SeamFramework.orgCommunity Documentation

Chapter 1. Seam Tutorial

1.1. Using the Seam examples
1.1.1. Running the examples on JBoss AS
1.1.2. Running the examples on Tomcat
1.1.3. Running the example tests
1.2. Your first Seam application: the registration example
1.2.1. Understanding the code
1.2.2. How it works
1.3. Clickable lists in Seam: the messages example
1.3.1. Understanding the code
1.3.2. How it works
1.4. Seam and jBPM: the todo list example
1.4.1. Understanding the code
1.4.2. How it works
1.5. Seam pageflow: the numberguess example
1.5.1. Understanding the code
1.5.2. How it works
1.6. A complete Seam application: the Hotel Booking example
1.6.1. Introduction
1.6.2. Overview of the booking example
1.6.3. Understanding Seam conversations
1.6.4. The Seam Debug Page
1.7. Nested conversations: extending the Hotel Booking example
1.7.1. Introduction
1.7.2. Understanding Nested Conversations
1.8. A complete application featuring Seam and jBPM: the DVD Store example
1.9. Bookmarkable URLs with the Blog example
1.9.1. Using "pull"-style MVC
1.9.2. Bookmarkable search results page
1.9.3. Using "push"-style MVC in a RESTful application

Seam provides a number of example applications demonstrating how to use the various features of Seam. This tutorial will guide you through a few of those examples to help you get started learning Seam. The Seam examples are located in the examples subdirectory of the Seam distribution. The registration example, which will be the first example we look at, is in the examples/registration directory.

Each example has the same directory structure:

The example applications run both on JBoss AS and Tomcat with no additional configuration. The following sections will explain the procedure in both cases. Note that all the examples are built and run from the Ant build.xml, so you'll need a recent version of Ant installed before you get started.

The examples are also configured for use on Tomcat 6.0. You will need to follow the instructions in Section 30.6.1, “Installing Embedded JBoss” for installing JBoss Embedded on Tomcat 6.0. JBoss Embedded is only required to run the Seam demos that use EJB3 components on Tomcat. There are also examples of non-EJB3 applications that can be run on Tomcat without the use of JBoss Embedded.

You'll need to set tomcat.home, in the shared build.properties file in the root folder of your Seam installation, to the location of your Tomcat installation. make sure you set the location of your Tomcat.

You'll need to use a different Ant target when using Tomcat. Use ant tomcat.deploy in example subdirectory to build and deploy any example for Tomcat.

On Tomcat, the examples deploy to URLs like /jboss-seam-example, so for the registration example the URL would be http://localhost:8080/jboss-seam-registration/. The same is true for examples that deploy as a WAR, as mentioned in the previous section.

The registration example is a simple application that lets a new user store his username, real name and password in the database. The example isn't intended to show off all of the cool functionality of Seam. However, it demonstrates the use of an EJB3 session bean as a JSF action listener, and basic configuration of Seam.

We'll go slowly, since we realize you might not yet be familiar with EJB 3.0.

The start page displays a very basic form with three input fields. Try filling them in and then submitting the form. This will save a user object in the database.

The example is implemented with two Facelets templates, one entity bean and one stateless session bean. Let's take a look at the code, starting from the "bottom".

We need an EJB entity bean for user data. This class defines persistence and validation declaratively, via annotations. It also needs some extra annotations that define the class as a Seam component.

Example 1.1. User.java

(1)@Entity

(2)@Name("user")
(3)@Scope(SESSION)
(4)@Table(name="users")
public class User implements Serializable
{
   private static final long serialVersionUID = 1881413500711441951L;
   
(5)   private String username;
   private String password;
   private String name;
   
   public User(String name, String password, String username)
   {
      this.name = name;
      this.password = password;
      this.username = username;
   }
   
(6)   public User() {}
   
(7)   @NotNull @Length(min=5, max=15)
   public String getPassword()
   {
      return password;
   }
   public void setPassword(String password)
   {
      this.password = password;
   }
   
   @NotNull
   public String getName()
   {
      return name;
   }
   public void setName(String name)
   {
      this.name = name;
   }
   
(8)   @Id @NotNull @Length(min=5, max=15)
   public String getUsername()
   {
      return username;
   }
   public void setUsername(String username)
   {
      this.username = username;
   }
}
1

The EJB3 standard @Entity annotation indicates that the User class is an entity bean.

2

A Seam component needs a component name specified by the @Name annotation. This name must be unique within the Seam application. When JSF asks Seam to resolve a context variable with a name that is the same as a Seam component name, and the context variable is currently undefined (null), Seam will instantiate that component, and bind the new instance to the context variable. In this case, Seam will instantiate a User the first time JSF encounters a variable named user.

3

Whenever Seam instantiates a component, it binds the new instance to a context variable in the component's default context. The default context is specified using the @Scope annotation. The User bean is a session scoped component.

4

The EJB standard @Table annotation indicates that the User class is mapped to the users table.

5

name, password and username are the persistent attributes of the entity bean. All of our persistent attributes define accessor methods. These are needed when this component is used by JSF in the render response and update model values phases.

6

An empty constructor is both required by both the EJB specification and by Seam.

7

The @NotNull and @Length annotations are part of the Hibernate Validator framework. Seam integrates Hibernate Validator and lets you use it for data validation (even if you are not using Hibernate for persistence).

8

The EJB standard @Id annotation indicates the primary key attribute of the entity bean.


The most important things to notice in this example are the @Name and @Scope annotations. These annotations establish that this class is a Seam component.

We'll see below that the properties of our User class are bound directly to JSF components and are populated by JSF during the update model values phase. We don't need any tedious glue code to copy data back and forth between the JSP pages and the entity bean domain model.

However, entity beans shouldn't do transaction management or database access. So we can't use this component as a JSF action listener. For that we need a session bean.

Most Seam application use session beans as JSF action listeners (you can use JavaBeans instead if you like).

We have exactly one JSF action in our application, and one session bean method attached to it. In this case, we'll use a stateless session bean, since all the state associated with our action is held by the User bean.

This is the only really interesting code in the example!

Example 1.2. RegisterAction.java

@Stateless    (1)
@Name("register")
public class RegisterAction implements Register
{
   @In
   private Use(2)r user;
   
   @PersistenceContext
   private Ent(3)ityManager em;
   
   @Logger
   private Log(4) log;
   
   public String register()
   {          (5)
      List existing = em.createQuery(
         "select username from User where username = #{user.username}")
         .getR(6)esultList();
         
      if (existing.size()==0)
      {
         em.persist(user);
         log.info("Registered new user #{user.username}");
         retur(7)n "/registered.xhtml";
      }       (8)
      else
      {
         FacesMessages.instance().add("User #{user.username} already exists");
         retur(9)n null;
      }
   }

}
1

The EJB @Stateless annotation marks this class as a stateless session bean.

2

The @In annotation marks an attribute of the bean as injected by Seam. In this case, the attribute is injected from a context variable named user (the instance variable name).

3

The EJB standard @PersistenceContext annotation is used to inject the EJB3 entity manager.

4

The Seam @Logger annotation is used to inject the component's Log instance.

5

The action listener method uses the standard EJB3 EntityManager API to interact with the database, and returns the JSF outcome. Note that, since this is a session bean, a transaction is automatically begun when the register() method is called, and committed when it completes.

6

Notice that Seam lets you use a JSF EL expression inside EJB-QL. Under the covers, this results in an ordinary JPA setParameter() call on the standard JPA Query object. Nice, huh?

7

The Log API lets us easily display templated log messages which can also make use of JSF EL expressions.

8

JSF action listener methods return a string-valued outcome that determines what page will be displayed next. A null outcome (or a void action listener method) redisplays the previous page. In plain JSF, it is normal to always use a JSF navigation rule to determine the JSF view id from the outcome. For complex application this indirection is useful and a good practice. However, for very simple examples like this one, Seam lets you use the JSF view id as the outcome, eliminating the requirement for a navigation rule. Note that when you use a view id as an outcome, Seam always performs a browser redirect.

9

Seam provides a number of built-in components to help solve common problems. The FacesMessages component makes it easy to display templated error or success messages. (As of Seam 2.1, you can use StatusMessages instead to remove the semantic dependency on JSF). Built-in Seam components may be obtained by injection, or by calling the instance() method on the class of the built-in component.


Note that we did not explicitly specify a @Scope this time. Each Seam component type has a default scope if not explicitly specified. For stateless session beans, the default scope is the stateless context, which is the only sensible value.

Our session bean action listener performs the business and persistence logic for our mini-application. In more complex applications, we might need require a separate service layer. This is easy to achieve with Seam, but it's overkill for most web applications. Seam does not force you into any particular strategy for application layering, allowing your application to be as simple, or as complex, as you want.

Note that in this simple application, we've actually made it far more complex than it needs to be. If we had used the Seam application framework controllers, we would have eliminated all of our application code. However, then we wouldn't have had much of an application to explain.

The view pages for a Seam application could be implemented using any technology that supports JSF. In this example we use Facelets, because we think it's better than JSP.


The only thing here that is specific to Seam is the <s:validateAll> tag. This JSF component tells JSF to validate all the contained input fields against the Hibernate Validator annotations specified on the entity bean.


This is a simple Facelets page using some inline EL. There's nothing specific to Seam here.

Since this is the first Seam app we've seen, we'll take a look at the deployment descriptors. Before we get into them, it is worth noting that Seam strongly values minimal configuration. These configuration files will be created for you when you create a Seam application. You'll never need to touch most of these files. We're presenting them now only to help you understand what all the pieces in the example are doing.

If you've used many Java frameworks before, you'll be used to having to declare all your component classes in some kind of XML file that gradually grows more and more unmanageable as your project matures. You'll be relieved to know that Seam does not require that application components be accompanied by XML. Most Seam applications require a very small amount of XML that does not grow very much as the project gets bigger.

Nevertheless, it is often useful to be able to provide for some external configuration of some components (particularly the components built in to Seam). You have a couple of options here, but the most flexible option is to provide this configuration in a file called components.xml, located in the WEB-INF directory. We'll use the components.xml file to tell Seam how to find our EJB components in JNDI:


This code configures a property named jndiPattern of a built-in Seam component named org.jboss.seam.core.init. The funny @ symbols are there because our Ant build script puts the correct JNDI pattern in when we deploy the application, which it reads from the components.properties file. You learn more about how this process works in Section 5.2, “Configuring components via components.xml.

The presentation layer for our mini-application will be deployed in a WAR. So we'll need a web deployment descriptor.


This web.xml file configures Seam and JSF. The configuration you see here is pretty much identical in all Seam applications.

Most Seam applications use JSF views as the presentation layer. So usually we'll need faces-config.xml. In our case, we are going to use Facelets for defining our views, so we need to tell JSF to use Facelets as its templating engine.


Note that we don't need any JSF managed bean declarations! Our managed beans are annotated Seam components. In Seam applications, the faces-config.xml is used much less often than in plain JSF. Here, we are simply using it to enable Facelets as the view handler instead of JSP.

In fact, once you have all the basic descriptors set up, the only XML you need to write as you add new functionality to a Seam application is orchestration: navigation rules or jBPM process definitions. Seam's stand is that process flow and configuration data are the only things that truly belong in XML.

In this simple example, we don't even need a navigation rule, since we decided to embed the view id in our action code.

When the form is submitted, JSF asks Seam to resolve the variable named user. Since there is no value already bound to that name (in any Seam context), Seam instantiates the user component, and returns the resulting User entity bean instance to JSF after storing it in the Seam session context.

The form input values are now validated against the Hibernate Validator constraints specified on the User entity. If the constraints are violated, JSF redisplays the page. Otherwise, JSF binds the form input values to properties of the User entity bean.

Next, JSF asks Seam to resolve the variable named register. Seam uses the JNDI pattern mentioned earlier to locate the stateless session bean, wraps it as a Seam component, and returns it. Seam then presents this component to JSF and JSF invokes the register() action listener method.

But Seam is not done yet. Seam intercepts the method call and injects the User entity from the Seam session context, before allowing the invocation to continue.

The register() method checks if a user with the entered username already exists. If so, an error message is queued with the FacesMessages component, and a null outcome is returned, causing a page redisplay. The FacesMessages component interpolates the JSF expression embedded in the message string and adds a JSF FacesMessage to the view.

If no user with that username exists, the "/registered.xhtml" outcome triggers a browser redirect to the registered.xhtml page. When JSF comes to render the page, it asks Seam to resolve the variable named user and uses property values of the returned User entity from Seam's session scope.

Clickable lists of database search results are such an important part of any online application that Seam provides special functionality on top of JSF to make it easier to query data using EJB-QL or HQL and display it as a clickable list using a JSF <h:dataTable>. The messages example demonstrates this functionality.

The message list example has one entity bean, Message, one session bean, MessageListBean and one JSP.

Just like in the previous example, we have a session bean, MessageManagerBean, which defines the action listener methods for the two buttons on our form. One of the buttons selects a message from the list, and displays that message. The other button deletes a message. So far, this is not so different to the previous example.

But MessageManagerBean is also responsible for fetching the list of messages the first time we navigate to the message list page. There are various ways the user could navigate to the page, and not all of them are preceded by a JSF action — the user might have bookmarked the page, for example. So the job of fetching the message list takes place in a Seam factory method, instead of in an action listener method.

We want to cache the list of messages in memory between server requests, so we will make this a stateful session bean.

Example 1.11. MessageManagerBean.java

@Stateful
@Scope(SESSION)
@Name("messageManager")
public class MessageManagerBean implements Serializable, MessageManager
{
   @DataModel
   private Lis(1)t<Message> messageList;
   
   @DataModelSelection
   @Out(requir(2)ed=false)
   private Mes(3)sage message;
   
   @PersistenceContext(type=EXTENDED)
   private Ent(4)ityManager em;
   
   @Factory("messageList")
   public void(5) findMessages()
   {
      messageList = em.createQuery("select msg from Message msg order by msg.datetime desc")
                      .getResultList();
   }
   
   public void select()
   {          (6)
      message.setRead(true);
   }
   
   public void delete()
   {          (7)
      messageList.remove(message);
      em.remove(message);
      message=null;
   }
   
   @Remove
   public void(8) destroy() {}

}
1

The @DataModel annotation exposes an attibute of type java.util.List to the JSF page as an instance of javax.faces.model.DataModel. This allows us to use the list in a JSF <h:dataTable> with clickable links for each row. In this case, the DataModel is made available in a session context variable named messageList.

2

The @DataModelSelection annotation tells Seam to inject the List element that corresponded to the clicked link.

3

The @Out annotation then exposes the selected value directly to the page. So every time a row of the clickable list is selected, the Message is injected to the attribute of the stateful bean, and the subsequently outjected to the event context variable named message.

4

This stateful bean has an EJB3 extended persistence context. The messages retrieved in the query remain in the managed state as long as the bean exists, so any subsequent method calls to the stateful bean can update them without needing to make any explicit call to the EntityManager.

5

The first time we navigate to the JSP page, there will be no value in the messageList context variable. The @Factory annotation tells Seam to create an instance of MessageManagerBean and invoke the findMessages() method to initialize the value. We call findMessages() a factory method for messages.

6

The select() action listener method marks the selected Message as read, and updates it in the database.

7

The delete() action listener method removes the selected Message from the database.

8

All stateful session bean Seam components must have a method with no parameters marked @Remove that Seam uses to remove the stateful bean when the Seam context ends, and clean up any server-side state.


Note that this is a session-scoped Seam component. It is associated with the user login session, and all requests from a login session share the same instance of the component. (In Seam applications, we usually use session-scoped components sparingly.)

The JSP page is a straightforward use of the JSF <h:dataTable> component. Again, nothing specific to Seam.

Example 1.13. messages.jsp


<%@ taglib uri="http://java.sun.com/jsf/html" prefix="h" %>
<%@ taglib uri="http://java.sun.com/jsf/core" prefix="f" %>
<html>
 <head>
  <title>Messages</title>
 </head>
 <body>
  <f:view>
   <h:form>
     <h2>Message List</h2>
     <h:outputText value="No messages to display" 
                   rendered="#{messageList.rowCount==0}"/>
     <h:dataTable var="msg" value="#{messageList}" 
                  rendered="#{messageList.rowCount>0}">
        <h:column>
           <f:facet name="header">
              <h:outputText value="Read"/>
           </f:facet>
           <h:selectBooleanCheckbox value="#{msg.read}" disabled="true"/>
        </h:column>
        <h:column>
           <f:facet name="header">
              <h:outputText value="Title"/>
           </f:facet>
           <h:commandLink value="#{msg.title}" action="#{messageManager.select}"/>
        </h:column>
        <h:column>
           <f:facet name="header">
              <h:outputText value="Date/Time"/>
           </f:facet>
           <h:outputText value="#{msg.datetime}">
              <f:convertDateTime type="both" dateStyle="medium" timeStyle="short"/>
           </h:outputText>
        </h:column>
        <h:column>
           <h:commandButton value="Delete" action="#{messageManager.delete}"/>
        </h:column>
     </h:dataTable>
     <h3><h:outputText value="#{message.title}"/></h3>
     <div><h:outputText value="#{message.text}"/></div>
   </h:form>
  </f:view>
 </body>
</html>

The first time we navigate to the messages.jsp page, the page will try to resolve the messageList context variable. Since this context variable is not initialized, Seam will call the factory method findMessages(), which performs a query against the database and results in a DataModel being outjected. This DataModel provides the row data needed for rendering the <h:dataTable>.

When the user clicks the <h:commandLink>, JSF calls the select() action listener. Seam intercepts this call and injects the selected row data into the message attribute of the messageManager component. The action listener fires, marking the selected Message as read. At the end of the call, Seam outjects the selected Message to the context variable named message. Next, the EJB container commits the transaction, and the change to the Message is flushed to the database. Finally, the page is re-rendered, redisplaying the message list, and displaying the selected message below it.

If the user clicks the <h:commandButton>, JSF calls the delete() action listener. Seam intercepts this call and injects the selected row data into the message attribute of the messageList component. The action listener fires, removing the selected Message from the list, and also calling remove() on the EntityManager. At the end of the call, Seam refreshes the messageList context variable and clears the context variable named message. The EJB container commits the transaction, and deletes the Message from the database. Finally, the page is re-rendered, redisplaying the message list.

jBPM provides sophisticated functionality for workflow and task management. To get a small taste of how jBPM integrates with Seam, we'll show you a simple "todo list" application. Since managing lists of tasks is such core functionality for jBPM, there is hardly any Java code at all in this example.

The center of this example is the jBPM process definition. There are also two JSPs and two trivial JavaBeans (There was no reason to use session beans, since they do not access the database, or have any other transactional behavior). Let's start with the process definition:


If we view this process definition using the process definition editor provided by JBossIDE, this is what it looks like:

This document defines our business process as a graph of nodes. This is the most trivial possible business process: there is one task to be performed, and when that task is complete, the business process ends.

The first JavaBean handles the login screen login.jsp. Its job is just to initialize the jBPM actor id using the actor component. In a real application, it would also need to authenticate the user.


Here we see the use of @In to inject the built-in Actor component.

The JSP itself is trivial:


The second JavaBean is responsible for starting business process instances, and ending tasks.


In a more realistic example, @StartTask and @EndTask would not appear on the same method, because there is usually work to be done using the application in order to complete the task.

Finally, the core of the application is in todo.jsp:

Example 1.18. todo.jsp


<%@ taglib uri="http://java.sun.com/jsf/html" prefix="h" %>
<%@ taglib uri="http://java.sun.com/jsf/core" prefix="f" %>
<%@ taglib uri="http://jboss.com/products/seam/taglib" prefix="s" %>
<html>
<head>
<title>Todo List</title>
</head>
<body>
<h1>Todo List</h1>
<f:view>
   <h:form id="list">
      <div>
         <h:outputText value="There are no todo items." 
                       rendered="#{empty taskInstanceList}"/>
         <h:dataTable value="#{taskInstanceList}" var="task" 
                      rendered="#{not empty taskInstanceList}">
            <h:column>
                <f:facet name="header">
                    <h:outputText value="Description"/>
                </f:facet>
                <h:inputText value="#{task.description}"/>
            </h:column>
            <h:column>
                <f:facet name="header">
                    <h:outputText value="Created"/>
                </f:facet>
                <h:outputText value="#{task.taskMgmtInstance.processInstance.start}">
                    <f:convertDateTime type="date"/>
                </h:outputText>
            </h:column>
            <h:column>
                <f:facet name="header">
                    <h:outputText value="Priority"/>
                </f:facet>
                <h:inputText value="#{task.priority}" style="width: 30"/>
            </h:column>
            <h:column>
                <f:facet name="header">
                    <h:outputText value="Due Date"/>
                </f:facet>
                <h:inputText value="#{task.dueDate}" style="width: 100">
                    <f:convertDateTime type="date" dateStyle="short"/>
                </h:inputText>
            </h:column>
            <h:column>
                <s:button value="Done" action="#{todoList.done}" taskInstance="#{task}"/>
            </h:column>
         </h:dataTable>
      </div>
      <div>
      <h:messages/>
      </div>
      <div>
         <h:commandButton value="Update Items" action="update"/>
      </div>
   </h:form>
   <h:form id="new">
      <div>
         <h:inputText value="#{todoList.description}"/>
         <h:commandButton value="Create New Item" action="#{todoList.createTodo}"/>
      </div>
   </h:form>
</f:view>
</body>
</html>

Let's take this one piece at a time.

The page renders a list of tasks, which it gets from a built-in Seam component named taskInstanceList. The list is defined inside a JSF form.


Each element of the list is an instance of the jBPM class TaskInstance. The following code simply displays the interesting properties of each task in the list. For the description, priority and due date, we use input controls, to allow the user to update these values.


<h:column>
    <f:facet name="header">
       <h:outputText value="Description"/>
    </f:facet>
    <h:inputText value="#{task.description}"/>
</h:column>
<h:column>
    <f:facet name="header">
        <h:outputText value="Created"/>
    </f:facet>
    <h:outputText value="#{task.taskMgmtInstance.processInstance.start}">
        <f:convertDateTime type="date"/>
    </h:outputText>
</h:column>
<h:column>
    <f:facet name="header">
        <h:outputText value="Priority"/>
    </f:facet>
    <h:inputText value="#{task.priority}" style="width: 30"/>
</h:column>
<h:column>
    <f:facet name="header">
        <h:outputText value="Due Date"/>
    </f:facet>
    <h:inputText value="#{task.dueDate}" style="width: 100">
        <f:convertDateTime type="date" dateStyle="short"/>
    </h:inputText>
</h:column>

Note

Seam provides a default JSF date converter for converting a string to a date (no time). Thus, the converter is not necessary for the field bound to #{task.dueDate}.

This button ends the task by calling the action method annotated @StartTask @EndTask. It passes the task id to Seam as a request parameter:


<h:column>
    <s:button value="Done" action="#{todoList.done}" taskInstance="#{task}"/>
</h:column>

Note that this is using a Seam <s:button> JSF control from the seam-ui.jar package. This button is used to update the properties of the tasks. When the form is submitted, Seam and jBPM will make any changes to the tasks persistent. There is no need for any action listener method:


<h:commandButton value="Update Items" action="update"/>

A second form on the page is used to create new items, by calling the action method annotated @CreateProcess.


<h:form id="new">
    <div>
        <h:inputText value="#{todoList.description}"/>
        <h:commandButton value="Create New Item" action="#{todoList.createTodo}"/>
    </div>
</h:form>

After logging in, todo.jsp uses the taskInstanceList component to display a table of outstanding todo items for a the current user. Initially there are none. It also presents a form to enter a new entry. When the user types the todo item and hits the "Create New Item" button, #{todoList.createTodo} is called. This starts the todo process, as defined in todo.jpdl.xml.

The process instance is created, starting in the start state and immediately transition to the todo state, where a new task is created. The task description is set based on the user's input, which was saved to #{todoList.description}. Then, the task is assigned to the current user, which was stored in the seam actor component. Note that in this example, the process has no extra process state. All the state in this example is stored in the task definition. The process and task information is stored in the database at the end of the request.

When todo.jsp is redisplayed, taskInstanceList now finds the task that was just created. The task is shown in an h:dataTable. The internal state of the task is displayed in each column: #{task.description}, #{task.priority}, #{task.dueDate}, etc... These fields can all be edited and saved back to the database.

Each todo item also has "Done" button, which calls #{todoList.done}. The todoList component knows which task the button is for because each s:button specificies taskInstance="#{task}", referring to the task for that particular line of of the table. The @StartTast and @EndTask annotations cause seam to make the task active and immediately complete the task. The original process then transitions into the done state, according to the process definition, where it ends. The state of the task and process are both updated in the database.

When todo.jsp is displayed again, the now-completed task is no longer displayed in the taskInstanceList, since that component only display active tasks for the user.

For Seam applications with relatively freeform (ad hoc) navigation, JSF/Seam navigation rules are a perfectly good way to define the page flow. For applications with a more constrained style of navigation, especially for user interfaces which are more stateful, navigation rules make it difficult to really understand the flow of the system. To understand the flow, you need to piece it together from the view pages, the actions and the navigation rules.

Seam allows you to use a jPDL process definition to define pageflow. The simple number guessing example shows how this is done.

The example is implemented using one JavaBean, three JSP pages and a jPDL pageflow definition. Let's begin with the pageflow:

Example 1.20. pageflow.jpdl.xml

<pageflow-definition 
        xmlns="http://jboss.com/products/seam/pageflow"
        xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
        xsi:schemaLocation="http://jboss.com/products/seam/pageflow 
                            http://jboss.com/products/seam/pageflow-2.2.xsd"
        name="numberGuess">
   
   <start-page(1) name="displayGuess" view-id="/numberGuess.jspx">
      <redirect/>
      <transit(2)ion name="guess" to="evaluateGuess">
         <acti(3)on expression="#{numberGuess.guess}"/>
      </transition>
      <transition name="giveup" to="giveup"/>
      <transition name="cheat" to="cheat"/>
   </start-page>
              (4)
   <decision name="evaluateGuess" expression="#{numberGuess.correctGuess}">
      <transition name="true" to="win"/>
      <transition name="false" to="evaluateRemainingGuesses"/>
   </decision>
   
   <decision name="evaluateRemainingGuesses" expression="#{numberGuess.lastGuess}">
      <transition name="true" to="lose"/>
      <transition name="false" to="displayGuess"/>
   </decision>
   
   <page name="giveup" view-id="/giveup.jspx">
      <redirect/>
      <transition name="yes" to="lose"/>
      <transition name="no" to="displayGuess"/>
   </page>
   
   <process-state name="cheat">
      <sub-process name="cheat"/>
      <transition to="displayGuess"/>
   </process-state>
   
   <page name="win" view-id="/win.jspx">
      <redirect/>
      <end-conversation/>
   </page>
   
   <page name="lose" view-id="/lose.jspx">
      <redirect/>
      <end-conversation/>
   </page>
   
</pageflow-definition>
1

The <page> element defines a wait state where the system displays a particular JSF view and waits for user input. The view-id is the same JSF view id used in plain JSF navigation rules. The redirect attribute tells Seam to use post-then-redirect when navigating to the page. (This results in friendly browser URLs.)

2

The <transition> element names a JSF outcome. The transition is triggered when a JSF action results in that outcome. Execution will then proceed to the next node of the pageflow graph, after invocation of any jBPM transition actions.

3

A transition <action> is just like a JSF action, except that it occurs when a jBPM transition occurs. The transition action can invoke any Seam component.

4

A <decision> node branches the pageflow, and determines the next node to execute by evaluating a JSF EL expression.


Here is what the pageflow looks like in the JBoss Developer Studio pageflow editor:

Now that we have seen the pageflow, it is very, very easy to understand the rest of the application!

Here is the main page of the application, numberGuess.jspx:

Example 1.21. numberGuess.jspx


<<?xml version="1.0"?>
<jsp:root xmlns:jsp="http://java.sun.com/JSP/Page" 
          xmlns:h="http://java.sun.com/jsf/html"
          xmlns:f="http://java.sun.com/jsf/core"
          xmlns:s="http://jboss.com/products/seam/taglib"
          xmlns="http://www.w3.org/1999/xhtml"
          version="2.0">
  <jsp:output doctype-root-element="html" 
              doctype-public="-//W3C//DTD XHTML 1.0 Transitional//EN"
              doctype-system="http://www.w3c.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"/>
  <jsp:directive.page contentType="text/html"/>
  <html>
  <head>
    <title>Guess a number...</title>
    <link href="niceforms.css" rel="stylesheet" type="text/css" />
    <script language="javascript" type="text/javascript" src="niceforms.js" />
  </head>
  <body>
    <h1>Guess a number...</h1>
    <f:view>
      <h:form styleClass="niceform">
        
        <div>
        <h:messages globalOnly="true"/>
        <h:outputText value="Higher!" 
               rendered="#{numberGuess.randomNumber gt numberGuess.currentGuess}"/>
        <h:outputText value="Lower!" 
               rendered="#{numberGuess.randomNumber lt numberGuess.currentGuess}"/>
        </div>
        
        <div>
        I'm thinking of a number between 
        <h:outputText value="#{numberGuess.smallest}"/> and 
        <h:outputText value="#{numberGuess.biggest}"/>. You have 
        <h:outputText value="#{numberGuess.remainingGuesses}"/> guesses.
        </div>
        
        <div>
        Your guess: 
        <h:inputText value="#{numberGuess.currentGuess}" id="inputGuess" 
                     required="true" size="3" 
                     rendered="#{(numberGuess.biggest-numberGuess.smallest) gt 20}">
          <f:validateLongRange maximum="#{numberGuess.biggest}" 
                               minimum="#{numberGuess.smallest}"/>
        </h:inputText>
        <h:selectOneMenu value="#{numberGuess.currentGuess}" 
                         id="selectGuessMenu" required="true"
                         rendered="#{(numberGuess.biggest-numberGuess.smallest) le 20 and 
                                     (numberGuess.biggest-numberGuess.smallest) gt 4}">
          <s:selectItems value="#{numberGuess.possibilities}" var="i" label="#{i}"/>
        </h:selectOneMenu>
        <h:selectOneRadio value="#{numberGuess.currentGuess}" id="selectGuessRadio" 
                          required="true"
                          rendered="#{(numberGuess.biggest-numberGuess.smallest) le 4}">
          <s:selectItems value="#{numberGuess.possibilities}" var="i" label="#{i}"/>
        </h:selectOneRadio>
        <h:commandButton value="Guess" action="guess"/>
        <s:button value="Cheat" view="/confirm.jspx"/>
        <s:button value="Give up" action="giveup"/>
        </div>
        
        <div>
        <h:message for="inputGuess" style="color: red"/>
        </div>
        
      </h:form>
    </f:view>
  </body>
  </html>
</jsp:root>

Notice how the command button names the guess transition instead of calling an action directly.

The win.jspx page is predictable:


The lose.jspx looks roughly the same, so we'll skip over it.

Finally, we'll look at the actual application code:

Example 1.23. NumberGuess.java

@Name("numberGuess")
@Scope(ScopeType.CONVERSATION)
public class NumberGuess implements Serializable {
   
   private int randomNumber;
   private Integer currentGuess;
   private int biggest;
   private int smallest;
   private int guessCount;
   private int maxGuesses;
   private boolean cheated;
   
   @Create    (1)
   public void begin()
   {
      randomNumber = new Random().nextInt(100);
      guessCount = 0;
      biggest = 100;
      smallest = 1;
   }
   
   public void setCurrentGuess(Integer guess)
   {
      this.currentGuess = guess;
   }
   
   public Integer getCurrentGuess()
   {
      return currentGuess;
   }
   
   public void guess()
   {
      if (currentGuess>randomNumber)
      {
         biggest = currentGuess - 1;
      }
      if (currentGuess<randomNumber)
      {
         smallest = currentGuess + 1;
      }
      guessCount ++;
   }
   
   public boolean isCorrectGuess()
   {
      return currentGuess==randomNumber;
   }
   
   public int getBiggest()
   {
      return biggest;
   }
   
   public int getSmallest()
   {
      return smallest;
   }
   
   public int getGuessCount()
   {
      return guessCount;
   }
   
   public boolean isLastGuess()
   {
      return guessCount==maxGuesses;
   }

   public int getRemainingGuesses() {
      return maxGuesses-guessCount;
   }

   public void setMaxGuesses(int maxGuesses) {
      this.maxGuesses = maxGuesses;
   }

   public int getMaxGuesses() {
      return maxGuesses;
   }

   public int getRandomNumber() {
      return randomNumber;
   }

   public void cheated()
   {
      cheated = true;
   }
   
   public boolean isCheat() {
      return cheated;
   }
   
   public List<Integer> getPossibilities()
   {
      List<Integer> result = new ArrayList<Integer>();
      for(int i=smallest; i<=biggest; i++) result.add(i);
      return result;
   }
   
}
1

The first time a JSP page asks for a numberGuess component, Seam will create a new one for it, and the @Create method will be invoked, allowing the component to initialize itself.


The pages.xml file starts a Seam conversation (much more about that later), and specifies the pageflow definition to use for the conversation's page flow.


As you can see, this Seam component is pure business logic! It doesn't need to know anything at all about the user interaction flow. This makes the component potentially more reuseable.

We'll step through basic flow of the application. The game starts with the numberGuess.jspx view. When the page is first displayed, the pages.xml configuration causes conversation to begin and associates the numberGuess pageflow with that conversation. The pageflow starts with a start-page tag, which is a wait state, so the numberGuess.xhtml is rendered.

The view references the numberGuess component, causing a new instance to be created and stored in the conversation. The @Create method is called, initializing the state of the game. The view displays an h:form that allows the user to edit #{numberGuess.currentGuess}.

The "Guess" button triggers the guess action. Seam defers to the pageflow to handle the action, which says that the pageflow should transition to the evaluateGuess state, first invoking #{numberGuess.guess}, which updates the guess count and highest/lowest suggestions in the numberGuess component.

The evaluateGuess state checks the value of #{numberGuess.correctGuess} and transitions either to the win or evaluatingRemainingGuesses state. We'll assume the number was incorrect, in which case the pageflow transitions to evaluatingRemainingGuesses. That is also a decision state, which tests the #{numberGuess.lastGuess} state to determine whether or not the user has more guesses. If there are more guesses (lastGuess is false), we transition back to the original displayGuess state. Finally we've reached a page state, so the associated page /numberGuess.jspx is displayed. Since the page has a redirect element, Seam sends a redirect to the the user's browser, starting the process over.

We won't follow the state any more except to note that if on a future request either the win or the lose transition were taken, the user would be taken to either the /win.jspx or /lose.jspx. Both states specify that Seam should end the conversation, tossing away all the game state and pageflow state, before redirecting the user to the final page.

The numberguess example also contains Giveup and Cheat buttons. You should be able to trace the pageflow state for both actions relatively easily. Pay particular attention to the cheat transition, which loads a sub-process to handle that flow. Although it's overkill for this application, it does demonstrate how complex pageflows can be broken down into smaller parts to make them easier to understand.

The project structure is identical to the previous one, to install and deploy this application, please refer to Section 1.1, “Using the Seam examples”. Once you've successfully started the application, you can access it by pointing your browser to http://localhost:8080/seam-booking/

The application uses six session beans for to implement the business logic for the listed features.

  • AuthenticatorAction provides the login authentication logic.

  • BookingListAction retrieves existing bookings for the currently logged in user.

  • ChangePasswordAction updates the password of the currently logged in user.

  • HotelBookingAction implements booking and confirmation functionality. This functionality is implemented as a conversation, so this is one of the most interesting classes in the application.

  • HotelSearchingAction implements the hotel search functionality.

  • RegisterAction registers a new system user.

Three entity beans implement the application's persistent domain model.

  • Hotel is an entity bean that represent a hotel

  • Booking is an entity bean that represents an existing booking

  • User is an entity bean to represents a user who can make hotel bookings

We encourage you browse the sourcecode at your pleasure. In this tutorial we'll concentrate upon one particular piece of functionality: hotel search, selection, booking and confirmation. From the point of view of the user, everything from selecting a hotel to confirming a booking is one continuous unit of work, a conversation. Searching, however, is not part of the conversation. The user can select multiple hotels from the same search results page, in different browser tabs.

Most web application architectures have no first class construct to represent a conversation. This causes enormous problems managing conversational state. Usually, Java web applications use a combination of several techniques. Some state can be transfered in the URL. What can't is either thrown into the HttpSession or flushed to the database after every request, and reconstructed from the database at the beginning of each new request.

Since the database is the least scalable tier, this often results in an utterly unacceptable lack of scalability. Added latency is also a problem, due to the extra traffic to and from the database on every request. To reduce this redundant traffic, Java applications often introduce a data (second-level) cache that keeps commonly accessed data between requests. This cache is necessarily inefficient, because invalidation is based upon an LRU policy instead of being based upon when the user has finished working with the data. Furthermore, because the cache is shared between many concurrent transactions, we've introduced a whole raft of problem's associated with keeping the cached state consistent with the database.

Now consider the state held in the HttpSession. The HttpSession is great place for true session data, data that is common to all requests that the user has with the application. However, it's a bad place to store data related to individual series of requests. Using the session of conversational quickly breaks down when dealing with the back button and multiple windows. On top of that, without careful programming, data in the HTTP Session can grow quite large, making the HTTP session difficult to cluster. Developing mechanisms to isolate session state associated with different concurrent conversations, and incorporating failsafes to ensure that conversation state is destroyed when the user aborts one of the conversations by closing a browser window or tab is not for the faint hearted. Fortunately, with Seam, you don't have to worry about that.

Seam introduces the conversation context as a first class construct. You can safely keep conversational state in this context, and be assured that it will have a well-defined lifecycle. Even better, you won't need to be continually pushing data back and forth between the application server and the database, since the conversation context is a natural cache of data that the user is currently working with.

In this application, we'll use the conversation context to store stateful session beans. There is an ancient canard in the Java community that stateful session beans are a scalability killer. This may have been true in the early days of enterprise Java, but it is no longer true today. Modern application servers have extremely sophisticated mechanisms for stateful session bean state replication. JBoss AS, for example, performs fine-grained replication, replicating only those bean attribute values which actually changed. Note that all the traditional technical arguments for why stateful beans are inefficient apply equally to the HttpSession, so the practice of shifting state from business tier stateful session bean components to the web session to try and improve performance is unbelievably misguided. It is certainly possible to write unscalable applications using stateful session beans, by using stateful beans incorrectly, or by using them for the wrong thing. But that doesn't mean you should never use them. If you remain unconvinced, Seam allows the use of POJOs instead of stateful session beans. With Seam, the choice is yours.

The booking example application shows how stateful components with different scopes can collaborate together to achieve complex behaviors. The main page of the booking application allows the user to search for hotels. The search results are kept in the Seam session scope. When the user navigates to one of these hotels, a conversation begins, and a conversation scoped component calls back to the session scoped component to retrieve the selected hotel.

The booking example also demonstrates the use of RichFaces Ajax to implement rich client behavior without the use of handwritten JavaScript.

The search functionality is implemented using a session-scope stateful session bean, similar to the one we saw in the message list example.

Example 1.25. HotelSearchingAction.java

@Stateful     (1)
@Name("hotelSearch")
@Scope(ScopeType.SESSION)
@Restrict("#{i(2)dentity.loggedIn}")
public class HotelSearchingAction implements HotelSearching
{
   
   @PersistenceContext
   private EntityManager em;
   
   private String searchString;
   private int pageSize = 10;
   private int page;
   
   @DataModel (3)
   private List<Hotel> hotels;
   
   public void find()
   {
      page = 0;
      queryHotels();
   }
   public void nextPage()
   {
      page++;
      queryHotels();
   }
      
   private void queryHotels()
   {
      hotels = 
          em.createQuery("select h from Hotel h where lower(h.name) like #{pattern} " + 
                         "or lower(h.city) like #{pattern} " + 
                         "or lower(h.zip) like #{pattern} " +
                         "or lower(h.address) like #{pattern}")
            .setMaxResults(pageSize)
            .setFirstResult( page * pageSize )
            .getResultList();
   }
   
   public boolean isNextPageAvailable()
   {
      return hotels!=null && hotels.size()==pageSize;
   }
   
   public int getPageSize() {
      return pageSize;
   }
   
   public void setPageSize(int pageSize) {
      this.pageSize = pageSize;
   }
   
   @Factory(value="pattern", scope=ScopeType.EVENT)
   public String getSearchPattern()
   {
      return searchString==null ? 
            "%" : '%' + searchString.toLowerCase().replace('*', '%') + '%';
   }
   
   public String getSearchString()
   {
      return searchString;
   }
   
   public void setSearchString(String searchString)
   {
      this.searchString = searchString;
   }
              (4)
   @Remove
   public void destroy() {}
}
1

The EJB standard @Stateful annotation identifies this class as a stateful session bean. Stateful session beans are scoped to the conversation context by default.

2

The @Restrict annotation applies a security restriction to the component. It restricts access to the component allowing only logged-in users. The security chapter explains more about security in Seam.

3

The @DataModel annotation exposes a List as a JSF ListDataModel. This makes it easy to implement clickable lists for search screens. In this case, the list of hotels is exposed to the page as a ListDataModel in the conversation variable named hotels.

4

The EJB standard @Remove annotation specifies that a stateful session bean should be removed and its state destroyed after invocation of the annotated method. In Seam, all stateful session beans must define a method with no parameters marked @Remove. This method will be called when Seam destroys the session context.


The main page of the application is a Facelets page. Let's look at the fragment which relates to searching for hotels:

Example 1.26. main.xhtml

<div class="section">
  
    <span class="errors">
       <h:messages globalOnly="true"/>
    </span>
    
    <h1>Search Hotels</h1>

    <h:form id="searchCriteria">
    <fieldset> 
       <h:inputText id="searchString" value="#{hotelSearch.searchString}" 
                    style="width: 165px;">
         <a:support event="onkeyup" actionListener="#{hotelSearch.find}" 
              (1)      reRender="searchResults" />
       </h:inputText>
       &#160;
       <a:commandButton id="findHotels" value="Find Hotels" action="#{hotelSearch.find}" 
                        reRender="searchResults"/>
       &#160;
       <a:stat(2)us>
          <f:facet name="start">
             <h:graphicImage value="/img/spinner.gif"/>
          </f:facet>
       </a:status>
       <br/>
       <h:outputLabel for="pageSize">Maximum results:</h:outputLabel>&#160;
       <h:selectOneMenu value="#{hotelSearch.pageSize}" id="pageSize">
          <f:selectItem itemLabel="5" itemValue="5"/>
          <f:selectItem itemLabel="10" itemValue="10"/>
          <f:selectItem itemLabel="20" itemValue="20"/>
       </h:selectOneMenu>
    </fieldset>
    </h:form>
    
</div>

<a:outputPanel(3) id="searchResults">
  <div class="section">
    <h:outputText value="No Hotels Found"
                  rendered="#{hotels != null and hotels.rowCount==0}"/>
    <h:dataTable id="hotels" value="#{hotels}" var="hot" 
                 rendered="#{hotels.rowCount>0}">
        <h:column>
            <f:facet name="header">Name</f:facet>
            #{hot.name}
        </h:column>
        <h:column>
            <f:facet name="header">Address</f:facet>
            #{hot.address}
        </h:column>
        <h:column>
            <f:facet name="header">City, State</f:facet>
            #{hot.city}, #{hot.state}, #{hot.country}
        </h:column> 
        <h:column>
            <f:facet name="header">Zip</f:facet>
            #{hot.zip}
        </h:column>
        <h:column>
            <f:facet name="header">Action</f:facet>
            <s(4):link id="viewHotel" value="View Hotel" 
                    action="#{hotelBooking.selectHotel(hot)}"/>
        </h:column>
    </h:dataTable>
    <s:link value="More results" action="#{hotelSearch.nextPage}" 
            rendered="#{hotelSearch.nextPageAvailable}"/>
  </div>
</a:outputPanel>    
1

The RichFaces Ajax <a:support> tag allows a JSF action event listener to be called by asynchronous XMLHttpRequest when a JavaScript event like onkeyup occurs. Even better, the reRender attribute lets us render a fragment of the JSF page and perform a partial page update when the asynchronous response is received.

2

The RichFaces Ajax <a:status> tag lets us display an animated image while we wait for asynchronous requests to return.

3

The RichFaces Ajax <a:outputPanel> tag defines a region of the page which can be re-rendered by an asynchronous request.

4

The Seam <s:link> tag lets us attach a JSF action listener to an ordinary (non-JavaScript) HTML link. The advantage of this over the standard JSF <h:commandLink> is that it preserves the operation of "open in new window" and "open in new tab". Also notice that we use a method binding with a parameter: #{hotelBooking.selectHotel(hot)}. This is not possible in the standard Unified EL, but Seam provides an extension to the EL that lets you use parameters on any method binding expression.

If you're wondering how navigation occurs, you can find all the rules in WEB-INF/pages.xml; this is discussed in Section 6.7, “Navigation”.


This page displays the search results dynamically as we type, and lets us choose a hotel and pass it to the selectHotel() method of the HotelBookingAction, which is where the really interesting stuff is going to happen.

Now let's see how the booking example application uses a conversation-scoped stateful session bean to achieve a natural cache of persistent data related to the conversation. The following code example is pretty long. But if you think of it as a list of scripted actions that implement the various steps of the conversation, it's understandable. Read the class from top to bottom, as if it were a story.

Example 1.27. HotelBookingAction.java

@Stateful
@Name("hotelBooking")
@Restrict("#{identity.loggedIn}")
public class HotelBookingAction implements HotelBooking
{
   
   @Persistenc(1)eContext(type=EXTENDED)
   private EntityManager em;
   
   @In 
   private User user;
   
   @In(required=false) @Out
   private Hotel hotel;
   
   @In(required=false) 
   @Out(requir(2)ed=false)
   private Booking booking;
     
   @In
   private FacesMessages facesMessages;
      
   @In
   private Events events;
   
   @Logger 
   private Log log;
   
   private boolean bookingValid;
   
   @Begin     (3)
   public void selectHotel(Hotel selectedHotel)
   {
      hotel = em.merge(selectedHotel);
   }
   
   public void bookHotel()
   {      
      booking = new Booking(hotel, user);
      Calendar calendar = Calendar.getInstance();
      booking.setCheckinDate( calendar.getTime() );
      calendar.add(Calendar.DAY_OF_MONTH, 1);
      booking.setCheckoutDate( calendar.getTime() );
   }
   
   public void setBookingDetails()
   {
      Calendar calendar = Calendar.getInstance();
      calendar.add(Calendar.DAY_OF_MONTH, -1);
      if ( booking.getCheckinDate().before( calendar.getTime() ) )
      {
         facesMessages.addToControl("checkinDate", "Check in date must be a future date");
         bookingValid=false;
      }
      else if ( !booking.getCheckinDate().before( booking.getCheckoutDate() ) )
      {
         facesMessages.addToControl("checkoutDate", 
                                    "Check out date must be later than check in date");
         bookingValid=false;
      }
      else
      {
         bookingValid=true;
      }
   }
   
   public boolean isBookingValid()
   {
      return bookingValid;
   }
   
   @End       (4)
   public void confirm()
   {
      em.persist(booking);
      facesMessages.add("Thank you, #{user.name}, your confimation number " + 
                        " for #{hotel.name} is #{booki g.id}");
      log.info("New booking: #{booking.id} for #{user.username}");
      events.raiseTransactionSuccessEvent("bookingConfirmed");
   }
   
   @End
   public void cancel() {}
   
   @Remove    (5)
   public void destroy() {}
1

This bean uses an EJB3 extended persistence context, so that any entity instances remain managed for the whole lifecycle of the stateful session bean.

2

The @Out annotation declares that an attribute value is outjected to a context variable after method invocations. In this case, the context variable named hotel will be set to the value of the hotel instance variable after every action listener invocation completes.

3

The @Begin annotation specifies that the annotated method begins a long-running conversation, so the current conversation context will not be destroyed at the end of the request. Instead, it will be reassociated with every request from the current window, and destroyed either by timeout due to conversation inactivity or invocation of a matching @End method.

4

The @End annotation specifies that the annotated method ends the current long-running conversation, so the current conversation context will be destroyed at the end of the request.

5

This EJB remove method will be called when Seam destroys the conversation context. Don't forget to define this method!


HotelBookingAction contains all the action listener methods that implement selection, booking and booking confirmation, and holds state related to this work in its instance variables. We think you'll agree that this code is much cleaner and simpler than getting and setting HttpSession attributes.

Even better, a user can have multiple isolated conversations per login session. Try it! Log in, run a search, and navigate to different hotel pages in multiple browser tabs. You'll be able to work on creating two different hotel reservations at the same time. If you leave any one conversation inactive for long enough, Seam will eventually time out that conversation and destroy its state. If, after ending a conversation, you backbutton to a page of that conversation and try to perform an action, Seam will detect that the conversation was already ended, and redirect you to the search page.

Long-running conversations make it simple to maintain consistency of state in an application even in the face of multi-window operation and back-buttoning. Unfortunately, simply beginning and ending a long-running conversation is not always enough. Depending on the requirements of the application, inconsistencies between what the user's expectations and the reality of the application’s state can still result.

The nested booking application extends the features of the hotel booking application to incorporate the selection of rooms. Each hotel has available rooms with descriptions for a user to select from. This requires the addition of a room selection page in the hotel reservation flow.

The user now has the option to select any available room to be included in the booking. As with the hotel booking application we saw previously, this can lead to issues with state consistency. As with storing state in the HTTPSession, if a conversation variable changes it affects all windows operating within the same conversation context.

To demonstrate this, let’s suppose the user clones the room selection screen in a new window. The user then selects the Wonderful Room and proceeds to the confirmation screen. To see just how much it would cost to live the high-life, the user returns to the original window, selects the Fantastic Suite for booking, and again proceeds to confirmation. After reviewing the total cost, the user decides that practicality wins out and returns to the window showing Wonderful Room to confirm.

In this scenario, if we simply store all state in the conversation, we are not protected from multi-window operation within the same conversation. Nested conversations allow us to achieve correct behavior even when context can vary within the same conversation.

Now let's see how the nested booking example extends the behavior of the hotel booking application through use of nested conversations. Again, we can read the class from top to bottom, as if it were a story.

Example 1.28. RoomPreferenceAction.java

@Stateful
@Name("roomPreference")
@Restrict("#{identity.loggedIn}")
public class RoomPreferenceAction implements RoomPreference 
{

   @Logger 
   private Log log;

   @In private Hotel hotel;
   
   @In private Booking booking;

   @DataModel(value="availableRooms")
   private List<Room> availableRooms;

   @DataModelSelection(value="availableRooms")
   private Room roomSelection;
    
   @In(required=false, value="roomSelection")
   @Out(required=false, value="roomSelection")
   private Room room;

   @Factory("availableRooms")
   public void(1) loadAvailableRooms()
   {
      availableRooms = hotel.getAvailableRooms(booking.getCheckinDate(), booking.getCheckoutDate());
      log.info("Retrieved #0 available rooms", availableRooms.size());
   }

   public BigDecimal getExpectedPrice()
   {
      log.info("Retrieving price for room #0", roomSelection.getName());
      
      return booking.getTotal(roomSelection);
   }

   @Begin(nest(2)ed=true)
   public String selectPreference()
   {
      log.info("Room selected");
      
      this.roo(3)m = this.roomSelection;
      
      return "payment";
   }

   public String requestConfirmation()
   {
      // all validations are performed through the s:validateAll, so checks are already
      // performed
      log.info("Request confirmation from user");
      
      return "confirm";
   }

   @End(beforeRedirect=true)
   public Stri(4)ng cancel()
   {
      log.info("ending conversation");

      return "cancel";
   }

   @Destroy @Remove                                                                      
   public void destroy() {}    
}
1

The hotel instance is injected from the conversation context. The hotel is loaded through an extended persistence context so that the entity remains managed throughout the conversation. This allows us to lazily load the availableRooms through an @Factory method by simply walking the association.

2

When @Begin(nested=true) is encountered, a nested conversation is pushed onto the conversation stack. When executing within a nested conversation, components still have access to all outer conversation state, but setting any values in the nested conversation’s state container does not affect the outer conversation. In addition, nested conversations can exist concurrently stacked on the same outer conversation, allowing independent state for each.

3

The roomSelection is outjected to the conversation based on the @DataModelSelection. Note that because the nested conversation has an independent context, the roomSelection is only set into the new nested conversation. Should the user select a different preference in another window or tab a new nested conversation would be started.

4

The @End annotation pops the conversation stack and resumes the outer conversation. The roomSelection is destroyed along with the conversation context.


When we begin a nested conversation it is pushed onto the conversation stack. In the nestedbooking example, the conversation stack consists of the outer long-running conversation (the booking) and each of the nested conversations (room selections).

Example 1.29. rooms.xhtml

<div class="section">
    <h1>Room Preference</h1>
</div>

<div class="section">
    <h:form id="room_selections_form">
        <div class="section">
            <h:outputText styleClass="output" 
                value="No rooms available for the dates selected: " 
                rendered="#{availableRooms != null and availableRooms.rowCount == 0}"/>
            <h:outputText styleClass="output" 
                value="Rooms available for the dates selected: " 
                rendered="#{availableRooms != null and availableRooms.rowCount > 0}"/>
                
            <h:outputText styleClass="output" value="#{booking.checkinDate}"/> -
            <h:outputText styleClass="output" value="#{booking.checkoutDate}"/>
            
            <br/><br/>
              (1)
            <h:dataTable value="#{availableRooms}" var="room" 
                    rendered="#{availableRooms.rowCount > 0}">
                <h:column>
                    <f:facet name="header">Name</f:facet>
                    #{room.name}
                </h:column>
                <h:column>
                    <f:facet name="header">Description</f:facet>
                    #{room.description}
                </h:column>
                <h:column>
                    <f:facet name="header">Per Night</f:facet>
                    <h:outputText value="#{room.price}">
                        <f:convertNumber type="currency" currencySymbol="$"/>
                    </h:outputText>
                </h:column>
              (2)  <h:column>
                    <f:facet name="header">Action</f:facet>
                    <h:commandLink id="selectRoomPreference" 
                        action="#{roomPreference.selectPreference}">Select</h:commandLink>
                </h:column>
            </h:dataTable>
        </div>
        <div class="entry">
            <div class="label">&#160;</div>
            <d(3)iv class="input">
                <s:button id="cancel" value="Revise Dates" view="/book.xhtml"/>
            </div>
        </div>    
    </h:form>
</div>
1

When requested from EL, the #{availableRooms} are loaded by the @Factory method defined in RoomPreferenceAction. The @Factory method will only be executed once to load the values into the current context as a @DataModel instance.

2

Invoking the #{roomPreference.selectPreference} action results in the row being selected and set into the @DataModelSelection. This value is then outjected to the nested conversation context.

3

Revising the dates simply returns to the /book.xhtml. Note that we have not yet nested a conversation (no room preference has been selected), so the current conversation can simply be resumed. The <s:button> component simply propagates the current conversation when displaying the /book.xhtml view.


Now that we have seen how to nest a conversation, let's see how we can confirm the booking once a room has been selected. This can be achieved by simply extending the behavior of the HotelBookingAction.

Example 1.30. HotelBookingAction.java

@Stateful
@Name("hotelBooking")
@Restrict("#{identity.loggedIn}")
public class HotelBookingAction implements HotelBooking
{
   
   @PersistenceContext(type=EXTENDED)
   private EntityManager em;
   
   @In 
   private User user;
   
   @In(required=false) @Out
   private Hotel hotel;
   
   @In(required=false) 
   @Out(required=false)
   private Booking booking;
   
   @In(required=false)
   private Room roomSelection;
   
   @In
   private FacesMessages facesMessages;
      
   @In
   private Events events;
   
   @Logger 
   private Log log;
   
   @Begin
   public void selectHotel(Hotel selectedHotel)
   {
      log.info("Selected hotel #0", selectedHotel.getName());
      hotel = em.merge(selectedHotel);
   }
   
   public String setBookingDates()
   {
      // the result will indicate whether or not to begin the nested conversation
      // as well as the navigation.  if a null result is returned, the nested
      // conversation will not begin, and the user will be returned to the current
      // page to fix validation issues
      String result = null;

      Calendar calendar = Calendar.getInstance();
      calendar.add(Calendar.DAY_OF_MONTH, -1);

      // validate what we have received from the user so far
      if ( booking.getCheckinDate().before( calendar.getTime() ) )
      {
         facesMessages.addToControl("checkinDate", "Check in date must be a future date");
      }
      else if ( !booking.getCheckinDate().before( booking.getCheckoutDate() ) )
      {
         facesMessages.addToControl("checkoutDate", "Check out date must be later than check in date");
      }
      else
      {
         result = "rooms";
      }

      return result;
   }
   
   public void bookHotel()
   {      
      booking = new Booking(hotel, user);
      Calendar calendar = Calendar.getInstance();
      booking.setCheckinDate( calendar.getTime() );
      calendar.add(Calendar.DAY_OF_MONTH, 1);
      booking.setCheckoutDate( calendar.getTime() );
   }
   
   @End(root=true)
   public void(1) confirm()
   {
      // on confirmation we set the room preference in the booking.  the room preference
      // will be injected based on the nested conversation we are in.
      booking.setRoomPreference(roomSelection);
              (2)
      em.persist(booking);
      facesMessages.add("Thank you, #{user.name}, your confimation number for #{hotel.name} is #{booking.id}");
      log.info("New booking: #{booking.id} for #{user.username}");
      events.raiseTransactionSuccessEvent("bookingConfirmed");
   }
   
   @End(root=t(3)rue, beforeRedirect=true)
   public void cancel() {}
   
   @Destroy @Remove
   public void destroy() {}
}
1

Annotating an action with @End(root=true) ends the root conversation which effectively destroys the entire conversation stack. When any conversation is ended, its nested conversations are ended as well. As the root is the conversation that started it all, this is a simple way to destroy and release all state associated with a workspace once the booking is confirmed.

2

The roomSelection is only associated with the booking on user confirmation. While outjecting values to the nested conversation context will not impact the outer conversation, any objects injected from the outer conversation are injected by reference. This means that any changing to these objects will be reflected in the parent conversation as well as other concurrent nested conversations.

3

By simply annotating the cancellation action with @End(root=true, beforeRedirect=true) we can easily destroy and release all state associated with the workspace prior to redirecting the user back to the hotel selection view.


Feel free to deploy the application, open many windows or tabs and attempt combinations of various hotels with various room preferences. Confirming a booking always results in the correct hotel and room preference thanks to the nested conversation model.

The DVD Store demo application shows the practical usage of jBPM for both task management and pageflow.

The user screens take advantage of a jPDL pageflow to implement searching and shopping cart functionality.

The administration screens take use jBPM to manage the approval and shipping cycle for orders. The business process may even be changed dynamically, by selecting a different process definition!

The Seam DVD Store demo can be run from dvdstore directory, just like the other demo applications.

Seam makes it very easy to implement applications which keep state on the server-side. However, server-side state is not always appropriate, especially in for functionality that serves up content. For this kind of problem we often want to keep application state in the URL so that any page can be accessed at any time through a bookmark. The blog example shows how to a implement an application that supports bookmarking throughout, even on the search results page. This example demonstrates how Seam can manage application state in the URL as well as how Seam can rewrite those URLs to be even

The Blog example demonstrates the use of "pull"-style MVC, where instead of using action listener methods to retrieve data and prepare the data for the view, the view pulls data from components as it is being rendered.

This snippet from the index.xhtml facelets page displays a list of recent blog entries:


If we navigate to this page from a bookmark, how does the #{blog.recentBlogEntries} data used by the <h:dataTable> actually get initialized? The Blog is retrieved lazily — "pulled" — when needed, by a Seam component named blog. This is the opposite flow of control to what is used in traditional action-based web frameworks like Struts.


This is good so far, but what about bookmarking the result of form submissions, such as a search results page?

The blog example has a tiny form in the top right of each page that allows the user to search for blog entries. This is defined in a file, menu.xhtml, included by the facelets template, template.xhtml:


Then the form would have looked like this:


<div id="search">
   <h:form>
      <h:inputText value="#{searchAction.searchPattern}"/>
      <h:commandButton value="Search" action="searchResults"/>
   </h:form>
</div>

But when we redirect, we need to include the values submitted with the form in the URL to get a bookmarkable URL like http://localhost:8080/seam-blog/search/. JSF does not provide an easy way to do this, but Seam does. We use two Seam features to accomplish this: page parameters and URL rewriting. Both are defined in WEB-INF/pages.xml:


The page parameter instructs Seam to link the request parameter named searchPattern to the value of #{searchService.searchPattern}, both whenever a request for the Search page comes in and whenever a link to the search page is generated. Seam takes responsibility for maintaining the link between URL state and application state, and you, the developer, don't have to worry about it.

Without URL rewriting, the URL for a search on the term book would be http://localhost:8080/seam-blog/seam/search.xhtml?searchPattern=book. This is nice, but Seam can make the URL even simpler using a rewrite rule. The first rewrite rule, for the pattern /search/{searchPattern}, says that any time we have a URL for search.xhtml with a searchPattern request parameter, we can fold that URL into the simpler URL. So,the URL we saw earlier, http://localhost:8080/seam-blog/seam/search.xhtml?searchPattern=book can be written instead as http://localhost:8080/seam-blog/search/book.

Just like with page parameters, URL rewriting is bi-directional. That means that Seam forwards requests for the simpler URL to the the right view, and it also automatically generates the simpler view for you. You never need to worry about constructing URLs. It's all handled transparently behind the scenes. The only requirement is that to use URL rewriting, the rewrite filter needs to be enabled in components.xml.

<web:rewrite-filter view-mapping="/seam/*" />

The redirect takes us to the search.xhtml page:


<h:dataTable value="#{searchResults}" var="blogEntry">
  <h:column>
     <div>
        <s:link view="/entry.xhtml" propagation="none" value="#{blogEntry.title}">
           <f:param name="blogEntryId" value="#{blogEntry.id}"/>
        </s:link>
        posted on 
        <h:outputText value="#{blogEntry.date}">
            <f:convertDateTime timeZone="#{blog.timeZone}" locale="#{blog.locale}" type="both"/>
        </h:outputText>
     </div>
  </h:column>
</h:dataTable>

Which again uses "pull"-style MVC to retrieve the actual search results using Hibernate Search.

@Name("searchService")

public class SearchService 
{
   
   @In
   private FullTextEntityManager entityManager;
   
   private String searchPattern;
   
   @Factory("searchResults")
   public List<BlogEntry> getSearchResults()
   {
      if (searchPattern==null || "".equals(searchPattern) ) {
         searchPattern = null;
         return entityManager.createQuery("select be from BlogEntry be order by date desc").getResultList();
      }
      else
      {
         Map<String,Float> boostPerField = new HashMap<String,Float>();
         boostPerField.put( "title", 4f );
         boostPerField.put( "body", 1f );
         String[] productFields = {"title", "body"};
         QueryParser parser = new MultiFieldQueryParser(productFields, new StandardAnalyzer(), boostPerField);
         parser.setAllowLeadingWildcard(true);
         org.apache.lucene.search.Query luceneQuery;
         try
         {
            luceneQuery = parser.parse(searchPattern);
         }
         catch (ParseException e)
         {
            return null;
         }
         return entityManager.createFullTextQuery(luceneQuery, BlogEntry.class)
               .setMaxResults(100)
               .getResultList();
      }
   }
   public String getSearchPattern()
   {
      return searchPattern;
   }
   public void setSearchPattern(String searchPattern)
   {
      this.searchPattern = searchPattern;
   }
}

Very occasionally, it makes more sense to use push-style MVC for processing RESTful pages, and so Seam provides the notion of a page action. The Blog example uses a page action for the blog entry page, entry.xhtml. Note that this is a little bit contrived, it would have been easier to use pull-style MVC here as well.

The entryAction component works much like an action class in a traditional push-MVC action-oriented framework like Struts:

@Name("entryAction")

@Scope(STATELESS)
public class EntryAction
{
   @In Blog blog;
   
   @Out BlogEntry blogEntry;
   
   public void loadBlogEntry(String id) throws EntryNotFoundException
   {
      blogEntry = blog.getBlogEntry(id);
      if (blogEntry==null) throw new EntryNotFoundException(id);
   }
   
}

Page actions are also declared in pages.xml:


<pages>
   ...

    <page view-id="/entry.xhtml"> 
        <rewrite pattern="/entry/{blogEntryId}" />
        <rewrite pattern="/entry" />
        
        <param name="blogEntryId" 
               value="#{blogEntry.id}"/>
        
        <action execute="#{entryAction.loadBlogEntry(blogEntry.id)}"/>
    </page>
    
    <page view-id="/post.xhtml" login-required="true">
        <rewrite pattern="/post" />
        
        <action execute="#{postAction.post}"
                if="#{validation.succeeded}"/>
        
        <action execute="#{postAction.invalid}"
                if="#{validation.failed}"/>
        
        <navigation from-action="#{postAction.post}">
            <redirect view-id="/index.xhtml"/>
        </navigation>
    </page>

    <page view-id="*">
        <action execute="#{blog.hitCount.hit}"/>
    </page>

</pages>

Notice that the example is using page actions for post validation and the pageview counter. Also notice the use of a parameter in the page action method binding. This is not a standard feature of JSF EL, but Seam lets you use it, not just for page actions but also in JSF method bindings.

When the entry.xhtml page is requested, Seam first binds the page parameter blogEntryId to the model. Keep in mind that because of the URL rewriting, the blogEntryId parameter name won't show up in the URL. Seam then runs the page action, which retrieves the needed data — the blogEntry — and places it in the Seam event context. Finally, the following is rendered:


<div class="blogEntry">
    <h3>#{blogEntry.title}</h3>
    <div>
        <s:formattedText value="#{blogEntry.body}"/>
    </div>
    <p>
    [Posted on&#160;
    <h:outputText value="#{blogEntry.date}">
       <f:convertDateTime timeZone="#{blog.timeZone}" locale="#{blog.locale}" type="both"/>
    </h:outputText>]
    </p>
</div>

If the blog entry is not found in the database, the EntryNotFoundException exception is thrown. We want this exception to result in a 404 error, not a 505, so we annotate the exception class:

@ApplicationException(rollback=true)

@HttpError(errorCode=HttpServletResponse.SC_NOT_FOUND)
public class EntryNotFoundException extends Exception
{
   EntryNotFoundException(String id)
   {
      super("entry not found: " + id);
   }
}

An alternative implementation of the example does not use the parameter in the method binding:

@Name("entryAction")

@Scope(STATELESS)
public class EntryAction
{
   @In(create=true) 
   private Blog blog;
   
   @In @Out
   private BlogEntry blogEntry;
   
   public void loadBlogEntry() throws EntryNotFoundException
   {
      blogEntry = blog.getBlogEntry( blogEntry.getId() );
      if (blogEntry==null) throw new EntryNotFoundException(id);
   }
}

<pages>
   ...

   <page view-id="/entry.xhtml" action="#{entryAction.loadBlogEntry}">
      <param name="blogEntryId" value="#{blogEntry.id}"/>
   </page>
   
   ...
</pages>

It is a matter of taste which implementation you prefer.

The blog demo also demonstrates very simple password authentication, posting to the blog, page fragment caching and atom feed generation.