structures are often taught as a fundamental skill in academia, and the idea that data structures should be standard tools available with the programming language may not have occurred to the GoF authors. I maintain that the implementation of Chain of Responsibility as a chain (specifically, a linked list) adds nothing to the solution and can just as easily be implemented using a standard Java List, as shown below. Furthermore, youíll see that Iíve gone to some effort to separate the chain-management parts of the implementation from the various Strategies, so that the code can be more easily reused.
In StrategyPattern.java, above, what you probably want is to automatically find a solution. Chain of Responsibility provides a way to do this by chaining the Strategy objects together and providing a mechanism for them to automatically recurse through each one in the chain:
1. Implement Chain of Responsibility to create an "expert system" that solves problems by successively trying one solution after another until one matches. You should be able to dynamically add solutions to the expert system. The test for solution should just be a string match, but when a solution fits, the expert system should return the appropriate type of ProblemSolver object. What other pattern/patterns show up here?
2. Implement Chain of Responsibility to create a language translator which begins by searching for a local specialized translation system (which may know specifics about your problem domain), then a more global generalized system, and finally falls back on BabelFish if it canít translate everything. Note that each link in the chain may partially translate what itís able to.
3. Implement Chain of Responsibility to create an tool to help reformat Java source code by trying multiple approaches to breaking lines. Note that normal code and comments will probably need to be treated differently, leading to the possibility of implementing a Tree of Responsibility. Also note the similarity between this approach and the Composite design pattern; perhaps the more general description of this technique is a Composite of Strategies.
Use serialization to create an undo mechanism.
When dealing with multiple types which are interacting, a program can get particularly messy. For example, consider a system that parses and executes mathematical expressions. You want to be able to say Number + Number, Number * Number, etc., where Number is the base class for a family of numerical objects. But when you say a + b, and you donít know the exact type of either a or b, so how can you get them to interact properly?
The answer starts with something you probably donít think about: Java performs only single dispatching. That is, if you are performing an operation on more than one object whose type is unknown, Java can invoke the dynamic binding mechanism on only one of those types. This doesnít solve the problem, so you end up detecting some types manually and effectively producing your own dynamic binding behavior.
The solution is called multiple dispatching. Remember that polymorphism can occur only via member function calls, so if you want double dispatching to occur, there must be two member function calls: the first to determine the first unknown type, and the second to determine the second unknown type. With multiple dispatching, you must have a polymorphic method call to determine each of the types. Generally, youíll set up a configuration such that a single member function call produces more than one dynamic member function call and thus determines more than one type in the process. To get this effect, you need to work with more than one polymorphic method call: youíll need one call for each dispatch. The methods in the following example are called compete( ) and eval( ), and are both members of the same type. (In this case there will be only two dispatches, which is referred to as double dispatching). If you are working with two different type hierarchies that are interacting, then youíll have to have a polymorphic method call in each hierarchy.
Hereís an example of multiple dispatching:
The assumption is that you have a primary class hierarchy that is fixed; perhaps itís from another vendor and you canít make changes to that hierarchy. However, youíd like to add new polymorphic methods to that hierarchy, which means that normally youíd have to add something to the base class interface. So the dilemma is that you need to add methods to the base class, but you canít touch the base class. How do you get around this?
The design pattern that solves this kind of problem is called a ďvisitorĒ (the final one in the Design Patterns book), and it builds on the double dispatching scheme shown in the last section.
The visitor pattern allows you to extend the interface of the primary type by creating a separate class hierarchy of type Visitor to virtualize the operations performed upon the primary type. The objects of the primary type simply ďacceptĒ the visitor, then call the visitorís dynamically-bound member function.
1. Create a business-modeling environment with three types of Inhabitant: Dwarf (for engineers), Elf (for marketers) and Troll (for managers). Now create a class called Project that creates the different inhabitants and causes them to interact( ) with each other using multiple dispatching.
2. Modify the above example to make the interactions more detailed. Each Inhabitant can randomly produce a Weapon using getWeapon( ): a Dwarf uses Jargon or Play, an Elf uses InventFeature or SellImaginaryProduct, and a Troll uses Edict and Schedule. You must decide which weapons ďwinĒ and ďloseĒ in each interaction (as in PaperScissorsRock.java). Add a battle( ) member function to Project that takes two Inhabitants and matches them against each other. Now create a meeting( ) member function for Project that creates groups of Dwarf, Elf and Manager and battles the groups against each other until only members of one group are left standing. These are the ďwinners.Ē
Modify PaperScissorsRock.java to replace the double
dispatching with a table lookup. The easiest way to do this is to create a Map
of Maps, with the key of each Map the class of each object.
Then you can do the lookup by saying:
Notice how much easier it is to reconfigure the system. When is it more appropriate to use this approach vs. hard-coding the dynamic dispatches? Can you create a system that has the syntactic simplicity of use of the dynamic dispatch but uses a table lookup?
4. Modify Exercise 2 to use the table lookup technique described in Exercise 3.
This chapter looks at the value of crossing language boundaries. It is often very advantageous to solve a problem using more than one programming language, rather than being arbitrarily stuck using a single language. As youíll see in this chapter, a problem that is very difficult or tedious to solve in one language can often be solved quickly and easily in another. If you can combine the use of languages, you can create your product much more quickly and cheaply.
, an implementation of the Python language in pure Java byte codes.The most straightforward use of this idea is the Interpreter design pattern, which adds an interpreted language to your program to allow the end user to easily customize a solution. In Java, the easiest and most powerful way to do this is with Jython
Interpreter solves a particular problem Ė that of creating a scripting language for the user. But sometimes itís just easier and faster to temporarily step into another language to solve a particular aspect of your problem. Youíre not creating an interpreter, youíre just writing some code in another language. Again, Jython is a good example of this, but CORBA also allows you to cross language boundaries.
If the application user needs greater run time flexibility, for example to create scripts describing the desired behavior of the system, you can use the Interpreter design pattern. Here, you create and embed a language interpreter into your program.
Remember that each design pattern allows one or more factors to change, so itís important to first be aware of which factor is changing. Sometimes the end users of your application (rather than the programmers of that application) need complete flexibility in the way that they configure some aspect of the program. That is, they need to do some kind of simple programming. The interpreter pattern provides this flexibility by adding a language interpreter.
The problem is that developing your own language and building an interpreter is a time-consuming distraction from the process of developing your application. You must ask whether you want to finish writing your application or create a new language.† The best solution is to reuse code: embed an interpreter thatís already been built and debugged for you. The Python language can be freely embedded into your for-profit application without signing any license agreement, paying royalties, or dealing with strings of any kind. There are basically no restrictions at all when you're using Python.
Python is a language that is very easy to learn, very logical to read and write, supports functions and objects, has a large set of available libraries, and runs on virtually every platform. You can download Python and learn more about it by going to www.Python.org.
For solving Java problems, we will look at a special version of Python called Jython. This is generated entirely in Java byte codes, so incorporating it into your application is quite simple,† and itís as portable as Java is. It has an extremely clean interface with Java: Java can call Python classes, and Python can call Java classes.
Python is designed with classes from the ground up and is a truly pure object oriented language (both C++ and Java violate purity in various ways). Python scales up so that you can create very big programs without losing control of the code.
www.Python.org and follow the links and instructions. To install Jython, go to http://jython.sourceforge.net.† The download is a .class file, which will run an installer when you execute it with Java.† You also need to add jython.jar to your CLASSPATH. You can find further installation instructions at http://www.bruceeckel.com/TIPatterns/Building-Code.html.To install Python, go to
To get you started, here is a brief introduction for the experienced programmer (which is what you should be if youíre reading this book). You can refer to the full documentation at www.Python.org (especially the incredibly useful HTML page A Python Quick Reference), and also numerous books such as Learning Python by Mark Lutz and David Ascher (OíReilly, 1999).
Python is often referred to as a scripting language, but scripting languages tend to be limiting, especially in the scope of the problems that they solve. Python, on the other hand, is a programming language that also supports scripting. It is marvelous for scripting, and you may find yourself replacing all your batch files, shell scripts, and simple programs with Python scripts. But it is far more than a scripting language.
Python is designed to be very clean to write and especially to read. You will find that itís quite easy to read your own code long after youíve written it, and also to read other peopleís code. This is accomplished partially through clean, to-the-point syntax, but a major factor in code readability is indentation Ė scoping in Python is determined by indentation. For example:
The Ď#í denotes a comment that goes until the end of the line, just like C++ and Java Ď//í comments.
First notice that the basic syntax of Python is C-ish; notice the if statement. But in a C if, you would be required to use parentheses around the conditional, whereas they are not necessary in Python (but it wonít complain if you use them anyway).
The conditional clause ends with a colon, and this indicates that what follows will be a group of indented statements, which are the ďthenĒ part of the if statement. In this case, there is a ďprintĒ statement which sends the result to standard output, followed by an assignment to a variable named val. The subsequent statement is not indented so it is no longer part of the if. Indenting can nest to any level, just like curly braces in C++ or Java, but unlike those languages there is no option (and no argument) about where the braces are placed Ė the compiler forces everyoneís code to be formatted the same way, which is one of the main reasons for Pythonís consistent readability.
Python normally has only one statement per line (you can put more by separating them with semicolons), thus no terminating semicolon is necessary.† Even from the brief example above you can see that the language is designed to be as simple as possible, and yet still very readable.
With languages like C++ and Java, containers are add-on libraries and not integral to the language. In Python, the essential nature of containers for programming is acknowledged by building them into the core of the language: both lists and associative arrays (a.k.a. maps, dictionaries, hash tables) are fundamental data types. This adds much to the elegance of the language.
In addition, the for statement automatically iterates through lists rather than just counting through a sequence of numbers. This makes a lot of sense when you think about it, since youíre almost always using a for loop to step through an array or a container. Python formalizes this by automatically making for use an iterator that works through a sequence. Hereís an example:
The first line creates a list. You can print the list and it will look exactly as you put it in (in contrast, remember that I had to create a special Arrays2 class in Thinking in Java, 2nd Edition in order to print arrays in Java). Lists are like Java containers Ė you can add new elements to them (here, append( ) is used) and they will automatically resize themselves. The for statement creates an iterator x which takes on each value in the list.
You can create a list of numbers with the range( ) function, so if you really need to imitate Cís for, you can.
Notice that there arenít any type declarations Ė the object names simply appear, and Python infers their type by the way that you use them. Itís as if Python is designed so that you only need to press the keys that absolutely must. Youíll find after youíve worked with Python for a short while that youíve been using up a lot of brain cycles parsing semicolons, curly braces, and all sorts of other extra verbiage that was demanded by your non-Python programming language but didnít actually describe what your program was supposed to do.
To create a function in Python, you use the def keyword, followed by the function name and argument list, and a colon to begin the function body. Here is the first example turned into a function:
Notice there is no type information in the function signature Ė all it specifies is the name of the function and the argument identifiers, but no argument types or return types. Python is a weakly-typed language, which means it puts the minimum possible requirements on typing. For example, you could pass and return different types from the same function:
The only constraints on an object that is passed into the function are that the function can apply its operations to that object, but other than that, it doesnít care. Here, the same function applies the Ď+í operator to integers and strings:
When the operator Ď+í is used with strings, it means concatenation (yes, Python supports operator overloading, and it does a nice job of it).
The above example also shows a little bit about Python string handling,† which is the best of any language Iíve seen. You can use single or double quotes to represent strings, which is very nice because if you surround a string with double quotes, you can embed single quotes and vice versa:
Note that Python was not named after the snake, but rather the Monty Python comedy troupe, and so examples are virtually required to include Python-esque references.
The triple-quote syntax quotes everything, including newlines. This makes it particularly useful for doing things like generating web pages (Python is an especially good CGI language), since you can just triple-quote the entire page that you want without any other editing.
The Ďrí right before a string means ďraw,Ē which takes the backslashes literally so you donít have to put in an extra backslash.
Substitution in strings is exceptionally easy, since Python uses Cís printf( ) substitution syntax, but for any string at all. You simply follow the string with a Ď%í and the values to substitute:
As you can see in the second case, if you have more than one argument you surround them in parentheses (this forms a tuple, which is a list that cannot be modified).
All the formatting from printf( ) is available, including control over the number of decimal places and alignment. Python also has very sophisticated regular expressions.
Like everything else in Python, the definition of a class uses a minimum of additional syntax. You use the class keyword, and inside the body you use def to create methods. Hereís a simple class:
Both methods have ďselfĒ as their first argument. C++ and Java both have a hidden first argument in their class methods, which points to the object that the method was called for and can be accessed using the keyword this. Python methods also use a reference to the current object, but when you are defining a method you must explicitly specify the reference as the first argument. Traditionally, the reference is called self but you could use any identifier you want (if you do not use self you will probably confuse a lot of people, however). If you need to refer to fields in the object or other methods in the object, you must use self in the expression. However, when you call a method for an object as in x.show( ), you do not hand it the reference to the object Ė that is done for you.
Here, the first method is special, as is any identifier that begins and ends with double underscores. In this case, it defines the constructor, which is automatically called when the object is created, just like in C++ and Java. However, at the bottom of the example you can see that the creation of an object looks just like a function call using the class name. Pythonís spare syntax makes you realize that the new keyword isnít really necessary in C++ or Java, either.
All the code at the bottom is set off by an if clause, which checks to see if something called __name≠__ is equivalent to __main__. Again, the double underscores indicate special names. The reason for the if is that any file can also be used as a library module within another program (modules are described shortly). In that case, you just want the classes defined, but you donít want the code at the bottom of the file to be executed. This particular if statement is only true when you are running this file directly; that is, if you say on the command line:
However, if this file is imported as a module into another program, the __main__ code is not executed.
Something thatís a little surprising at first is that you define fields inside methods, and not outside of the methods like C++ or Java (if you create fields using the C++/Java style, they implicitly become static fields). To create an object field, you just name it Ė using self Ė inside of one of the methods (usually in the constructor, but not always), and space is created when that method is run. This seems a little strange coming from C++ or Java where you must decide ahead of time how much space your object is going to occupy, but it turns out to be a very flexible way to program.
Because Python is weakly typed, it doesnít really care about interfaces Ė all it cares about is applying operations to objects (in fact, Javaís interface keyword would be wasted in Python). This means that inheritance in Python is different from inheritance in C++ or Java, where you often inherit simply to establish a common interface. In Python, the only reason you inherit is to inherit an implementation Ė to re-use the code in the base class.
If youíre going to inherit from a class, you must tell Python to bring that class into your new file. Python controls its name spaces as aggressively as Java does, and in a similar fashion (albeit with Pythonís penchant for simplicity). Every time you create a file, you implicitly create a module (which is like a package in Java) with the same name as that file. Thus, no package keyword is needed in Python. When you want to use a module, you just say import and give the name of the module. Python searches the PYTHONPATH in the same way that Java searches the CLASSPATH (but for some reason, Python doesnít have the same kinds of pitfalls as Java does) and reads in the file. To refer to any of the functions or classes within a module, you give the module name, a period, and the function or class name. If you donít want the trouble of qualifying the name, you can say
from module import name(s)
Where ďname(s)Ē can be a list of names separated by commas.
You inherit a class (or classes Ė Python supports multiple inheritance) by listing the name(s) of the class inside parentheses after the name of the inheriting class. Note that the Simple class, which resides in the file (and thus, module) named SimpleClass is brought into this new name space using an import statement:
Simple2 is inherited from Simple, and in the constructor, the base-class constructor is called. In display( ), showMsg( ) can be called as a method of self, but when calling the base-class version of the method you are overriding, you must fully qualify the name and pass self in as the first argument, as shown in the base-class constructor call. This can also be seen in the overridden version of show( ).
In __main__, you will see (when you run the program) that the base-class constructor is called. You can also see that the showMsg( ) method is available in the derived class, just as you would expect with inheritance.
The class Different also has a method named show( ), but this class is not derived from Simple. The f( ) method defined in __main__ demonstrates weak typing: all it cares about is that show( ) can be applied to obj, and it doesnít have any other type requirements. You can see that f( ) can be applied equally to an object of a class derived from Simple and one that isnít, without discrimination. If youíre a C++ programmer, you should see that the objective of the C++ template feature is exactly this: to provide weak typing in a strongly-typed language. Thus, in Python you automatically get the equivalent of templates Ė without having to learn that particularly difficult syntax and semantics.
It turns out to be remarkably simple to use Jython to create an interpreted language inside your application. Consider the greenhouse controller example from Chapter 8 of Thinking in Java, 2nd edition. This is a situation where you want the end user Ė the person managing the greenhouse Ė to have configuration control over the system, and so a simple scripting language is the ideal solution.
To create the language, weíll simply write a set of Python classes, and the constructor of each will add itself to a (static) master list. The common data and behavior will be factored into the base class Event. Each Event object will contain an action string (for simplicity Ė in reality, youíd have some sort of functionality) and a time when the event is supposed to run. The constructor initializes these fields, and then adds the new Event object to a static list called events (defining it in the class, but outside of any methods, is what makes it static):
The constructor of each derived class calls the base-class constructor, which adds the new object to the list. The run( ) function sorts the list, which automatically uses the __cmp__( ) method that was defined in Event to base comparisons on time only. In this example, it only prints out the list, but in the real system it would wait for the time of each event to come up and then run the event.
The __main__ section performs a simple test on the classes.
The above file is now a module that can be included in another Python program to define all the classes it contains. But instead of an ordinary Python program, letís use Jython, inside of Java. This turns out to be remarkably simple: you import some Jython classes, create a PythonInterpreter object, and cause the Python files to be loaded:
The PythonInterpreter object is a complete Python interpreter that accepts commands from the Java program. One of these commands is execfile( ), which tells it to execute all the statements it finds in a particular file. By executing GreenHouseLanguage.py, all the classes from that file are loaded into our PythonInterpreter object, and so it now ďholdsĒ the greenhouse controller language. The Schedule.ghs file is the one created by the end user to control the greenhouse. Hereís an example:
This is the goal of the interpreter design pattern: to make the configuration of your program as simple as possible for the end user. With Jython you can achieve this with almost no effort at all.
One of the other methods available to the PythonInterpreter is exec( ), which allows you to send a command to the interpreter. Here, the run( ) function is called using exec( ).
http://jython.sourceforge.net and download and install Jython (actually, you only need jython.jar in your CLASSPATH). Once thatís in place, itís just like running any other Java program.Remember, to run this program you must go to
The prior example only creates and runs the interpreter using external scripts. In the rest of this chapter, we shall look at more sophisticated ways to interact with Jython. The simplest way to exercise more control over the PythonInterpreter object from within Java is to send data to the interpreter, and pull data back out.
To inject data into your Python program, the PythonInterpreter class has a deceptively simple method: set( ). However, set( ) takes many different data types and performs conversions upon them.† The following example is a reasonably thorough exercise of the various set( ) possibilities, along with comments that should give a fairly complete explanation:
As usual with Java, the distinction between real objects and primitive types causes trouble. In general, if you pass a regular object to set( ), it knows what to do with it, but if you want to pass in a primitive you must perform a conversion. One way to do this is to create a ďPyĒ type, such as PyInteger or PyFloat. but it turns out you can also use Javaís own object wrappers like Integer and Float, which is probably going to be a lot easier to remember.
Early in the program youíll see an exec( ) containing the Python statement:
The colon inside the indexing statement indicates a Python slice, which produces a range of elements from the original array. In this case, it produces an array containing the elements from number 5 until the end of the array. You could also say Ďa[3:5]í to produce elements 3 through 5, or Ďa[:5]í to produce the elements zero through 5. The reason a slice is used in this statement is to make sure that the Java String has really been converted to a Python string, which can also be treated as an array of characters.
You can see that itís possible, using exec( ), to create a Python function (although itís a bit awkward). The prt( ) function prints the whole array, and then (to make sure itís a real Python array), iterates through each element of the array and prints it. Finally, it prints the class of the array, so we can see what conversion has taken place (Python not only has run-time type information, it also has the equivalent of Java reflection). The prt( ) function is used to print arrays that come from each of the Java primitive types.
Although a Java ArrayList does pass into the interpreter using set( ), and you can index into it as if it were an array, trying to create a slice fails. To completely convert it into an array, one approach is to simply extract a Java array using toArray( ), and pass that in. The set( ) method converts it to a PyArray Ė one of the classes provided with Jython Ė which can be treated as a Python array (you can also explicitly create a PyArray, but this seems unnecessary).
Finally, a Map is created and passed directly into the interpreter. While it is possible to do simple things like index into the resulting object, itís not a real Python dictionary so you canít (for example) call the keys( ) method. There is no straightforward way to convert a Java Map into a Python dictionary, and so I wrote a utility called toPyDictionary( ) and made it a static method of com.bruceeckel.python.PyUtil. This also includes utilities to extract a Python array into a Java List, and a Python dictionary into a Java Map:
Here is the (black-box) unit testing code:
Weíll see the use of the extraction tools in the next section.
There are a number of different ways to extract data from the PythonInterpreter. If you simply call the get( ) method, passing it the object identifier as a string, it returns a PyObject (part of the org.python.core support classes). Itís possible to ďcastĒ it using the __tojava__( ) method, but there are better alternatives:
1. The convenience methods in the Py class, such as py2int( ), take a PyObject and convert it to a number of different types.
2. An overloaded version of get( ) takes the desired Java Class object as a second argument, and produces an object that has that run-time type (so you still need to perform a cast on the result in your Java code).
Using the second approach, getting an array from the PythonInterpreter is quite easy. This is especially useful because Python is exceptionally good at manipulating strings and files, and so you will commonly want to extract the results as an array of strings. For example, you can do a wildcard expansion of file names using Pythonís glob( ), as shown further down in the following code:
The last two examples show the extraction of Python tuples and lists into Java Lists, and Python dictionaries into Java Maps. Both of these cases require more processing than is provided in the standard Jython library, so I have again created utilities in com.bruceeckel.pyton.PyUtil: toList( ) to produce a List from a Python sequence, and toMap( ) to produce a Map from a Python dictionary. The PyUtil methods make it easier to take important data structures back and forth between Java and Python.
Itís also worth noting that you can have multiple PythonInterpreter objects in a program, and each one has its own name space:
When you run the program youíll see that the value of a is distinct within each PythonInterpreter.
Since you have the Java language at your disposal, and you can set and retrieve values in the interpreter, thereís a tremendous amount that you can accomplish with the above approach (controlling Python from Java).† But one of the amazing things about Jython is that it makes Java classes almost transparently available from within Jython. Basically, a Java class looks like a Python class. This is true for standard Java library classes as well as classes that you create yourself, as you can see here:
The ď=MĒ comment is recognized by the makefile generator tool (that I created for this book) as a replacement makefile command. This will be used instead of the commands that the extraction tool would normally place in the makefile.
Note that the import statements map to the Java package structure exactly as you would expect. In the first example, a Date( ) object is created as if it were a native Python class, and printing this object just calls toString( ).
ValGen implements the concept of a ďgeneratorĒ which is used a great deal in the C++ STL (Standard Template Library, part of the Standard C++ Library). A generator is an object that produces a new object every time its ďgeneration methodĒ is called, and it is quite convenient for filling containers. Here, I wanted to use it in a for iteration, and so I needed the generation method to be the one that is called by the iteration process. This is a special method called __getitem__( ), which is actually the overloaded operator for indexing, Ď[ ]í. A for loop calls this method every time it wants to move the iteration forward, and when the elements run out, __getitem__( ) throws an out-of-bounds exception and that signals the end of the for loop (in other languages, you would never use an exception for ordinary control flow, but in Python it seems to work quite well). This exception happens automatically when self.val[i] runs out of elements, so the __getitem__( ) code turns out to be simple. The only complexity is that __getitem__( ) appears to return two objects instead of just one. What Python does is automatically package multiple return values into a tuple, so you still only end up returning a single object (in C++ or Java you would have to create your own data structure to accomplish this). In addition, in the for loop where ValGen is used, Python automatically ďunpacksĒ the tuple so that you can have multiple iterators in the for. These are the kinds of syntax simplifications that make Python so endearing.
The map and set objects are instances of Javaís HashMap and HashSet, again created as if those classes were just native Python components. In the for loop, the put( ) and add( ) methods work just like they do in Java. Also, indexing into a Java Map uses the same notation as for dictionaries, but note that to iterate through the keys in a Map you must use the Map method keySet( ) rather than the Python dictionary method keys( ).
The final part of the example shows the use of a Java class that I created from scratch, to demonstrate how trivial it is. Notice also that Jython intuitively understands JavaBeans properties, since you can either use the getVal( ) and setVal( ) methods, or assign to and read from the equivalent val property. Also, getChars( ) returns a Character in Java, and this becomes an array in Python.
The easiest way to use Java classes that you create for use inside a Python program is to put them inside a package. Although Jython can also import unpackaged java classes (import JavaClass), all such unpackaged java classes will be treated as if they were defined in different packages so they can only see each otherís public methods.
Java packages translate into Python modules, and Python must import a module in order to be able to use the Java class. Here is the Java code for JavaClass:
. Because Python is such a powerful, flexible, dynamic language it is an ideal tool for automated test frameworks, without making any changes to the Java code thatís being tested.You can see that this is just an ordinary Java class, without any awareness that it will be used in a Jython program. For this reason, one of the important uses of Jython is in testing Java code
Inner classes becomes attributes on the class object. Instances of static inner classes can be create by the usual call:
Non-static inner classes must have an outer class instance supplied explicitly as the first argument:
Jython wraps the Java libraries so that any of them can be used directly or via inheritance. In addition, Python shorthand simplifies coding.
As an example, consider the HTMLButton.java example from Chapter 9 of Thinking in Java, 2nd edition (you presumably have already downloaded and installed the source code for that book from www.BruceEckel.com, since a number of examples in this book use libraries from that book). Here is its conversion to Jython:
If you compare the Java version of the program to the above Jython implementation, youíll see that Jython is shorter and generally easier to understand. For example, in the Java version to set up the frame you had to make several calls: the constructor for JFrame( ), the setVisible( ) method and the setDefaultCloseOperation( ) method, whereas in the above code all three of these operations are performed with a single constructor call.
Also notice that the JButton is configured with an actionListener( ) method inside the constructor, with the assignment to kapow. In addition, Jythonís JavaBean awareness means that a call to any method with a name that begins with ďsetĒ can be replaced with an assignment, as you can see above.
The only method that did not come over from Java is the pack( ) method, which seems to be essential in order to force the layout to happen properly. Itís also important that the call to pack( ) appear before the size setting.
You can easily inherit from standard Java library classes in Jython. Hereís the Dialogs.java example from Chapter 13 of Thinking in Java, 2nd edition, converted into Jython:
MyDialog is inherited from JDialog, and you can see named arguments being used in the call to the base-class constructor.
In the creation of the ďOKĒ JButton, note that the actionPerformed method is set right inside the constructor, and that the function is created using the Python lambda keyword. This creates a nameless function with the arguments appearing before the colon and the expression that generates the returned value after the colon. As you should know, the Java prototype for the actionPerformed( ) method only contains a single argument, but the lambda expression indicates two. However, the second argument is provided with a default value, so the function can be called with only one argument. The reason for the second argument is seen in the default value, because this is a way to pass self into the lambda expression, so that it can be used to dispose of the dialog.
Compare this code with the version thatís published in Thinking in Java, 2nd edition. Youíll find that Python language features allow a much more succinct and direct implementation.
Although it does not directly relate to the original problem of this chapter (creating an interpreter), Jython has the additional ability to create Java classes directly from your Jython code. This can produce very useful results, as you are then able to treat the results as if they are native Java classes, albeit with Python power under the hood.
To produce Java classes from Python code, Jython comes with a compiler called jythonc.
The process of creating Python classes that will produce Java classes is a bit more complex than when calling Java classes from Python, because the methods in Java classes are strongly typed, while Python functions and methods are weakly typed. Thus, you must somehow tell jythonc that a Python method is intended to have a particular set of argument types and that its return value is a particular type. You accomplish this with the ď@sigĒ string, which is placed right after the beginning of the Python method definition (this is the standard location for the Python documentation string). For example:
The Python definition doesnít specify any return type, but the @sig string gives the full type information about what is being passed and returned. The jythonc compiler uses this information to generate the correct Java code.
Thereís one other set of rules you must follow in order to get a successful compilation: you must inherit from a Java class or interface in your Python class (you do not need to specify the @sig signature for methods defined in the superclass/interface). If you do not do this, you wonít get your desired methods Ė unfortunately, jythonc gives you no warnings or errors in this case, but you wonít get what you want. If you donít see whatís missing, it can be very frustrating.
In addition, you must import the appropriate java class and give the correct package specification.† In the example below, java is imported so you must inherit from java.lang.Object, but you could also say from java.lang import Object and then youíd just inherit from Object without the package specification. Unfortunately, you donít get any warnings or errors if you get this wrong, so you must be patient and keep trying.
Here is an example of a Python class created to produce a Java class. This also introduces the Ď=Tí directive for the makefile builder tool, which specifies a different target than the one that is normally used by the tool. In this case, the Python file is used to build a Java .class file, so the class file is the desired makefile target. To accomplish this, the default makefile command is replaced using the Ď=Mí directive (notice how you can break across lines using Ď\í):
First note that PythonToJavaClass is inherited from java.lang.Object; if you donít do this you will quietly get a Java class without the right signatures. You are not required to inherit from Object; any other Java class will do.
This class is designed to demonstrate different† arguments and return values, to provide you with enough examples that youíll be able to easily create your own signature strings. The first three of these are fairly self-explanatory, but note the full qualification of the Java name in the signature string.
In returnArray( ), a Python array must be returned as a Java array. To do this, the Jython array( ) function (from the jarray module) must be used, along with the type of the class for the resulting array. Any time you need to return an array to Java, you must use array( ), as seen in the methods ints( ) and doubles( ).
The last methods show how to pass arguments in from Java. Basic types happen automatically as long as you specify them in the @sig string, but you must use objects and you cannot pass in primitives (that is, primitives must be ensconced in wrapper objects, such as Integer).
In argIn3( ), you can see that a Java List is transparently converted to something that behaves just like a Python array, but is not a true array because you cannot take a slice from it. If you want a true Python array, then you must create and pass a PyArray as in argIn4( ), where the slice is successful. Similarly, a Java Map must come in as a PyDictionary in order to be treated as a Python dictionary.
Here is the Java program to exercise the Java classes produced by the above Python code. This also introduces the Ď=Dí directive for the makefile builder tool, which specifies a dependency in addition to those detected by the tool. Here, you canít compile TestPythonToJavaClass.java until PythonToJavaClass.class is available:
For Python support, youíll usually only need to import the classes in org.python.core. Everything else in the above example is fairly straightforward, as PythonToJavaClass appears, from the Java side, to be just another Java class. dumpClassInfo( ) uses reflection to verify that the method signatures specified in PythonToJavaClass.py have come through properly.
Part of the trick of creating Java classes from Python code is the @sig information in the method documentation strings. But thereís a second problem which stems from the fact that Python has no ďpackageĒ keyword Ė the Python equivalent of packages (modules) are implicitly created based on the file name. However, to bring the resulting class files into the Java program, jythonc must be given information about how to create the Java package for the Python code. This is done on the jythonc command line using the --package flag, followed by the package name you wish to produce (including the separation dots, just as you would give the package name using the package keyword in a Java program). This will put the resulting .class files in the appropriate subdirectory off of the current directory. Then you only need to import the package in your Java program, as shown above (youíll need Ď.í in your CLASSPATH in order to run it from the code directory).
Here are the make dependency rules that I used to build the above example (the backslashes at the ends of the lines are understood by make to be line continuations). These rules are encoded into the above Java and Python files using the comment syntax thatís understood by my makefile builder tool:
The first target, TestPythonToJavaClass.class, depends on both TestPythonToJavaClass.java and the PythonToJavaClass.class, which is the Python code thatís converted to a class file. This latter, in turn, depends on the Python source code. Note that itís important that the directory where the target lives be specified, so that the makefile will create the Java program with the minimum necessary amount of rebuilding.
An alternative to Jython is the Java-Python Extension (JPE), which directly connects with your native C-Python implementation.
Jython runs entirely within the JavaVM, which produces two fundamental limitations: Jython cannot be called from CPython, and native Python extensions are not accessible from JPython. JPE is linked with the Python C libraries, so JPE can be called from C-Python, and native Python extensions can be called from Java through JPE.
If you need to access the features on your native platform, JPE might be the easiest solution. You can find JPE at http://www.arakne.com/jpe.htm.
This chapter has arguably gone much deeper into Jython than required to use the interpreter design pattern. Indeed, once you decide that you need to use interpreter and that youíre not going to get lost inventing your own language, the solution of installing Jython is quite simple, and you can at least get started by following the GreenHouseController example.
Of course, that example is often too simple and you may need something more sophisticated, often requiring more interesting data to be passed back and forth. When I encountered the limited documentation, I felt it necessary to come up with a more thorough examination of Jython.
In the process, note that there could be another equally powerful design pattern lurking in here, which could perhaps be called multiple languages. This is based on the experience of having each language solve a certain class of problems better than the other; by combining languages you can solve problems much faster than with either language by itself. CORBA is another way to bridge across languages, and at the same time bridging between computers and operating systems.
To me, Python and Java present a very potent combination for program development because of Javaís architecture and tool set, and Pythonís extremely rapid development (generally considered to be 5-10 times faster than C++ or Java). Python is usually slower, however, but even if you end up re-coding parts of your program for speed, the initial fast development will allow you to more quickly flesh out the system and uncover and solve the critical sections. And often, the execution speed of Python is not a problem Ė in those cases itís an even bigger win. A number of commercial products already use Java and Jython, and because of the terrific productivity leverage I expect to see this happen more in the future.
1. Modify GreenHouseLanguage.py so that it checks the times for the events and runs those events at the appropriate times.
2. Modify GreenHouseLanguage.py so that it calls a function for action instead of just printing a string.
3. Create a Swing application with a JTextField (where the user will enter commands) and a JTextArea (where the command results will be displayed). Connect to a PythonInterpreter object so that the output will be sent to the JTextArea (which should scroll). Youíll need to locate the PythonInterpreter command that redirects the output to a Java stream.
4. Modify GreenHouseLanguage.py to add a master controller class (instead of the static array inside Event) and provide a run( ) method for each of the subclasses. Each run( ) should create and use an object from the standard Java library during its execution. Modify GreenHouseController.java to use this new class.
5. Modify the resulting GreenHouseLanguage.py from exercise two to produce Java classes (add the @sig documentation strings to produce the correct Java signatures, and create a makefile to build the Java .class files). Write a Java program that uses these classes.
While State has a way to allow the client programmer to change the implementation, StateMachine imposes a structure to automatically change the implementation from one object to the next. The current implementation represents the state that a system is in, and the system behaves differently from one state to the next (because it uses State). Basically, this is a ďstate machineĒ using objects.
The code that moves the system from one state to the next is often a Template Method, as seen in the following framework for a basic state machine. We start by defining a tagging interface for input objects:
Each state can be run( ) to perform its behavior, and (in this design) you can also pass it an Input object so it can tell you what new state to move to based on that Input. The key distinction between this design and the next is that here, each State object decides what other states it can move to, based on the Input, whereas in the subsequent design all of the state transitions are held in a single table. Another way to put it is that here, each State object has its own little State table, and in the subsequent design there is a single master state transition table for the whole system.
The StateMachine keeps track of the current state, which is initialized by the constructor. The runAll( ) method takes an Iterator to a list of Input objects (an Iterator is used here for convenience and simplicity; the important issue is that the input information comes from somewhere). This method not only moves to the next state, but it also calls run( ) for each state object Ė thus you can see itís an expansion of the idea of the State pattern, since run( ) does something different depending on the state that the system is in.
Iíve also treated runAll( ) as a template method. This is typical, but certainly not required Ė you could concievably want to override it, but typically the behavior change will occur in Stateís run( ) instead.
At this point the basic framework for this style of StateMachine (where each state decides the next states) is complete. As an example, Iíll use a fancy mousetrap that can move through several states in the process of trapping a mouse. The mouse classes and information are stored in the mouse package, including a class representing all the possible moves that a mouse can make, which will be the inputs to the state machine:
Youíll note that hashCode( ) and equals( ) have been overriden so that MouseAction objects can be used as keys in a HashMap, but in the first version of the mousetrap we wonít do this. Also, each possible move by a mouse is enumerated as a static MouseAction object.
For creating test code, a sequence of mouse inputs is provided from a text file:
To read this file in a generic fashion, here is a general-purpose tool called StringList:
This StringList only holds Objects, just as an ArrayList does, so we need an adapter to turn the Strings into MouseActions:
The MouseMoveList looks a bit like a decorator, and acts a bit like an adapter. However, an adapter changes one interface to another, and a decorator adds functionality or data. MouseMoveList changes the contents of the container, so it might be thought of as a Transformer.
With these tools in place, itís now possible to create the first version of the mousetrap program. Each State subclass defines itís run( ) behavior, and also establishes its next state with an if-else clause:
The StateMachine class simply defines all the possible states as static objects, and also sets up the initial state. The UnitTest creates a MouseTrap and then tests it with all the inputs from a MouseMoveList.
While theuse of if-else statements inside the next( ) methods is perfectly reasonable, managing a large number of these could become difficult. Another approach is to create tables inside each State object defining the various next states based on the input.
Initially, this seems like it ought to be quite simple. You should be able to define a static table in each State subclass that defines the transitions in terms of the other State objects. However, it turns out that this approach generates cyclic initialization dependencies. To solve the problem, Iíve had to delay the initialization of the tables until the first time that the next( ) method is called for a particular State object. Initially, the next( ) methods can appear a little strange because of this.
The StateT class is an implementation of State (so that the same StateMachine class can be used from the previous example) that adds a Map and a method to initialize the map from a two-dimensional array. The next( ) method has a base-class implementation which must be called from the overridden derived class next( ) methods after they test for a null Map (and initialize it if itís null):
The rest of the code is identical Ė the difference is in the next( ) methods and the StateT class.
If you have to create and maintain a lot of State classes, this approach is an improvement, since itís easier to quickly read and understand the state transitions from looking at the table.
1. Modify MouseTrap2Test.java so that the state table information is loaded from an external text file, which only contains the state table information.
The advantage of the previous design is that all the information about a state, including the state transition information, is located within the state class itself. This is generally a good design principle.
However, in a pure state machine, the machine can be completely represented by a single state-transition table. This has the advantage of locating all the information about the state machine in a single place, which means that you can more easily create and maintain the table based on a classic state-transition diagram.
The classic state-transition diagram uses a circle to represent each state, and lines from the state pointing to all states that state can transition into. Each transition line is annotated with conditions for transition and an action during transition. Hereís what it looks like:
(Simple State Machine Diagram)
∑ Direct translation of state diagram
∑ Vector of change: the state diagram representation
∑ Reasonable implementation
∑ No excess of states (you could represent every single change with a new state)
∑ Simplicity and flexibility
∑ States are trivial Ė no information or functions/data, just an identity
∑ Not like the State pattern!
∑ The machine governs the move from state to state
∑ Similar to flyweight
∑ Each state may move to many others
∑ Condition & action functions must also be external to states
∑ Centralize description in a single table containing all variations, for ease of configuration
∑ State Machine & Table-Driven Code
∑ Implements a vending machine
∑ Uses several other patterns
∑ Separates common state-machine code from specific application (like template method)
∑ Each input causes a seek for appropriate solution (like chain of responsibility)
∑ Tests and transitions are encapsulated in function objects (objects that hold functions)
∑ Java constraint: methods are not first-class objects
The State class is distinctly different from before, since it is really just a placeholder with a name. Thus it is not inherited from previous State classes:
In the state transition diagram, an input is tested to see if it meets the condition necessary to transfer to the state under question. As before, the Input is just a tagging interface:
The Condition evaluates the Input to decide whether this row in the table is the correct transition:
If the Condition returns true, then the transition to a new state is made, and as that transition is made some kind of action occurs (in the previous state machine design, this was the run( ) method):
With these classes in place, we can set up a 3-dimensional table whereeach row completely describes a state. The first element in the row is the current state, and the rest of the elements are each a row indicating what the type of the input can be, the condition that must be satisfied in order for this state change to be the correct one, the action that happens during transition, and the new state to move into. Note that the Input object is not just used for its type, it is also a Messenger object that carries information to the Condition and Transition objects:
See ListPerformance.java example in TIJ from Chapter 9
1. Apply TransitionTable.java to the ďWasherĒ problem.
2. Create a StateMachine system whereby the current state along with input information determines the next state that the system will be in. To do this, each state must store a reference back to the proxy object (the state controller) so that it can request the state change. Use a HashMap to create a table of states, where the key is a String naming the new state and the value is the new state object. Inside each state subclass override a method nextState( ) that has its own state-transition table. The input to nextState( ) should be a single word that comes from a text file containing one word per line.
3. Modify the previous exercise so that the state machine can be configured by creating/modifying a single multi-dimensional array.
4. Modify the ďmoodĒ exercise from the previous session so that it becomes a state machine using StateMachine.java
5. Create an elevator state machine system using StateMachine.java
6. Create a heating/air-conditioning system using StateMachine.java
7. A generator is an object that produces other objects, just like a factory, except that the generator function doesnít require any arguments. Create a MouseMoveGenerator which produces correct MouseMove actions as outputs each time the generator function is called (that is, the mouse must move in the proper sequence, thus the possible moves are based on the previous move Ė itís another state machine). Add a method iterator( ) to produce an iterator, but this method should take an int argument that specifies the number of moves to produce before hasNext( ) returns false.
This chapter will look at the process of solving a problem by applying design patterns in an evolutionary fashion. That is, a first cut design will be used for the initial solution, and then this solution will be examined and various design patterns will be applied to the problem (some of which will work, and some of which wonít). The key question that will always be asked in seeking improved solutions is ďwhat will change?Ē
 (although he tends to talk about pieces of code more than pattern-level designs). You start with a solution, and then when you discover that it doesnít continue to meet your needs, you fix it. Of course, this is a natural tendency but in computer programming itís been extremely difficult to accomplish with procedural programs, and the acceptance of the idea that we can refactor code and designs adds to the body of proof that object-oriented programming is ďa good thing.ĒThis process is similar to what Martin Fowler talks about in his book Refactoring: Improving the Design of Existing Code
The nature of this problem is that the trash is thrown unclassified into a single bin, so the specific type information is lost. But later, the specific type information must be recovered to properly sort the trash. In the initial solution, RTTI (described in Chapter 12 of Thinking in Java, 2nd edition) is used.
This is not a trivial design because it has an added constraint. Thatís what makes it interestingóitís more like the messy problems youíre likely to encounter in your work. The extra constraint is that the trash arrives at the trash recycling plant all mixed together. The program must model the sorting of that trash. This is where RTTI comes in: you have a bunch of anonymous pieces of trash, and the program figures out exactly what type they are.
In the source code listings available for this book, this file will be placed in the subdirectory recyclea that branches off from the subdirectory refactor. The unpacking tool takes care of placing it into the correct subdirectory. The reason for doing this is that this chapter rewrites this particular example a number of times and by putting each version in its own directory (using the default package in each directory so that invoking the program is easy), the class names will not clash.
Several ArrayList objects are created to hold Trash references. Of course, ArrayLists actually hold Objects so theyíll hold anything at all. The reason they hold Trash (or something derived from Trash) is only because youíve been careful to not put in anything except Trash. If you do put something ďwrongĒ into the ArrayList, you wonít get any compile-time warnings or errorsóyouíll find out only via an exception at run time.
When the Trash references are added, they lose their specific identities and become simply Object references (they are upcast). However, because of polymorphism the proper behavior still occurs when the dynamically-bound methods†are called through the Iterator sorter, once the resulting Object has been cast back to Trash. sumValue( ) also takes an Iterator to perform operations on every object in the ArrayList.
It looks silly to upcast the types of Trash into a container holding base type references, and then turn around and downcast. Why not just put the trash into the appropriate receptacle in the first place? (Indeed, this is the whole enigma of recycling). In this program it would be easy to repair, but sometimes a systemís structure and flexibility can benefit greatly from downcasting.
The program satisfies the design requirements: it works. This might be fine as long as itís a one-shot solution. However, a useful program tends to evolve over time, so you must ask, ďWhat if the situation changes?Ē For example, cardboard is now a valuable recyclable commodity, so how will that be integrated into the system (especially if the program is large and complicated). Since the above type-check coding in the switch statement could be scattered throughout the program, you must go find all that code every time a new type is added, and if you miss one the compiler wonít give you any help by pointing out an error.
The key to the misuse of RTTI here is that every type is tested. If youíre looking for only a subset of types because that subset needs special treatment, thatís probably fine. But if youíre hunting for every type inside a switch statement, then youíre probably missing an important point, and definitely making your code less maintainable. In the next section weíll look at how this program evolved over several stages to become much more flexible. This should prove a valuable example in program design.
The solutions in Design Patterns are organized around the question ďWhat will change as this program evolves?Ē This is usually the most important question that you can ask about any design. If you can build your system around the answer, the results will be two-pronged: not only will your system allow easy (and inexpensive) maintenance, but you might also produce components that are reusable, so that other systems can be built more cheaply. This is the promise of object-oriented programming, but it doesnít happen automatically; it requires thought and insight on your part. In this section weíll see how this process can happen during the refinement of a system.
The answer to the question ďWhat will change?Ē for the recycling system is a common one: more types will be added to the system. The goal of the design, then, is to make this addition of types as painless as possible. In the recycling program, weíd like to encapsulate all places where specific type information is mentioned, so (if for no other reason) any changes can be localized to those encapsulations. It turns out that this process also cleans up the rest of the code considerably.
This brings up a general object-oriented design principle that I first heard spoken by Grady Booch: ďIf the design is too complicated, make more objects.Ē This is simultaneously counterintuitive and ludicrously simple, and yet itís the most useful guideline Iíve found. (You might observe that ďmaking more objectsĒ is often equivalent to ďadd another level of indirection.Ē) In general, if you find a place with messy code, consider what sort of class would clean that up. Often the side effect of cleaning up the code will be a system that has better structure and is more flexible.
Consider first the place where Trash objects are created, which is a switch statement inside main( ):
This is definitely messy, and also a place where you must change code whenever a new type is added. If new types are commonly added, a better solution is a single method that takes all of the necessary information and produces a reference to an object of the correct type, already upcast to a trash object. In Design Patterns this is broadly referred to as a creational pattern (of which there are several). The specific pattern that will be applied here is a variant of the Factory Method. Here, the factory method is a static member of Trash, but more commonly it is a method that is overridden in the derived class.
The idea of the factory method is that you pass it the essential information it needs to know to create your object, then stand back and wait for the reference (already upcast to the base type) to pop out as the return value. From then on, you treat the object polymorphically. Thus, you never even need to know the exact type of object thatís created. In fact, the factory method hides it from you to prevent accidental misuse. If you want to use the object without polymorphism, you must explicitly use RTTI and casting.
But thereís a little problem, especially when you use the more complicated approach (not shown here) of making the factory method in the base class and overriding it in the derived classes. What if the information required in the derived class requires more or different arguments? ďCreating more objectsĒ solves this problem. To implement the factory method, the Trash class gets a new method called factory. To hide the creational data, thereís a new class called Messenger that carries all of the necessary information for the factory method to create the appropriate Trash object (weíve started referring to Messenger as a design pattern, but itís simple enough that you may not choose to elevate it to that status). Hereís a simple implementation of Messenger:
A Messenger objectís only job is to hold information for the factory( ) method. Now, if thereís a situation in which factory( ) needs more or different information to create a new type of Trash object, the factory( ) interface doesnít need to be changed. The Messenger class can be changed by adding new data and new constructors, or in the more typical object-oriented fashion of subclassing.
The factory( ) method for this simple example looks like this:
Here, the determination of the exact type of object is simple, but you can imagine a more complicated system in which factory( ) uses an elaborate algorithm. The point is that itís now hidden away in one place, and you know to come to this place when you add new types.
The creation of new objects is now much simpler in main( ):
A Messenger object is created to pass the data into factory( ), which in turn produces some kind of Trash object on the heap and returns the reference thatís added to the ArrayList bin. Of course, if you change the quantity and type of argument, this statement will still need to be modified, but that can be eliminated if the creation of the Messenger object is automated. For example, an ArrayList of arguments can be passed into the constructor of a Messenger object (or directly into a factory( ) call, for that matter). This requires that the arguments be parsed and checked at run time, but it does provide the greatest flexibility.
You can see from this code what ďvector of changeĒ problem the factory is responsible for solving: if you add new types to the system (the change), the only code that must be modified is within the factory, so the factory isolates the effect of that change.
A problem with the design above is that it still requires a central location where all the types of the objects must be known: inside the factory( ) method. If new types are regularly being added to the system, the factory( ) method must be changed for each new type. When you discover something like this, it is useful to try to go one step further and move all of the information about the typeóincluding its creationóinto the class representing that type. This way, the only thing you need to do to add a new type to the system is to inherit a single class.
To move the information concerning type creation into each specific type of Trash, the ďprototypeĒ pattern (from the Design Patterns book) will be used. The general idea is that you have a master sequence of objects, one of each type youíre interested in making. The objects in this sequence are used only for making new objects, using an operation thatís not unlike the clone( ) scheme built into Javaís root class Object. In this case, weíll name the cloning method tClone( ). When youíre ready to make a new object, presumably you have some sort of information that establishes the type of object you want to create, then you move through the master sequence comparing your information with whatever appropriate information is in the prototype objects in the master sequence. When you find one that matches your needs, you clone it.
In this scheme there is no hard-coded information for creation. Each object knows how to expose appropriate information and how to clone itself. Thus, the factory( ) method doesnít need to be changed when a new type is added to the system.
One approach to the problem of prototyping is to add a number of methods to support the creation of new objects. However, in Java 1.1 thereís already support for creating new objects if you have a reference to the Class object. With Java 1.1 reflection (introduced in Chapter 12 of Thinking in Java, 2nd edition) you can call a constructor even if you have only a reference to the Class object. This is the perfect solution for the prototyping problem.
The list of prototypes will be represented indirectly by a list of references to all the Class objects you want to create. In addition, if the prototyping fails, the factory( ) method will assume that itís because a particular Class object wasnít in the list, and it will attempt to load it. By loading the prototypes dynamically like this, the Trash class doesnít need to know what types it is working with, so it doesnít need any modifications when you add new types. This allows it to be easily reused throughout the rest of the chapter.
The basic Trash class and sumValue( ) remain as before. The rest of the class supports the prototyping pattern. You first see two inner classes (which are made static, so they are inner classes only for code organization purposes) describing exceptions that can occur. This is followed by an ArrayList called trashTypes, which is used to hold the Class references.
In Trash.factory( ), the String inside the Messenger object id (a different version of the Messenger class than that of the prior discussion) contains the type name of the Trash to be created; this String is compared to the Class names in the list. If thereís a match, then thatís the object to create. Of course, there are many ways to determine what object you want to make. This one is used so that information read in from a file can be turned into objects.
Once youíve discovered which kind of Trash to create, then the reflection methods come into play. The getConstructor( ) method takes an argument thatís an array of Class references. This array represents the arguments, in their proper order, for the constructor that youíre looking for. Here, the array is dynamically created using the Java 1.1†array-creation syntax:
This code assumes that every Trash type has a constructor that takes a double (and notice that double.class is distinct from Double.class). Itís also possible, for a more flexible solution, to call getConstructors( ), which returns an array of the possible constructors.
What comes back from getConstructor( ) is a reference to a Constructor object (part of java.lang.reflect). You call the constructor dynamically with the method newInstance( ), which takes an array of Object containing the actual arguments. This array is again created using the Java 1.1†syntax:
In this case, however, the double must be placed inside a wrapper class so that it can be part of this array of objects. The process of calling newInstance( ) extracts the double, but you can see it is a bit confusingóan argument might be a double or a Double, but when you make the call you must always pass in a Double. Fortunately, this issue exists only for the primitive types.
Once you understand how to do it, the process of creating a new object given only a Class reference is remarkably simple. Reflection also allows you to call methods in this same dynamic fashion.
Of course, the appropriate Class reference might not be in the trashTypes list. In this case, the return in the inner loop is never executed and youíll drop out at the end. Here, the program tries to rectify the situation by loading the Class object dynamically and adding it to the trashTypes list. If it still canít be found something is really wrong, but if the load is successful then the factory method is called recursively to try again.
As youíll see, the beauty of this design is that this code doesnít need to be changed, regardless of the different situations it will be used in (assuming that all Trash subclasses contain a constructor that takes a single double argument).
To fit into the prototyping scheme, the only thing thatís required of each new subclass of Trash is that it contain a constructor that takes a double argument. Java reflection handles everything else.
Here are the different types of Trash, each in their own file but part of the Trash package (again, to facilitate reuse within the chapter):
And hereís a new type of Trash:
You can see that, other than the constructor, thereís nothing special about any of these classes.
The information about Trash objects will be read from an outside file. The file has all of the necessary information about each piece of trash on a single line in the form Trash:weight, such as:
Note that the class path must be included when giving the class names, otherwise the class will not be found.
This file is read using the previously-defined StringList tool, and each line is picked aparat usingthe String method indexOf( ) to produce the index of the Ď:í. This is first used with the String method substring( ) to extract the name of the trash type, and next to get the weight that is turned into a double with the static Double.valueOf( ) method. The trim( ) method removes white space at both ends of a string.
The Trash parser is placed in a separate file since it will be reused throughout this chapter:
In RecycleA.java, an ArrayList was used to hold the Trash objects. However, other types of containers can be used as well. To allow for this, the first version of fillBin( ) takes a reference to a Fillable, which is simply an interface that supports a method called addTrash( ):
Anything that supports this interface can be used with fillBin. Of course, Collection doesnít implement Fillable, so it wonít work. Since Collection is used in most of the examples, it makes sense to add a second overloaded fillBin( ) method that takes a Collection. Any Collection can then be used as a Fillable object using an adapter class:
You can see that the only job of this class is to connect Fillableís addTrash( ) method to Collectionís add( ). With this class in hand, the overloaded fillBin( ) method can be used with a Collection in ParseTrash.java:
This approach works for any container class thatís used frequently. Alternatively, the container class can provide its own adapter that implements Fillable. (Youíll see this later, in DynaTrash.java.)
Now you can see the revised version of RecycleA.java using the prototyping technique:
All of the Trash objects, as well as the ParseTrash and support classes, are now part of the package refactor.trash, so they are simply imported.
The process of opening the data file containing Trash descriptions and the parsing of that file have been wrapped into the static method ParseTrash.fillBin( ), so now itís no longer a part of our design focus. You will see that throughout the rest of the chapter, no matter what new classes are added, ParseTrash.fillBin( ) will continue to work without change, which indicates a good design.
In terms of object creation, this design does indeed severely localize the changes you need to make to add a new type to the system. However, thereís a significant problem in the use of RTTI that shows up clearly here. The program seems to run fine, and yet it never detects any cardboard, even though there is cardboard in the list! This happens because of the use of RTTI, which looks for only the types that you tell it to look for. The clue that RTTI is being misused is that every type in the system is being tested, rather than a single type or subset of types. As you will see later, there are ways to use polymorphism instead when youíre testing for every type. But if you use RTTI a lot in this fashion, and you add a new type to your system, you can easily forget to make the necessary changes in your program and produce a difficult-to-find bug. So itís worth trying to eliminate RTTI in this case, not just for aesthetic reasonsóit produces more maintainable code.
With creation out of the way, itís time to tackle the remainder of the design: where the classes are used. Since itís the act of sorting into bins thatís particularly ugly and exposed, why not take that process and hide it inside a class? This is the principle of ďIf you must do something ugly, at least localize the ugliness inside a class.Ē It looks like this:
The TrashSorter object initialization must now be changed whenever a new type of Trash is added to the model. You could imagine that the TrashSorter class might look something like this:
That is, TrashSorter is an ArrayList of references to ArrayLists of Trash references, and with add( ) you can install another one, like so:
Now, however, sort( ) becomes a problem. How does the statically-coded method deal with the fact that a new type has been added? To solve this, the type information must be removed from sort( ) so that all it needs to do is call a generic method that takes care of the details of type. This, of course, is another way to describe a dynamically-bound method. So sort( ) will simply move through the sequence and call a dynamically-bound method for each ArrayList. Since the job of this method is to grab the pieces of trash it is interested in, itís called grab(Trash). The structure now looks like:
TrashSorter needs to call each grab( ) method and get a different result depending on what type of Trash the current ArrayList is holding. That is, each ArrayList must be aware of the type it holds. The classic approach to this problem is to create a base ďTrash binĒ class and inherit a new derived class for each different type you want to hold. If Java had a parameterized type mechanism that would probably be the most straightforward approach. But rather than hand-coding all the classes that such a mechanism should be building for us, further observation can produce a better approach.
A basic OOP design principle is ďUse data members for variation in state, use polymorphism for variation in behavior.Ē Your first thought might be that the grab( ) method certainly behaves differently for an ArrayList that holds Paper than for one that holds Glass. But what it does is strictly dependent on the type, and nothing else. This could be interpreted as a different state, and since Java has a class to represent type (Class) this can be used to determine the type of Trash a particular Tbin will hold.
The constructor for this Tbin requires that you hand it the Class of your choice. This tells the ArrayList what type it is supposed to hold. Then the grab( ) method uses Class BinType and RTTI to see if the Trash object youíve handed it matches the type itís supposed to grab.
Here is the new version of the program:
Tbin contains a Class reference type which establishes in the constructor what what type it should grab. The grab() method checks this type against the object you pass it. Note that in this design, grab() only accepts Trash objects so you get compile-time type checking on the base type, but you could also just accept Object and it would still work.
TbinList holds a set of Tbin references, so that sort( ) can iterate through the Tbins when itís looking for a match for the Trash object youíve handed it. If it doesnít find a match, it creates a new Tbin for the type that hasnít been found, and makes a recursive call to itself Ė the next time around, the new bin will be found.
Notice the genericity of this code: it doesnít change at all if new types are added. If the bulk of your code doesnít need changing when a new type is added (or some other change occurs) then you have an easily extensible system.
The above design is certainly satisfactory. Adding new types to the system consists of adding or modifying distinct classes without causing code changes to be propagated throughout the system. In addition, RTTI is not ďmisusedĒ as it was in RecycleA.java. However, itís possible to go one step further and take a purist viewpoint about RTTI and say that it should be eliminated altogether from the operation of sorting the trash into bins.
To accomplish this, you must first take the perspective that all type-dependent activitiesósuch as detecting the type of a piece of trash and putting it into the appropriate binóshould be controlled through polymorphism and dynamic binding.
The previous examples first sorted by type, then acted on sequences of elements that were all of a particular type. But whenever you find yourself picking out particular types, stop and think. The whole idea of polymorphism (dynamically-bound method calls) is to handle type-specific information for you. So why are you hunting for types?
The answer is something you probably donít think about: Java performs only single dispatching. That is, if you are performing an operation on more than one object whose type is unknown, Java will invoke the dynamic binding mechanism on only one of those types. This doesnít solve the problem, so you end up detecting some types manually and effectively producing your own dynamic binding behavior.
The solution is called multiple dispatching, which means setting up a configuration such that a single method call produces more than one dynamic method call and thus determines more than one type in the process. To get this effect, you need to work with more than one type hierarchy: youíll need a type hierarchy for each dispatch. The following example works with two hierarchies: the existing Trash family and a hierarchy of the types of trash bins that the trash will be placed into. This second hierarchy isnít always obvious and in this case it needed to be created in order to produce multiple dispatching (in this case there will be only two dispatches, which is referred to as double dispatching).
Remember that polymorphism can occur only via method calls, so if you want double dispatching to occur, there must be two method calls: one used to determine the type within each hierarchy. In the Trash hierarchy there will be a new method called addToBin( ), which takes an argument of an array of TypedBin. It uses this array to step through and try to add itself to the appropriate bin, and this is where you'll see the double dispatch.
The new hierarchy is TypedBin, and it contains its own method called add( ) that is also used polymorphically. But here's an additional twist: add( ) is overloaded to take arguments of the different types of trash. So an essential part of the double dispatching scheme also involves overloading.
Redesigning the program produces a dilemma: itís now necessary for the base class Trash to contain an addToBin( ) method. One approach is to copy all of the code and change the base class. Another approach, which you can take when you donít have control of the source code, is to put the addToBin( ) method into an interface, leave Trash alone, and inherit new specific types of Aluminum, Paper, Glass, and Cardboard. This is the approach that will be taken here.
Most of the classes in this design must be public, so they are placed in their own files. Hereís the interface:
In each particular subtype of Aluminum, Paper, Glass, and Cardboard, the addToBin( ) method in the interface TypedBinMember is implemented, but it looks like the code is exactly the same in each case:
The code in each addToBin( ) calls add( ) for each TypedBin object in the array. But notice the argument: this. The type of this is different for each subclass of Trash, so the code is different. (Although this code will benefit if a parameterized type mechanism is ever added to Java.) So this is the first part of the double dispatch, because once youíre inside this method you know youíre Aluminum, or Paper, etc. During the call to add( ), this information is passed via the type of this. The compiler resolves the call to the proper overloaded version of add( ). But since tb[i] produces a reference to the base type TypedBin, this call will end up calling a different method depending on the type of TypedBin thatís currently selected. That is the second dispatch.
Hereís the base class for TypedBin:
You can see that the overloaded add( ) methods all return false. If the method is not overloaded in a derived class, it will continue to return false, and the caller (addToBin( ), in this case) will assume that the current Trash object has not been added successfully to a container, and continue searching for the right container.
In each of the subclasses of TypedBin, only one overloaded method is overridden, according to the type of bin thatís being created. For example, CardboardBin overrides add(DDCardboard). The overridden method adds the trash object to its container and returns true, while all the rest of the add( ) methods in CardboardBin continue to return false, since they havenít been overridden. This is another case in which a parameterized type mechanism in Java would allow automatic generation of code. (With C++ templates, you wouldnít have to explicitly write the subclasses or place the addToBin( ) method in Trash.)
Since for this example the trash types have been customized and placed in a different directory, youíll need a different trash data file to make it work. Hereís a possible DDTrash.dat:
Hereís the rest of the program:
TrashBinSet encapsulates all of the different types of TypedBins, along with the sortIntoBins( ) method, which is where all the double dispatching takes place. You can see that once the structure is set up, sorting into the various TypedBins is remarkably easy. In addition, the efficiency of two dynamic method calls is probably better than any other way you could sort.
Notice the ease of use of this system in main( ), as well as the complete independence of any specific type information within main( ). All other methods that talk only to the Trash base-class interface will be equally invulnerable to changes in Trash types.
The changes necessary to add a new type are relatively isolated: you modify TypedBin, inherit the new type of Trash with its addToBin( ) method, then inherit a new TypedBin (this is really just a copy and simple edit), and finally add a new type into the aggregate initialization for TrashBinSet.
Now consider applying a design pattern that has an entirely different goal to the trash sorting problem.
For this pattern, we are no longer concerned with optimizing the addition of new types of Trash to the system. Indeed, this pattern makes adding a new type of Trash more complicated. The assumption is that you have a primary class hierarchy that is fixed; perhaps itís from another vendor and you canít make changes to that hierarchy. However, youíd like to add new polymorphic methods to that hierarchy, which means that normally youíd have to add something to the base class interface. So the dilemma is that you need to add methods to the base class, but you canít touch the base class. How do you get around this?
The design pattern that solves this kind of problem is called a ďvisitorĒ (the final one in the Design Patterns book), and it builds on the double dispatching scheme shown in the last section.
The visitor pattern allows you to extend the interface of the primary type by creating a separate class hierarchy of type Visitor to virtualize the operations performed upon the primary type. The objects of the primary type simply ďacceptĒ the visitor, then call the visitorís dynamically-bound method. It looks like this:
Now, if v is a Visitable reference to an Aluminum object, the code:
uses double dispatching to cause two polymorphic method calls: the first one to select Aluminumís version of accept( ), and the second one within accept( ) when the specific version of visit( ) is called dynamically using the base-class Visitor reference v.
This configuration means that new functionality can be added to the system in the form of new subclasses of Visitor. The Trash hierarchy doesnít need to be touched. This is the prime benefit of the visitor pattern: you can add new polymorphic functionality to a class hierarchy without touching that hierarchy (once the accept( ) methods have been installed). Note that the benefit is helpful here but not exactly what we started out to accomplish, so at first blush you might decide that this isnít the desired solution.
But look at one thing thatís been accomplished: the visitor solution avoids sorting from the master Trash sequence into individual typed sequences. Thus, you can leave everything in the single master sequence and simply pass through that sequence using the appropriate visitor to accomplish the goal. Although this behavior seems to be a side effect of visitor, it does give us what we want (avoiding RTTI).
The double dispatching in the visitor pattern takes care of determining both the type of Trash and the type of Visitor. In the following example, there are two implementations of Visitor: PriceVisitor to both determine and sum the price, and WeightVisitor to keep track of the weights.
You can see all of this implemented in the new, improved version of the recycling program.
As with DoubleDispatch.java, the Trash class is left alone and a new interface is created to add the accept( ) method:
Since thereís nothing concrete in the Visitor base class, it can be created as an interface:
At this point, you could †follow the same approach that was used for double dispatching and create new subtypes of Aluminum, Paper, Glass, and Cardboard that implement the accept( ) method. For example, the new Visitable Aluminum would look like this:
However, we seem to be encountering an ďexplosion of interfaces:Ē basic Trash, special versions for double dispatching, and now more special versions for visitor. Of course, this ďexplosion of interfacesĒ is arbitraryóone could simply put the additional methods in the Trash class. If we ignore that we can instead see an opportunity to use the Decorator pattern: it seems like it should be possible to create a Decorator that can be wrapped around an ordinary Trash object and will produce the same interface as Trash and add the extra accept( ) method. In fact, itís a perfect example of the value of Decorator.
:The double dispatch creates a problem, however. Since it relies on overloading of both accept( ) and visit( ), it would seem to require specialized code for each different version of the accept( ) method. With C++ templates, this would be fairly easy to accomplish (since templates automatically generate type-specialized code) but Java has no such mechanismóat least it does not appear to. However, reflection allows you to determine type information at run time, and it turns out to solve many problems that would seem to require templates (albeit not as simply). Hereís the decorator that does the trick
[[ Description of Reflection use† ]]
[[Note that a Dynamic Proxy might also be applied here.]]
The only other tool we need is a new type of Fillable adapter that automatically decorates the objects as they are being created from the original Trash.dat file. But this might as well be a decorator itself, decorating any kind of Fillable:
Now you can wrap it around any kind of existing Fillable, or any new ones that havenít yet been created.
The rest of the program creates specific Visitor types and sends them through a single list of Trash objects:
In Test( ), note how visitability is added by simply creating a different kind of bin using the decorator. Also notice that the FillableCollection adapter has the appearance of being used as a decorator (for ArrayList) in this situation. However, it completely changes the interface of the ArrayList, whereas the definition of Decorator is that the interface of the decorated class must still be there after decoration.
Note that the shape of the client code (shown in the Test class) has changed again, from the original approaches to the problem. Now thereís only a single Trash bin. The two Visitor objects are accepted into every element in the sequence, and they perform their operations. The visitors keep their own internal data to tally the total weights and prices.
Finally, thereís no run time type identification other than the inevitable cast to Trash when pulling things out of the sequence. This, too, could be eliminated with the implementation of parameterized types in Java.
One way you can distinguish this solution from the double dispatching solution described previously is to note that, in the double dispatching solution, only one of the overloaded methods, add( ), was overridden when each subclass was created, while here each one of the overloaded visit( ) methods is overridden in every subclass of Visitor.
Thereís a lot more code here, and thereís definite coupling between the Trash hierarchy and the Visitor hierarchy. However, thereís also high cohesion within the respective sets of classes: they each do only one thing (Trash describes Trash, while Visitor describes actions performed on Trash), which is an indicator of a good design. Of course, in this case it works well only if youíre adding new Visitors, but it gets in the way when you add new types of Trash.
Low coupling between classes and high cohesion within a class is definitely an important design goal. Applied mindlessly, though, it can prevent you from achieving a more elegant design. It seems that some classes inevitably have a certain intimacy with each other. These often occur in pairs that could perhaps be called couplets; for example, containers and iterators. The Trash-Visitor pair above appears to be another such couplet.
Various designs in this chapter attempt to remove RTTI, which might give you the impression that itís ďconsidered harmfulĒ (the condemnation used for poor, ill-fated goto, which was thus never put into Java). This isnít true; it is the misuse of RTTI that is the problem. The reason our designs removed RTTI is because the misapplication of that feature prevented extensibility, while the stated goal was to be able to add a new type to the system with as little impact on surrounding code as possible. Since RTTI is often misused by having it look for every single type in your system, it causes code to be non-extensible: when you add a new type, you have to go hunting for all the code in which RTTI is used, and if you miss any you wonít get help from the compiler.
However, RTTI doesnít automatically create non-extensible code. Letís revisit the trash recycler once more. This time, a new tool will be introduced, which I call a TypeMap. It contains a HashMap that holds ArrayLists, but the interface is simple: you can add( ) a new object, and you can get( ) an ArrayList containing all the objects of a particular type. The keys for the contained HashMap are the types in the associated ArrayList. The beauty of this design is that the TypeMap dynamically adds a new pair whenever it encounters a new type, so whenever you add a new type to the system (even if you add the new type at run time), it adapts.
Our example will again build on the structure of the Trash types in package refactor.Trash (and the Trash.dat file used there can be used here without change):
Although powerful, the definition for TypeMap is simple. It contains a HashMap, and the add( ) method does most of the work. When you add( ) a new object, the reference for the Class object for that type is extracted. This is used as a key to determine whether an ArrayList holding objects of that type is already present in the HashMap. If so, that ArrayList is extracted and the object is added to the ArrayList. If not, the Class object and a new ArrayList are added as a key-value pair.
You can get an Iterator of all the Class objects from keys( ), and use each Class object to fetch the corresponding ArrayList with get( ). And thatís all there is to it.
The filler( ) method is interesting because it takes advantage of the design of ParseTrash.fillBin( ), which doesnít just try to fill an ArrayList but instead anything that implements the Fillable interface with its addTrash( ) method. All filler( ) needs to do is to return a reference to an interface that implements Fillable, and then this reference can be used as an argument to fillBin( ) like this:
To produce this reference, an anonymous inner class (described in Chapter 8 of Thinking in Java, 2nd edition) is used. You never need a named class to implement Fillable, you just need a reference to an object of that class, thus this is an appropriate use of anonymous inner classes.
An interesting thing about this design is that even though it wasnít created to handle the sorting, fillBin( ) is performing a sort every time it inserts a Trash object into bin.
Much of class DynaTrash should be familiar from the previous examples. This time, instead of placing the new Trash objects into a bin of type ArrayList, the bin is of type TypeMap, so when the trash is thrown into bin itís immediately sorted by TypeMapís internal sorting mechanism. Stepping through the TypeMap and operating on each individual ArrayList becomes a simple matter.
As you can see, adding a new type to the system wonít affect this code at all, and the code in TypeMap is completely independent. This is certainly the smallest solution to the problem, and arguably the most elegant as well. It does rely heavily on RTTI, but notice that each key-value pair in the HashMap is looking for only one type. In addition, thereís no way you can ďforgetĒ to add the proper code to this system when you add a new type, since there isnít any code you need to add.
Coming up with a design such as TrashVisitor.java that contains a larger amount of code than the earlier designs can seem at first to be counterproductive. It pays to notice what youíre trying to accomplish with various designs. Design patterns in general strive to separate the things that change from the things that stay the same. The ďthings that changeĒ can refer to many different kinds of changes. Perhaps the change occurs because the program is placed into a new environment or because something in the current environment changes (this could be: ďThe user wants to add a new shape to the diagram currently on the screenĒ). Or, as in this case, the change could be the evolution of the code body. While previous versions of the trash sorting example emphasized the addition of new types of Trash to the system, TrashVisitor.java allows you to easily add new functionality without disturbing the Trash hierarchy. Thereís more code in TrashVisitor.java, but adding new functionality to Visitor is cheap. If this is something that happens a lot, then itís worth the extra effort and code to make it happen more easily.
The discovery of the vector of change is no trivial matter; itís not something that an analyst can usually detect before the program sees its initial design. The necessary information will probably not appear until later phases in the project: sometimes only at the design or implementation phases do you discover a deeper or more subtle need in your system. In the case of adding new types (which was the focus of most of the ďrecycleĒ examples) you might realize that you need a particular inheritance hierarchy only when you are in the maintenance phase and you begin extending the system!
One of the most important things that youíll learn by studying design patterns seems to be an about-face from what has been promoted so far in this book. That is: ďOOP is all about polymorphism.Ē This statement can produce the ďtwo-year-old with a hammerĒ syndrome (everything looks like a nail). Put another way, itís hard enough to ďgetĒ polymorphism, and once you do, you try to cast all your designs into that one particular mold.
What design patterns say is that OOP isnít just about polymorphism. Itís about ďseparating the things that change from the things that stay the same.Ē Polymorphism is an especially important way to do this, and it turns out to be helpful if the programming language directly supports polymorphism (so you donít have to wire it in yourself, which would tend to make it prohibitively expensive). But design patterns in general show other ways to accomplish the basic goal, and once your eyes have been opened to this you will begin to search for more creative designs.
Since the Design Patterns book came out and made such an impact, people have been searching for other patterns. You can expect to see more of these appear as time goes on. Here are some sites recommended by Jim Coplien, of C++ fame (http://www.bell-labs.com/~cope), who is one of the main proponents of the patterns movement:
Also note there has been a yearly conference on design patterns, called PLOP, that produces a published proceedings, the third of which came out in late 1997 (all published by Addison-Wesley).
1. Add a class Plastic to TrashVisitor.java.
2. Add a class Plastic to DynaTrash.java.
3. Create a decorator like VisitableDecorator, but for the multiple dispatching example, along with an ďadapter decoratorĒ class like the one created for VisitableDecorator. Build the rest of the example and show that it works.
A number of more challenging projects for you to solve. [[Some of these may turn into examples in the book, and so at some point might disappear from here]]
First, create a Blackboard (cite reference) which is an object on which anyone may record information. This particular blackboard draws a maze, and is used as information comes back about the structure of a maze from the rats that are investigating it.
Now create the maze itself. Like a real maze, this object reveals very little information about itself ó given a coordinate, it will tell you whether there are walls or spaces in the four directions immediately surrounding that coordinate, but no more. For starters, read the maze in from a text file but consider hunting on the internet for a maze-generating algorithm. In any event, the result should be an object that, given a maze coordinate, will report walls and spaces around that coordinate. Also, you must be able to ask it for an entry point to the maze.
Finally, create the maze-investigating Rat class. Each rat can communicate with both the blackboard to give the current information and the maze to request new information based on the current position of the rat. However, each time a rat reaches a decision point where the maze branches, it creates a new rat to go down each of the branches. Each rat is driven by its own thread. When a rat reaches a dead end, it terminates itself after reporting the results of its final investigation to the blackboard.
The goal is to completely map the maze, but you must also determine whether the end condition will be naturally found or whether the blackboard must be responsible for the decision.
An example implementation by Jeremy Meyer:
The maze initialization file:
A discussion of algorithms to create mazes as well as Java source code to implement them:
A discussion of algorithms for collision detection and other individual/group moving behavior for autonomous physical objects:
Create a pair of decorators for I/O Readers and Writers that encode (for the Writer decorator) and decode (for the reader decorator) XML
Contains tools needed to build the book etc. Some of these may be temporary and disappear when the code base is moved to CVS.
Ant comes with an extension API so that you can create your own tasks by writing them in Java. You can find full details in the official Ant documentation and in the published books on Ant.
As an alternative, you can simply write a Java program and call it from Ant; this way, you donít have to learn the extension API. For example, to compile the code in this book, we need to verify that the version of Java that the user is running is JDK 1.3 or greater, so we created the following program:
This simply uses System.getProperty( ) to discover the Java version, and throws an exception if it isnít at least 1.3. When Ant sees the exception, it will halt. Now you can include the following in any buildfile where you want to check the version number:
If you use this approach to adding tools, you can write them and test them quickly, and if itís justified, you can invest the extra effort and write an Ant extension.
Although useful, the Arrays class stops short of being fully functional. For example, it would be nice to be able to easily print the elements of an array without having to code a for loop by hand every time. And as youíll see, the fill( ) method only takes a single value and places it in the array, so if you wanted, for example, to fill an array with randomly generated numbers, fill( ) is no help.
Thus it makes sense to supplement the Arrays class with some additional utilities, which will be placed in the package com.bruceeckel.util for convenience. These will print an array of any type and fill an array with values or objects that are created by an object called a generator that you can define.
Because code needs to be created for each primitive type as well as Object, thereís a lot of nearly duplicated code. For example, a ďgeneratorĒ interface is required for each type because the return type of next( ) must be different in each case:
Arrays2 contains a variety of toString( ) methods, overloaded for each type. These methods allow you to easily print an array. The toString( ) code introduces the use of StringBuffer instead of String objects. This is a nod to efficiency; when youíre assembling a string in a method that might be called a lot, itís wiser to use the more efficient StringBuffer rather than the more convenient String operations. Here, the StringBuffer is created with an initial value, and Strings are appended. Finally, the result is converted to a String as the return value:
To fill an array of elements using a generator, the fill( ) method takes a reference to an appropriate generator interface, which has a next( ) method that will somehow produce an object of the right type (depending on how the interface is implemented). The fill( ) method simply calls next( ) until the desired range has been filled. Now you can create any generator by implementing the appropriate interface and use your generator with fill( ).
Random data generators are useful for testing, so a set of inner classes is created to implement all the primitive generator interfaces, as well as a String generator to represent Object. You can see that RandStringGenerator uses RandCharGenerator to fill an array of characters, which is then turned into a String. The size of the array is determined by the constructor argument.
 This is a tongue-in-cheek reference to an event in China after the death of Mao-Tze Tung, when four persons including Maoís widow made a power play, and were demonized by the Chinese Communist Party under that name.
 From Mark Johnson.
 But be warned: the examples are in C++.
 A free email publication. See www.BruceEckel.com to subscribe.
 From an email from Kevlin Henney.
 Shalloway, Design Patterns Explained, and Alexandrescu, Advanced C++ Design (??)
 In the Python language, all functions are already objects and so the Command pattern is often redundant.
 Page 235.
 The original version of this was called JPython, but the project changed and the name was changed to emphasize the distinctness of the new version.
 Changing the registry setting python.security.respectJavaAccessibility = true to false makes testing even more powerful because it allows the test script to use *all* methods, even protected and package-private.
 No mice were harmed in the creation of this example.
 Addison-Wesley, 1999.
 This was a solution created by Jaroslav Tulach in a design patterns class that I gave in Prague.
 The C++ programmer will note how much the code could be collapsed with the use of default arguments and templates. The Python programmer will note that this entire library would be largely unnecessary in that language.