Marrying GWT with Spring the Generic Way

During the last days I've played around with GWT and Spring. Although there are already plenty of blog posts describing how to tunnel GWT Services through a Spring DispatcherServlet, they have all one disadvantage in common: They require to do a lot of explicit Spring Bean Definitions. In this post I describe a nice mechanism to automatically create those Bean Definitions, freeing the Bean Config of redundant, boilerplate code.

Transparent Handling of Timestamp Differences with Hibernate

SQL does not define unified methods to calculate differences between timestamp values. Many database vendors provide somehow their own solution for this problem, but when it comes to generalizing the persistence layer with Hibernate, we are faced again with the missing transparent approach. This article shows a concise solution proven to work transparently on Oracle 10g and HSQL.

In our case, we had

1. A model class, let's call it Task. Task has two fields startTime & endTime, both of type java.sql.Timestamp. And several other fields of different primitive types.
2. Task was subject to a Grails application (Although the persistence layer was not GORM, but plain Hibernate with Spring due to several other reasons).
3. We had to display lists of tasks, paginated and orderable by either all fields or the duration, which is actually the difference between endTime and startTime.
4. In production mode the data is stored in an Oracle 10g Database
5. Unit & Integration Tests were done with an In Memory HSQL Database.

These requirements implied the following constraints:

• We have to avoid redundancy. Storing the duration also as a field of Task is not an option. If it could be calculated, then it has to be calculated.
• We want to handle pagination and ordering of the fields and duration similar. We could have introduced some dispatching logic, like if duration is not envolved we use query A otherwise we use query B. But that's not a proper solution and most likely not very well maintainable. We just need a single query which can cope with both cases. Ultimately this means, we have a property name and tell Hibernate to order the result according to this property.
• We need mechanism to cope with Oracle & HSQL.

The solution we found consists of several aspects. Let's go through them top down:

Adding a property formula to the Hibernate Mapping

Since we have to handle Hibernate properties and the calculated value in the same way, we need to add duration as an "artificial property" to the Hibernate Mapping of Task. Luckily Hibernate provides a mechanism for this use case: Properties could consist of a formula, which is basically the call to a valid HQL function. We simply need to define

<hibernate-mapping>
    <class name="foo.Task" table="TASKS">
        ...
        <property 
            name="duration" 
            formula="MY_DIFF_TIME_SECONDS(START_DATE, END_DATE)" 
        />
        ...
    </class>
</hibernate-mapping>

What do we do here? We assume there is a registered function called MY_DIFF_TIME_SECONDS(x,y), which calculates the difference between two timestamps in seconds. This method is called with Task's properties START_DATE and END_DATE.

Adding custom functions to the Hibernate Dialects

We have expected that MY_DIFF_TIME_SECONDS(x,y) exists. Since it is not given by an Intelligent Creator for free, we need to take care that such a function will be created. Hibernate itself does actually the same thing. There are different Hibernate Dialects for different database vendors. Those dialects register several types and function which have individual names in the database with a unified name for Hibernate. We just have to apply this to our function as well.

    public class MyOracle10gDialect extends Oracle10gDialect {

     public MyOracle10gDialect()
     {
      super();
      registerFunction(
          "MY_DIFF_TIME_SECONDS", 
       new SQLFunctionTemplate(
           Hibernate.INTEGER, 
           "MY_DIFF_TIME_SECONDS(?1,?2)"));
     }
    }

This is the example of our custom tailored Oracle Dialect. We extend Hibernate's Dialect, make sure that super() is called and just add our new function. We do the same with MyHsqlDialect by extending HSQLDialect but just don't show it here.

Implementing the functions in the Database

Up to now we've created the unified facade for the Java world using Hibernate, now we need to look at the other side: Calculating timestamp differences in both Databases. This is, as outlined above, the actually problem we are facing. It is possible to do that in Oracle and HSQL, but with two completely different approaches. But since we have unified everything from Hibernate's point of view, we simply have to take care that the databases get a function MY_DIFF_TIME_SECONDS(x,y) to calculate our timestamp differences.

Let's do it for Oracle first:

    CREATE OR REPLACE FUNCTION MY_DIFF_TIME_SECONDS(
        start_date IN DATE,end_date in DATE)
    return NUMBER
    IS
    BEGIN
      if start_date is null 
      then 
        return null;
      end if;
      if end_date is null 
      then 
        return null;
      end if;
      return (end_date - start_date)*24*60*60;
    END;
    /

This is PL/SQL and it does what we need. Besides some defensive null checks we simply utlize the Minus operater, which is also defined on Oracle's date types. Its result is a fraction of a day, hence we need to shift the result to become our difference in seconds.

HSQL does not know PL/SQL (lucky dude!). But it enables us to add arbitrary Java functions which could be called from SQL. Fine thing:

        
    import java.sql.Timestamp;
    import org.joda.time.Period;
    
    public class HsqlStoredFunctions {

     /**
      * Returns the difference between two timestamps in seconds.
      * 
      * @param start the early time
      * @param end the later time
      * @return if start and end != null then the difference is returned,
      *         otherwise null is returned
      * 
      */
     public static Integer diffTimeInSeconds(Timestamp start,Timestamp end){
      if(start == null){
       return null;
      }  
      if(end == null){
       return null;
      }

      Period period = new Period(start.getTime(),end.getTime());

      return period.toStandardSeconds().getSeconds();
     }
    }

This is the function. It has to be static, but that's actually the only requirement. We make it aware to HSQL by:

    public class HsqlDataSourceInitializer implements InitializingBean{

     private JdbcTemplate jdbcTemplate;
     public void setDataSource(DataSource dataSource){
      this.jdbcTemplate = new JdbcTemplate(dataSource);
     }

     @Override
     public void afterPropertiesSet() throws Exception {

      String aliasDiffTime = "CREATE ALIAS MY_DIFF_TIME_SECONDS FOR " +
        " \"mypackage.testing.HsqlStoredFunctions.diffTimeInSeconds\"";
      jdbcTemplate.execute(aliasDiffTime);
     }
    }

Here we see actually two quite interesting features. First the function is registered with the shown SQL statement. Secondly we can also see when we do this registration. Just reconsider: We use HSQL only in our testing scenario. In our particular case the whole persistence layer (and all other layers of the application as well) is nicely decoupled via Spring. Hence the difference between the production and the test environment is the Spring Bean configuration of the DataSource. But since the whole setup is now silently down by Spring we need a hook to register our function when the In Memory Database is created. Therefore HsqlDataSourceInitializer is a Spring Bean which implements InitializingBean. The method afterPropertiesSet() is called after all Spring Beans have been instantiated. Since the DataSource of HSQL Database is nothing but another Spring Bean, we can safely assume that it is up and running.

Setting the sort order on Hibernate Criteria Queries

Now we have a brief look on how we set the order of Hibernate Criteria Queries to show that we have our required unified solution. We've got this nice helper class:

 
    final class CriteriaUtil {

     private static final String ORDER_ASC = "asc";
     private static final String ORDER_DESC = "desc";

     public final Criteria withOrder(Criteria criteria,String sort,String order){

      if(criteria == null){
       throw new NullPointerException();
      }

      if(order != null){
       if(!(order.equals(ORDER_ASC) || order.equals(ORDER_DESC) )){
        throw new IllegalArgumentException("Unknown order direction: " + order);
       }
      }

      if(sort != null){
       if(order != null && order.equals(ORDER_DESC)){
        criteria.addOrder(Property.forName(sort).desc());
       }
       else {
        criteria.addOrder(Property.forName(sort).asc());
       }
      }
      return criteria;
     }
    }

And we can use the attributes as provided by Grails' build-in pagination mechanism to extend our Criteria Queries.

Finally

Although we've just shown a solution for Oracle and HSQL, this approach should be applicable to other Databases as well, as long as it is possible to register custom functions and to calculate the timestamp difference in terms of these functions.

The Different Faces of Polymorphism

Others might think polymorphism is the skill used by Odo from Deep Space Nine when he shifts his form. But we computer geeks now it better. Polymorphism is the one stop show that makes writing software so much fun. Or to put in another way, it provides us with the power of generalization, loosely coupling and ultimately maintainability. But polymorphism does not just come in the shape of the too well known interfaces with its precious virtual methods. Different paradigms provide us with different mechanisms to wield the Sword of the Many Forms. In this article we will give an introduction to several of these incarnations. Namely C++'s compile time polymorphism and that what they call Duck Typing in Dynamic Typed Languages.

Let's start with the basic question: What the hell is polymorphism? It is the ability to specify behavioral contracts on the one hand and to fulfill them on the other. You can say what you expect from someone by defining an contract. He in turn could realize your contract however he wants, as long as he sticks to your defined expectations. With this simple mechanism we basically created the founding of what makes software development possible in the large scope.

Let's start simple.

Runtime Polymorphism

Yeah, we are familiar with this one. It's our daily work. Let us just exercise through it. Here is our contract:

    public interface Phone {
        void call(String phoneNumber);
    }

We simply expect a Phone. If someone provides us something fulfilling the Phone's demands, we can make a call:

    public final class Client {
        void makePhoneCall(Phone phone){
            phone.call("555-6832");
        }
    }

Then there is a very simple Phone implementation to complete the example

    public final class ToyPhone implements Phone {
        public void call(String phoneNumber){
            System.out.println("Ring Ring");
        }
    }

And finally the full scenario breaks down to:

    Phone phone = new ToyPhone();
    Client client = new Client();
    client.makePhoneCall(phone);

All in all not sophisticated, but just all we need to explain the general idea. You see: With the interface we define our contract. Our Client relies on that we give him something like a Phone, then he will just use it. How does it work? This is due to the way Java implements public methods. Actually they are virtual methods, which means, somewhere under the hood there is a vtable, which maps a class' method to a memory address where the implementation of that method could be found. This is all static runtime information. From an execution flow's point of view calling a method means getting the right address from the vtable and then jump there and continue (if we leave out compiler optimization...). Static means, that the compiler already knows where to look into the vtable, but it will contain the proper value only at runtime, of course. This is no magic at all...

Compile Time Polymorphism

Ah ha. Now it is getting more exciting. Let's play the same game, just with C++ meta programming:

    template<typename PHONE> class Client {
    public: 
     void makePhoneCall(PHONE& phone){
      phone.call("555-3223");
     }
    };

    class MobilePhone {
    public: 
     void call(const std::string& phoneNumber);
    };

    void MobilePhone::call(const std::string& phoneNumber){
     std::cout << "Hello?" << std::endl;
    }

    int main(){
     Client<MobilePhone> client;
     MobilePhone phone;
     client.makePhoneCall(phone);
    }

Let's see what we've got here. We see the client, yes that's fine. Ok, this is a templatized class, maybe we would not have expected it, but we just accept it for now. What else? Yes, the implementation of our Phone. This is a C++ class. But, wait: call() is NOT virtual, what the heck... hm let's wait a further second and finish the example. Ok, main() will run our example. And it does what we expect: Create an instance of the client, create an instance of the phone and then let the client do his call. But something is still odd... Yes! Where is the so called contract? It's not there!? And this compiles? Obviously yes... Hm... Now we owe some explanations!

What we see here is called meta programming used by the modern C++ guys and has been first described by Andrei Alexandrescu in Modern C++ Design. He explains how to utilize C++ template compiler for writing programs rather than the actual C++ compiler. Of course this blog post cannot satisfy all which has to be said about meta programming, but just so much: In our particular case we just imply an interface. You and I know, there is nothing which defines our contract explicitly. Psst! But the template compiler does not know that. It just knows that it creates a template instance for the type MobilePhone, and that PHONE and therefore MobilePhone is called on the method call(). While compiling the instantiated code the C++ compiler in turn does its usual job. MobilePhone has such method, hence everything works out fine.

I've heard people argue "What's that nonsense! Templates are for generalizing on types! Like the STL does it!" But think about it: Yes templates are used for generalization. For sure. But in this case we don't generalize on types, we generalize on behaviour. And it turns out more than fine. Are there any benefits? Yes! We've already mentioned the most crucial one: There are no vcalls involved. No vcalls means, the C++ compiler could apply plenty of its optimization strategies, which results in aweful cool runtime behaviour. Eager to see more? Boost and CGAL are awesome examples of modern C++ design.

Actually the C++ guys find this concept of compile time polymorphism that attractive, that they try to incorporate it into a future C++ standard under the term Concepts. Although they've currently postponed it from C++0x.

Duck Typing

Finally we come to our Dynamic Typed Languages. Here they say, if it talks like a duck and walks like a duck then it is a duck. What does this mean? Have a look.

The important aspect here is, besides others, that an instance is not checked for its type during compile time as it is done with Java or C++ for example, but only on runtime, just at the moment a method is called on that instance. For me as a Java Enthusiast it's obvious to give an example in Groovy:

    class Client {
        def makePhoneCall(def phone){
            phone.call('555-3126')
        }
    }
    
    class PhoneBooth {
        def call(def phoneNumber){
            println('Beep...')
        }
    }
    
    static def main(){
        def client = new Client()
        def phone = new PhoneBooth()
        client.makePhoneCall(phone)
    }

It's nearly the same game as with C++ meta programming. We have no explicit definition of our contract. This is just stated implicitly. But since method dispatching is done just when we execute call() on phone, and it's instance PhoneBooth owns such a method, everything works out fine again. But why is it helpful? Because it yields efficient code. Of course, yes, there are a lot of reasons why it is helpful to have a proper static type system. I will not deny it. I also favor it of course. But there are a certain amount of valid use cases, which have other desires, like implementation efficiency. Consider a unit test for example. Maybe you have decoupled everything loosely enough. Your class under test just knows other interfaces. But you still have to create stubs or mocks for those. This is acceptable of course (at least better than to avoid it), but even then it is boiler plate. It does not support the readability of your test. On the other hand, you can easily craft a stub for your interface out of maps and closures. Very dense code, very readable.

Finally

What have we learned today? Polymorphism is cool. And besides this there are different paradigms having different styles of polymorphism. Some have explicitly defined contracts, others don't. But they all obey to the same principle. A contract defines expectations. Implementations fulfill this contract.

Builder Pattern - The Swiss Army Knife of Object Creation, Part 2

Are you still eager to learn how we use the Builder Pattern in Java to build hierarchical structures? Like the Groovy People, but without the dynamic Groovyness? In my last article I prepared everything we need for today. Hence take your pen and paper and follow me into the auditorium. The others are already waiting for us...

Builder Pattern - The Swiss Army Knife of Object Creation

It comes out that one of my new hobbys is feeding and taking care of unpopular design patterns. Up to just recently the Builder Pattern was one of this unloved creatures, but the enormous success of Groovy's awesome transformation of a single Design Pattern to a complete new world of DSLs converted the Builder to everybody's new Darling. This is the first article in a series of some, which discusses the Builder Pattern in static typed Languages like Java. This article gives a brief introduction to the Builder Pattern in general. With Groovy's dynamic typed nature, we are able to declaratively construct structures like this

def swing = new SwingBuilder()

def sharedPanel = {
     swing.panel() {
        label("Shared Panel")
    }
}

count = 0
def textlabel
def frame = swing.frame(
    title:'Frame', 
    defaultCloseOperation:JFrame.EXIT_ON_CLOSE, 
    pack:true, show:true) {
    vbox {
        textlabel = label("Click the button!")
        button(
            text:'Click Me',
            actionPerformed: {
                count++
                textlabel.text = "Clicked ${count} time(s)."
                println "Clicked!"
            }
        )
        widget(sharedPanel())
        widget(sharedPanel())
    }
}

In Groovy's ecosystem this concept nearly became a new religion. Domain models are declared, rather than manually plugged together. Grails and Griffon showcase a mutltiverse of applications. Beginning from configuring Database connections, Spring Beans even good ol' Log4j up to whole GUIs via SwingBuilder in Griffon. Today we declare software. And that's ok. Because it eases reading, writing and understanding.

The point is, is this only possible, if you utilize those new, fancy dynamic typed languages? Couldn't we build structures the declarative way even in our old, poor static languages? It would be sad, if the answer is no. The world is full of legacy application, Joe the Plumber Coder has to work with. Even poor Joe should have the oppertunity to write cool code. Hence gladly the answer is yes. But guys, to be honest, it is not that straight 'Just implement BuilderSupport' road. The path I show you gets dirty, filled with stacks and temporary structures, but is definitely worth the effort. (And since I'm always honest to you: We don't even get there today, but just the next time...)

But first things first: Let's talk about the basics of the Builder Pattern. We should have the same understanding, at least. The Builder Pattern is one of the so called Creational Patterns. It has one single purpose, and that's creating structures. But it does it with some super awesome fancyness. To see how it does it, we need to make one step back and have to reconsider why we actually need help in constructing objects. Surely, instantiating a plain new foo.Bar() is for beginners. But we are talking here about complex structures. Let's make up some real example. Have you ever handcrafted java.awt.GridBagLayout? The last time I did, I was pretty annoyed about building the GridBagConstraints:

JComponent content = getContent();
GridBagConstraints c11 = new GridBagConstraints();
c11.gridx = 1;
c11.gridy = 1;
c11.weighty = 1.0;
c11.weightx = 1.0;

c11.fill = GridBagConstraints.HORIZONTAL;
c11.anchor = GridBagConstraints.FIRST_LINE_START;

backgroundPanel.add(content,c11);

Do you see what happens? Maybe not on the first look (yes, you do, you are a pro, but please do so as if you didn't), because it is very noisy. What we actually wanted to do is adding our content component to the backgroundPanel. Therefore we have to have to create the GridBagConstraints instance, and assign those constraints we want to apply to our content component. But since GridBagConstraints is a dull Pojo, we need to set each value by hand. This means, for each constraint we want to set, we need to write c11.someConstraint = someContraintValue . Yes of course, it could be worse, like we could have to implement it with punch cards, but that's what I just called noisy. When I want to construct something, I want to have it so obvious and redundant free as possible. Constantly repeating c11. = does not kill anyone, but it's also not the best idea we can come up with.

Ok, this is just one of the problems we have here. The next one jumps in our face where we set fill and anchor. and here the annoyance factor is already huger. GridBagLayout is as the rest of Swing from Java's Pre Enum Era (As a matter of fact even before Josh Bloch discovered the Type Safe Enum Pattern): Enumeration Values are encoded as Int Constants. This is bad but it gets worth: GridBagConstraints hosts several different Enumerations in the same scope. On one hand there are those values related to fill, like HORIZONTAL and there are enum values considering anchor like FIRST_LINE_START. Since they are intermixed in the same Naming scope, namely GridBagConstraints, auto complete is no help at all, when it comes to guessing what to assign to fill. You have to look up in the API Docs to see your options. Nothing flaws coding more then breaking your flow of coding if you really have to step into Documentation. And finally you can argue: What's that nonsense with setters? That's why POJOs have Constructors, just use them, and your whole argumenation vanishes. Maybe, but actually i don't like these types of Constructors

GridBagConstraints(
int gridx, int gridy, int gridwidth, int gridheight, 
double weightx, double weighty, int anchor, int fill, 
Insets insets, int ipadx, int ipady) 

Just imagine an instantiation and then your desire to maintain it. Which one was gridwidth? The first or the second? Mh, let's look into Javadocs, and then count the parameters... No, sorry Sir. I waste my time differently. But up to now I've promised much, and proofed nothing. So be so kind and have a look:

backgroundPanel.add(
    getContent(),
    new GbcBuilder()
        .grid(1,1)
        .weight(1.0,1.0)
        .fillHorizontal()
        .anchorLastLineEnd()
        .build() );

Do you see now what I meant with noise? Nothing of it left. When we get to building the constraints, GbcBuilder comes up to our expectations. We even did some helpful things here: Grid values are only assigned in pairs, why bother with two lines if we could have one? Do you see any arbitrary Int Constants anywhere around? No? Auto complete just offers us fillHorizontal(). 'Huh, fill & horizontal, i guess that's what i need, let's take it.' And how did we do this it? Let me show you:

public class GbcBuilder {

    private GridBagConstraints constraints;

    public GbcBuilder(){
        constraints = new GridBagConstraints();
    }

    public GridBagConstraints build(){
        return constraints;
    }

    public GbcBuilder grid(int gridX,int gridY){
        constraints.gridx = gridX;
        constraints.gridy = gridY;
        return this;
    }

    public GbcBuilder weight(double x, double y){
        constraints.weightx = x;
        constraints.weighty = y;
        return this;
    }

    public GbcBuilder fillHorizontal(){
        constraints.fill = GridBagConstraints.HORIZONTAL;
        return this;
    }

    public GbcBuilder anchorLastLineEnd(){
        constraints.anchor = GridBagConstraints.LAST_LINE_END;
        return this;
    }
}

This is actually a super short version of a GridBagConstraintsBuilder. A proper solution would be a little bit huger. Nevertheless, you already see the important aspects. The builder does what he promises: He builds what he should, holds the instance internally and finally spills it out. By constantly returning itself, the Builder gives us this fancy mechanism to repetitively call him. Quite useful feature. But I've already mentioned the drawback. Providing a builder is expensive in terms of implementation time. Is does not pay out if we talk about very specialized structures, which might be initialized only once in your code. But for your working horse model classes, which are used everywhere around the corner, maybe this is quite helpful. This should be enough for now. Today we've learned some interesting things, namely the Groovy People love the Builder, they have invented whole Domain Specific Languages with it. Furthermore the Builder helps us with the proper construction of those structures, which are complex to fill by encapsulating the knowledge how to fill them. Although I've promised, I have not explained how to build complex hierarchies of objects, like we've seen in Groovy and there are even some noteworthy words about Builders and Immutability to say. But we will keep this for the next time. I hope you come back.

Here you can find Part 2.

Problems starting apps on the iPhone 3GS

I had suddenly a very strange problem. While being in the video app, the phone suddenly hung up, the boot (?) apple logo appeared, nothing happened, a few seconds later i was in the unlock screen. From that point on I could not start any downloaded apps anymore (evernote, tap tap revenge, twitterific and so forth). I've resetted the settings to default, this haddened helped. Then I uninstalled one of those apps, downloaded it again from the app store and now everything runs smooth again.

Playing around with Java Scenegraph and Swing

I've just invested *some* hours to play around with

• scenegraph
• macwidgets

The results are pretty amazing, in my honest opinion. Basically I've got a SGPanel with an underlying SGNode tree. Most of the nodes are FXComponents and therefore contain swing components again. I did a quick screencast to demo the implemented effects. Currently I can slide pages to any direction in an iphone like style and overlay other components, in this case a semi transparent dialog, which has some fancy fade out effects. All in all quite cool for good ol' swing.

Visiting the Family - An Advanced Application of the Visitor Pattern

If we want to apply the Visitor Pattern in a complex class hierarchy, the plain Visitor allows us only to visit those classes, which are leafs in the inheritance tree. In this article I'll show a fancy mechanism to overcome this limitation. Not yet bored? Then dive into my explanations.

This is the second post in a loosely scheduled series of articles about the Visitor Pattern. Please find the first one here.

One of those things we've learned last time is the simple fact, that the Visitor will only visit "leafs" in a complex class hierarchy. Otherwise the polymorphic type dispatching will not work. Well, but we are sometimes faced with situations, where this limitation is not acceptable. Let's look at some fancy example, to illustrate my explanations. And since we all spent our two years military service with the imperial guards, here's an obvious taxonomy of different star wars vehicle types. Usually complex type systems consist of several levels of abstraction, like our example. Here we have three:

1. Either everything is the same, namely a Vehicle
2. or it is a Vehicle Class like StarShip or Walker,
3. or it is a concrete Vehicle like, well ... the BWing, for instance.

I think you get the point. With our current knowledge we could only visit concrete vehicles. But what do we do if we are a fleet commander and want to command all of our units to start moving. And moving means StarShips start flying and Walker go walking, have a look:

        
    // StarShips...
    public abstract class StarShip extends Vehicle{

        public void fly(){
            System.out.println("Yeaahhh, I'm flying, Baby.");
        }
    }    
    
    // Walker
    public abstract class Walker extends Vehicle{

        public void walk(){
            System.out.println("I'm going for a walk.");
        }
    }    

As a fleet commander we know for sure that we can utilize the Visitor Pattern to command our Vehicle [] fleet = { new BWing(), new XWing(), new AtAt() }.

A fleet lieutenant would do it like this:

    // clumsy fleet lieutenant's way to command his units
    public class ClumsyVisitor implements Visitor{
        public void visit(BWing bwing) {
            bwing.fly();
        }

        public void visit(XWing xwing) {
            xwing.fly();
        }

        public void visit(AtAt atat) {
            atat.walk();
        }
    }    

But we would not have become commander if we would unnecessarily repeat ourselves. As you've already realized, this is code duplication since we don't operate on the correct level of abstraction. The proper way is calling methods on the Vehicle Classes, not on the concrete Vehicles. But again, how to we do that? Here's the solution:

    public abstract class AbstractVehicleClassVisitor
            implements Visitor {

        public abstract void visit(Walker walker);
        public abstract void visit(StarShip starShip);

        public void visit(BWing bwing) {
            visit((StarShip) bwing);
        }

        public void visit(XWing xwing) {
            visit((StarShip) xwing);
        }

        public void visit(AtAt atat) {
            visit((Walker) atat);
        }
    }    

Tata: We've just introduced the proper level of abstraction to our Visitor as well by deriving a new type from Visitor, the AbstractVehicleClassVisitor. Or to be more precise, we defined the missing visit(Walker) and visit(StarShip) methods and delegate the 'leaf visits' to these new abstract methods. Visiting Vehicle Classes is now done by extending the AbstractVehicleVisitor. Hence our fleet commander command simply looks like:

    // that's how we command!
    public class MoveVisitor 
    extends AbstractVehicleClassVisitor{

        public void visit(Walker walker) {
            walker.walk();
        }

        public void visit(StarShip starShip) {
            starShip.fly();
        }
    }  

Calling for the troops breaks down to

    Vehicle [] fleet = { new BWing(), new XWing(), new AtAt() };
    for(Vehicle unit : fleet) {
        unit.accept(new MoveVisitor());
    }
    
    /* Which would output
    Yeaahhh, I'm flying, Baby.
    Yeaahhh, I'm flying, Baby.
    I'm going for a walk.
    */

Yep, that's all. Easy, isn't?

My Two Cents about Coding in English

Jens Schauder blogged about English as the Lingua Franca in Software projects. He concludes that non native speakers should force themselves into doing their work in English to compete in a more and more international market.

After working for three years in an international project I absolutely agree with his opinion. Having no other chance to communicate with my japanese and indian colleagues than in english, the first few weeks were really hard. But I've learned several things:

1. It does not matter whether you speak 100% grammatically correct. It's more important to talk fluent and confident rather than being sure, that every grammar rule you've learned 10 years ago is correctly applied. My impression is people respect me because of my technical reputation, not because I do speak absolutely error free. (I doubt that I do, this blog gives many proofs)
2. The more you do things, the better you get. And the more confident you get. As Jens said, it's right, start reading things english, write things in english, and if you have the chance, do speak as much english as you can.

But meanwhile I've also seen the other side of the medal: If your code is commented in your mother tongue - or even more worse: Implemented in your mother tongue - then you've got a more or less limited market of people how can replace you. If the code is in english you extend the potential candidates to the rest of the industrialized world. And if you now consider that people in east europe, asia and india cost about a fourth of your salary, then you might start worrying. Currently I do not worry about these aspects, but if you were stuck in your job for 20 years, have not learned contemporary technologies then you might not be attractive for a modern, agile market and maybe you have a different opinion about this topic. But on the other hand, if you are one these people, you might not read blogs as well :-)

Hail to the Visitor Pattern

In my opinion the Visitor Pattern is one of the often underestimated patterns from the GOF book. For instance in Head First Design Patterns it just gets it share in the orphaned further patterns section. But applied correctly, the Visitor is a powerful tool helping you to dispatch complex type systems without violating the Open Closed Principle. Interested? Then continue reading!

Let's fetch up those of you, who are not familiar with the Visitor at all. This pattern enables injecting methods into types. 'Why the hell, do I want to do this?' you might ask. Well, consider you have some micky mouse type hierarchy of geometry related classes:

And you want to add functionality to these classes, like drawing. You could add a purely virtual method draw() to the super class and provide an implementation in all of the sub classes, since all of our subclasses are drawn differently, of course. Yes, indeed you could do this. But what if such a method would introduce dependencies you damn want to avoid with your plain classes? To stay in our example, let's assume we want draw our Geometries with QT. Hence we would implement QT depending code in draw(). This would make our classes dependent of QT. But what if we don't want this? Because we also want to allow drawing it in AWT? what now? Add drawMeWithAWT() as well? And then add new dependencies? I don't know about you, but surely I don't want this! Therefore, what else can we do.

Ah, I've got an idea. There must have been a reason, why god (or at least James Gosling) has invented instanceof in Java:

public void drawGeometry(Geometry g){
    
    if(g instanceof Polygon){
        drawPolygon((Polygon) g);
    }else if(g instanceof Circle){
        drawCircle((Circle) g);
    }else if(g instanceof Line){
        drawLine((Line) g);
    }else {
        throw new RuntimeException("Unknown type");
    }
}

Ok, let's look at the code closely. Does your brain hurt and do your eyes bleed already? No? Look again! They should! Please say loud to yourself: 'This is bad code!'. Why? Because it's the prototype example of violating the Open Closed Principle: You have some generalization (here Geometry) and make assumptions about existing sub types. The overall idea of generalization is well, generalize the fact that there are sub types. But why is it so awful evil? Because it is not maintainable. What happens if you start adding new sub types to Geometry? You would have to extend every single dispatching block and add a new else if for your new type. And code which relies on this idiom does not stop with a single block of such if else nonsense. No. It will spread through the whole code. And I would bet a serious amount of money (if I would have a serious amount of money) that you miss at least one place where you have to add your new sub type.

The rescue to this problem is polymorphism. But we've already been at this point. So we don't want polymorphism in the type hierarchy itself like our draw() method. But what now? Well, please step aside for the Visitor pattern. It lets you write specialized handling for each type, but prevents you from implementing the type dispatching on your own. How? Have a look:

The trick is SubElement's implementation of accept:

 
public void accept(Visitor visitor){ 
    visitor.visit(this); 
} 

This is the magic polymorphic dispatching code! Let's see it in action:

Element elem = new SubElement();
elem.accept(new ConcreteVisitor());

Ok, not much action. But all we need for the explanation. We treat our SubElement as a general Element. Calling the polymorphic accept actually on SubElement will take care that the visit(SubElement) is finally executed. Hence we did our dispatching, but without any if or even else. That's the power of polymorphism!

Now let's go back to our Geometry world and see a more complex example:

Here we have our DrawVisitor, who is responsible for drawing. The class hierarchy just depends on Visitor and therefore knows nothing about our concrete DrawVisitor. This goal has been reached. But, wait, we get even more, for free. People usually advice you to use the Visitor on a stable class hierarchy, hence on such, which does not frequently add new classes. Of course, this is correct. Let's assume our classes would belong to a framework and therefore we would not know anything about the client. if we constantly add new classes to the model and change the interface people could become quite angry with us.

But (and that's a very huge but in my opinion) if you change your model, add new classes and add the respective visit(NewClass) methods to the Visitor then you get a compile time check whether you have taken care of handling the new classes in all of your visitor's implementations. And having problems reported by the compiler is one of the major advantages of a static type system, in my honest opinion.

You might have noticed that Visitor visits all concrete classes, but not the abstract super class. This is the necessary otherwise the whole dispatching mechanism does not work. But on the other hand this introduces some challenges on more complex type systems, where we have several levels of abstraction. Even here we can use the visitor and build some handy tools with it to savely navigate such hierarchies, but I think this might be an issue for another post.

Summarizing what we've learned today: The Visitor pattern enables us to dispatch type hierarchies without violating the Open Closed principle by injecting methods in classes and relying on good old runtime polymorphism.

Or do you have a complete different opinion on this topic?

Using ant to manage larger Java projects

In this post I'll give a brief introduction to my experience of managing larger java projects with ant by sticking to the DRY principle.

One of the first steps during setting up a java project is surely the carefully planning of the build system. A build system's purpose is doing the always repeating tasks that occur during a project's life circle. The obvious ones are creating temporary directories, calling the compiler, packaging everything to a jar. If the complexity of the projects grows, it is quite certain that the requirements to the build system grow as well. Surely we want to include unit testing, javadoc generation. And what about automatic coverage analysis? Or ordering a pizza? Yes. Surely it should be able to cope with these challenges (maybe not the pizza thing, but i would love that...). There are roughly 2 approaches to get a build system in java: Either rely on the functionality your favored IDE provides, or use one of widely accepted build systems like ant or maven. In my opinion there are a lot of reasons not to choose the first one. Most importantly a particular project should not depend on a certain IDE. Maybe there's a common agreement in your company to focus on IDE, but this must not be the common case. As for every other aspect of software development as well, you should try to avoid dependencies, or least the unnecessary ones. On the other hand it is good to avoid complexity as well. If your source code builds in a shell, then it will most likely build every else. Hence, let's use some established build system like ant or maven.

I am using ant for about eight years now, and I am quite happy with it. My experiences are even that good, that I avoided the master the fully learning curve towards maven up to now. But this post is not about ant and maven. Actually I want to talk about how to maintain more complex projects. Let's start small. In a common ant project you will most likely have a (very simplified) structure like

/root 
    /src 
    /build 
        /bin 
        /jars
    build.xml

Your source code lives in src, it gets compiled to build/bin and jared to build/jars. You would have some targets, let's say init, compile, jar, which create some dirs, compile the sources and, who would have guessed, pack them. Anyways. But reality tends to be more complex than mickey mouse examples. If your project is larger, it could be a good advice to break your code up into different libraries. And even more important, with clear policies how to put things into libraries. There are several aspects which would affect this decision. Since your architecture is most likely organized in different components, it is wise to reflect this structure in your source code too. You should already have some policies for dependencies of your components, like

• more specialized components have to depend on more generalized ones
• don't rely on transitive dependencies
• permit cyclic dependencies

These rules are also valid on the actual library level.

But how should we reflect this in the directory layout? I've made good experiences with the following approach:

/root
   /libs
     /commons
        /src
        /build
           ...
        build.xml
     /derived
        /src
        /build
        ...   
        build.xml

But now we are getting to the root of the matter: All build files will more or less do the same. If we just copy them we would surely violate the DRY principle, and that's something we should avoid. Let's define what we need: We want to omit to redundantly specify the build process. Hence we want to have something like a template, which we always use, but, and that's important, it should leave enough freedom to make local modifications to this template. Why? Let's assume we have such a template mechanism specified somewhere and we call libraries with it. This is fine, as long as the libraries always have the same requirements. But what if we have one particular lib which differs from these requirements? let's say it consists of generated code, therefore it would have to call a source code generator after init but before compile? we could put this special target into the template as well and somehow ignore it as long as there's nothing to generate. But in my opinion this violates OCP. It would be better to provide some "hooks" when calling the template to do the special things.

Luckily ant has some nice tool in its toolbox to overcome this problem: The import target. This will let us specify some super build file, import its target and *tata* override some targets (if we want to). Let's think about import as ant's way of polymorphism. The super class (build file!) defines some abstract behaviour, the base class individualizes it to its specific needs. Therefore we will define the overall and general build steps in our template file and put wisely chosen hooks into it, which could be overridden by the importing build file. A very simplified super build file might look like

    <project name="Complex Ant Demo Master Libs">

        <!-- general library related properties -->


        <property name="dir.src" location="${dir.library}/src"/>
        <property name="dir.build" location="${dir.library}/build"/>

        <property name="dir.bin" location="${dir.build}/bin"/>
        <property name="dir.jars" location="${dir.build}/jars"/>

        <property name="file.lib.jar" value="${dir.jars}/${name.library}.jar"/>


        <target name="hook.init.pre"/>
        <target name="hook.init.post"/>

        <target name="init" description="Initializes the Library">

            <antcall target="hook.init.pre" />

            <mkdir dir="${dir.build}"/>    
            <mkdir dir="${dir.bin}"/>
            <mkdir dir="${dir.jars}"/>

            <antcall target="hook.init.post" />

        </target>

        <target name="hook.compiler.pre"/>
        <target name="hook.compile.post"/>

        <target name="compile" description="Compiles Sources" depends="init">
            <antcall target="hook.compiler.pre"/>

            <javac srcdir="${dir.src}" destdir="${dir.bin}"/>

            <antcall target="hook.compile.post" />
        </target>

        <target name="hook.jar.pre"/>
        <target name="hook.jar.post"/>

        <target name="jar" description="Creates Jars" depends="compile">
            <antcall target="hook.jar.pre" />

            <jar basedir="${dir.build}" destfile="${file.lib.jar}"/>

            <antcall target="hook.jar.post" />
        </target>    

        <target name="hook.clean.pre"/>
        <target name="hook.clean.post"/>

        <target name="clean" description="Cleans Up">

            <antcall target="hook.clean.pre"/>

            <delete dir="${dir.build}" />

            <antcall target="hook.clean.post"/>
        </target>

    </project> 

That's a long piece of example. But let's go through focussing on the interesting steps. First of all, we define some variables, which we will use later. Please note, that we implicitly assume there are variables called ${dir.library} and ${name.library}. We will see later where they come from. It's getting interesting in line 15. There we define those empty targets hook.init.pre and hook.init.post. But there purpose becomes clear in the init target below: First we call the (possible empty) pre hook, then we do our init action, finally the post hook. This mechanism is used in each target. (to be honest: I've seen this idea in the netbeans build system first...)

Let's assume we have library called "common", without any special treatment. Its build file is very simple now:

    <project name="Common Lib" basedir=".">

        <property name="dir.library" location="${basedir}"/>
        <property name="name.library" value="common"/>
        <property name="dir.project-root" location="${basedir}/../.."/>

        <import file="${dir.project-root}/build-master-lib.xml"/>

    </project> 

We simply specify those two implicit variables, just define the location for the super build file and then import it. Now we can call all the imported targets on common's build file. For instance

ant jar

would result in the appropriate jar file.

Finally consider our described specialized case. Do you remember? We want to call the code generator before compile. That's how we do it:

    <project name="A Generated Lib" basedir=".">

        <property name="dir.library" location="${basedir}"/>
        <property name="name.library" value="generated"/>
        <property name="dir.project-root" location="${basedir}/../.."/>

        <import file="${dir.project-root}/build-master-lib.xml"/>

        <property name="dir.gen" location="${dir.src}"/>

        <!-- overrides target from build-master-lib.xml -->
        <target name="hook.post.init">
            <mkdir dir="${dir.gen}"/>
        </target>

        <target name="hook.compiler.pre">
            <echo>Calling the source code generator</echo>
        </target>

        <target name="hook.clean.post">
            <mkdir dir="${dir.gen}"/>
        </target>

    </project>

We simply redefined the hooks to do all the housekeeping we need.

Of course this is only the first step in setting up a full blown, generically managed ant project. But this should be enough for now.