Friday, August 24, 2007

Struts2 + Spring + JUnit

Hopefully this entry serves as some search engine friendly documentation on how one might unit test Struts 2 actions configured using Spring, something I would think many, many people want to do. This used to be done using StrutsTestCase in the Struts 1.x days but Webwork/Struts provides enough flexibility in its architecture to accommodate unit testing fairly easily. I’m not going to go over how the Spring configuration is setup. I’m assuming you have a struts.xml file which has actions configured like this:

<struts>
<package namespace="/site" extends="struts-default">
<action name="deletePerson" class="personAction"
method="deletePerson">
<result name="success">/WEB-INF/pages/person.jsp</result>
</action>
</package>
...
</struts>

You also might have an applicationContext.xml file where you might define your Spring beans like this.

<beans>
<bean id="personAction"
class="com.arsenalist.action.PersonAction"/>
...
</beans>

Then of course you also need to have an action which you want to test which might look something like:

public class PersonAction extend ActionSupport { 

private int id;

public int getId() {
return id;
}
public void setId(int id) {
this.id = id;
}
public String deletePerson() {
....
return SUCCESS;
}
}

Remember than in Struts 2, an action is usually called before and after various other interceptors are invoked. Interceptor configuration is usually specified in the struts.xml file. At this point we need to cover three different methods of how you might want to call your actions.



  1. Specify request parameters which are translated and mapped to the actions domain objects (id in the PersonAction class) and then execute the action while also executing all configured interceptors.
  2. Instead of specifying request parameters, directly specify the values of the domain objects and then execute the action while also executing all configured interceptors.
  3. Finally, you just might want to execute the action and not worry about executing the interceptors. Here you’ll specify the values of the actions domain objects and then execute the action.

Depending on what you’re testing and what scenario you want to reproduce, you should pick the one that suits the case. There’s an example of all three cases below. The best way I find to test all your action classes is to have one base class which sets up the Struts 2 environment and then your action test classes can extend it. Here’s a class that could be used as one of those base classes.


See the comments for a little more detail about whats going on. One point to note is that the class being extended here is junit.framework.TestCase and not org.apache.struts2.StrutsTestCase as one might expect. The reason for this is that StrutsTestCase is not really a well written class and does not provide enough flexibility in how we want the very core Dispatcher object to be created. Also, the interceptor example shown in the Struts documentation does not compile as there seems to have been some sort of API change. It’s been fixed in this example.

public class BaseStrutsTestCase extends TestCase {

private Dispatcher dispatcher;
protected ActionProxy proxy;
protected MockServletContext servletContext;
protected MockHttpServletRequest request;
protected MockHttpServletResponse response;

/**
* Created action class based on namespace and name
*/
protected T createAction(Class clazz, String namespace, String name)
throws Exception {

// create a proxy class which is just a wrapper around the action call.
// The proxy is created by checking the namespace and name against the
// struts.xml configuration
proxy = dispatcher.getContainer().getInstance(ActionProxyFactory.class).
createActionProxy(
namespace, name, null, true, false);

// set to true if you want to process Freemarker or JSP results
proxy.setExecuteResult(false);
// by default, don't pass in any request parameters
proxy.getInvocation().getInvocationContext().
setParameters(new HashMap());

// set the actions context to the one which the proxy is using
ServletActionContext.setContext(
proxy.getInvocation().getInvocationContext());
request = new MockHttpServletRequest();
response = new MockHttpServletResponse();
ServletActionContext.setRequest(request);
ServletActionContext.setResponse(response);
ServletActionContext.setServletContext(servletContext);
return (T) proxy.getAction();
}

protected void setUp() throws Exception {
String[] config = new String[] { "META-INF/applicationContext-aws.xml" };

// Link the servlet context and the Spring context
servletContext = new MockServletContext();
XmlWebApplicationContext appContext = new XmlWebApplicationContext();
appContext.setServletContext(servletContext);
appContext.setConfigLocations(config);
appContext.refresh();
servletContext.setAttribute(WebApplicationContext.
ROOT_WEB_APPLICATION_CONTEXT_ATTRIBUTE, appContext);

// Use spring as the object factory for Struts
StrutsSpringObjectFactory ssf = new StrutsSpringObjectFactory(
null, null, servletContext);
ssf.setApplicationContext(appContext);
//ssf.setServletContext(servletContext);
StrutsSpringObjectFactory.setObjectFactory(ssf);

// Dispatcher is the guy that actually handles all requests. Pass in
// an empty Map as the parameters but if you want to change stuff like
// what config files to read, you need to specify them here
// (see Dispatcher's source code)
dispatcher = new Dispatcher(servletContext,
new HashMap());
dispatcher.init();
Dispatcher.setInstance(dispatcher);
}
}

By extending the above class for our action test classes we can easily simulate any of the three scenarios listed above. I’ve added three methods to PersonActionTest which illustrate how to test the above three cases: testInterceptorsBySettingRequestParameters, testInterceptorsBySettingDomainObjects() and testActionAndSkipInterceptors(), respectively.

public class PersonActionTest extends BaseStrutsTestCase { 

/**
* Invoke all interceptors and specify value of the action
* class' domain objects directly.
* @throws Exception Exception
*/
public void testInterceptorsBySettingDomainObjects()
throws Exception {
PersonAction action = createAction(PersonAction.class,
"/site", "deletePerson");
pa.setId(123);
String result = proxy.execute();
assertEquals(result, "success");
}

/**
* Invoke all interceptors and specify value of action class'
* domain objects through request parameters.
* @throws Exception Exception
*/
public void testInterceptorsBySettingRequestParameters()
throws Exception {
createAction(PersonAction.class, "/site", "deletePerson");
Map params = new HashMap();
params.put("id", "123");
proxy.getInvocation().getInvocationContext().setParameters(params);
String result = proxy.execute();
assertEquals(result, "success");
}

/**
* Skip interceptors and specify value of action class'
* domain objects by setting them directly.
* @throws Exception Exception
*/
public void testActionAndSkipInterceptors() throws Exception {
PersonAction action = createAction(PersonAction.class,
"/site", "deletePerson");
action.setId(123);
String result = action.deletePerson();
assertEquals(result, "success");
}
}

The source code for Dispatcher is probably a good thing to look at if you want to configure your actions more specifically. There are options to specify zero-configuration, alternate XML files and others. Ideally the StrutsTestCaseHelper should be doing a lot more than what it does right now (creating a badly configured Dispatcher) and should allow creation of custom dispatchers and object factories. That’s the reason why I’m not using StrutsTestCase since all that does is make a couple calls using StrutsTestCaseHelper.


If you want to test your validation, its pretty easy. Here’s a snippet of code that might do that:

 public void testValidation() throws Exception {
SomeAction action = createAction(SomeAction.class,
"/site", "someAction");
// lets forget to set a required field: action.setId(123);
String result = proxy.invoke();
assertEquals(result, "input");
assertTrue("Must have one field error",
action.getFieldErrors().size() == 1);
}

This example uses Struts 2.0.8 and Spring 2.0.5.

Thursday, August 2, 2007

Familiarity with Stripes

Taken from the official site Stipes.


Stripes is a presentation framework for building web applications using the latest Java technologies. The main driver behind Stripes is that web application development in Java is just too much work! It seems like every existing framework requires gobs of configuration. Struts is pretty feature-light and has some serious architectural issues. Others, like WebWork 2 and Spring-MVC are much better, but still require a lot of configuration, and seem to require you to learn a whole new language just to get started.

Stripes was born out of ongoing frustration with the lack of a high quality, easy to use web framework. Sure, Struts has its good points, but there are a lot of small things that really add up. A lot of small things that you learn to work around, and live with, without realizing how unproductive it's making you.

Up until recently it would have been difficult to create a framework that was better-enough to warrant competing with Struts. And with JSF on the horizon (perpetually?), and other web frameworks in play (WebWork, Tapestry, Wicket) some might question the rationale behind yet another framework. But with Java 1.5 and Servlet 2.4 I think the time has come. The rationale is plain and simple - I wanted a web framework that made it easy, no, fun, to write web applications in Java. The best way to demonstrate is perhaps by a comparison with Struts.

Number of artifacts

One of prime frustrations with Struts is the fact that just to implement a single page/form, I have to write or edit so many files. And I have to keep them in sync, or else things start going horribly wrong. With Struts I have to write JSP, Action, Form, a form-bean stanza in the struts-config.xml, an action stanza in the struts-config.xml, and if I'm going to do it the Struts way, a bunch of forward stanzas. And let's not go into the fact that since this is all stored in one xml file I'm continually facing merge conflicts with team mates. Yes, there are annotations for Struts, but they are just literal translations of what's in the XML file, and they don't feel natural to me.

Compare this with Stripes. I write JSP. I write ActionBean and annotate it with a @UrlBinding to specify the URL it should respond to, and one or more @HandlesEvent annotations to map events to methods. I'm done. All the information about the form and the action is in the same place.

Incremental development

Write a JSP with a Struts <html:form> tag on it and try to preview it in your container before you write your form-bean. Boom! You can't. Exception! Could not find form bean! Now write your form bean. Wait - that's still not good enough. Since everything is hooked together through the action stanza in the XML you better have an action, and edit that too.

Maybe it's just me, but I like to be able to write JSP, see if it looks ok, then go and write the back end components to go with it. It's not brain surgery, and Stripes lets you do it.

Property binding

Struts lets you use nested properties, and that's great. But say you have a property on your Form 'person.address.line1'. Struts will try and set it for you, but if person or address is null, you'll be very sad indeed. So you have to pre-instantiate any objects on which you want to use nested properties. This just seems like too much work.

Stripes will instantiate just about anything for you. Just so long as 'person' and 'address' have no-arg constructors Stripes will build them, join them together, set them on your ActionBean and then set the 'line1' property. It'll even instantiate implementations for any of the Collections interfaces, which leads us to...

Indexed Properties

The JavaBean specification is great for the most part, but it is clearly a child of the Java GUI side of the house. Struts' implementation of indexed properties matches the JavaBean specification. But in a disconnected web model, your Action won't know (and shouldn't have to know) how many indexed properties are coming back. To make it work you end up coding "smart" indexed getter and setter methods that extend lists and insert objects just to keep Struts happy.

Stripes does it all for you. You provide a getter and setter that use a List or Map type, using generics to inform Stripes what type of thing should be in the List or Map. Stripes will instantiate the List or Map for you if necessary, extend it when needed, manufacture new objects and put them in the list and, you know, set properties of objects in the list.

Validation

Other frameworks take shots at Struts for Validation. It's just fundamentally in the wrong place. Forget that the Struts Validator requires yet another XML file with yet more information about your Form that has to be kept in sync. Validation in Struts is divorced from Type Conversion. This is just plain wrong. Why validate that something is a well formed date, then spend a whole bunch more effort converting it to one? Well Struts almost does that. It expends a lot of effort validating it, tosses it all away and then performs some pretty weak type conversion. It's inefficient, and a pain to use. Which is why most of the Struts projects I've seen steer clear of Struts Validator.

In Stripes validation is tied closely to type conversion. A number of commonly used validations can be applied pre-conversion using a simple annotation. This includes things like required field checks, length checks, regex checking etc. Then type conversion happens, or tries to. If it fails, type converters produce validation errors that can be displayed to the user. So you can put your effort into creating good type converters (if the built in ones don't cover your needs) and be done.

Null mean Null (or does it)

Ask yourself this. You have an int property in your form bean and the user doesn't enter a value in a text box. What should your property be set to? Zero sounds like a sensible default right? Now, what if you have an Integer property? If the user doesn't submit anything I'd say that means null wouldn't you? Apparently Struts thinks it means zero again. And if there are validation errors, the form will even repopulate with the value zero! The only way to get around this is to declare your number properties as Strings and do your own conversion (which is a lot of effort if your dealing with multiple Locales).

Stripes takes the position that the HTML specification is a little inconsistent, and does what it can to make it more consistent for application developers. As a result, empty strings are treated the same as if the field didn't get submitted. The user didn't enter a value, so why should your ActionBean receive one and have to figure out what it means?

Formatting

The Struts FAQ says this (among other things) in answer to why the Struts tags provide so little formatting:

Great. JSTL doesn't have form tags, so while JSTL is great, it doesn't really help here. JSF and Struts don't mix so well even now, and that's a reason to leave people without formatting for years? And the last is just a cop-out. I want to be able to format data in the right way for each page. And it needs to be localizable etc.

Stripes provides high quality formatting support using the same java.text APIs that the JSTL formatting is built on. It's as similar to JSTL formatting as it can be, while being constrained to living on form tags (i.e. we can't have a separate tag for formatting a date vs. a number and so on).

Multi-Event Actions

If you want to have a form that submits multiple different events in Struts you have to either extend the DispatchAction or write your own support for it. And since the DispatchAction requires all buttons to have the same name, and uses the value to determine the method to invoke, it's a huge pain if you're using localized or even just externalized values for your buttons.

Stripes uses the name of the button itself, and has built in support for multi-event Actions. You can localize to your heart's content, and Stripes will detect which button was pressed and invoke the right method for you.

JSP / View Helpers

Struts doesn't really provide a good pattern for providing dynamic data to JSPs that are not the result of another Action. This leaves you with the options of writing a pre-action for the page or doing something outside of Struts. Then, what if another Action wants to forward to your page after it's completed processing. Do you chain the actions? Do you make the second action meet the dependencies of the page also?

Stripes has a neat way of handling this. A custom tag allows the use of ActionBeans as view helpers. It works similarly to the jsp:useBean tag in that if the ActionBean already exists it just gives you a reference to it. If it doesn't exist, the tag will bring it into existence, bind data out of the request on to it, and get it ready for use.

"HTML" Tags

Maybe it's just me who could never get used to it, but why do the Struts form input tags use 'property' instead of 'name'? And why is it 'styleClass' instead of 'class'? It also makes it hard to change a tag back and forth from a plain HTML tag to a Struts tag.

Stripes takes pains to make all the form input tags as close to (if not identical to) their HTML counterparts as possible.


Who had experience with Stripes? Does it make sense to use it? Your comments are welcome.

Friday, July 27, 2007

Using OSF Layers in Your JWeb Application

This article will discuss one strategy for combining frameworks using three popular open source frameworks. For the presentation layer we will use Struts; for our business layer we will use Spring; and for our persistence layer we will use Hibernate. You should be able to substitute any one of these frameworks in your application and get the same effect. Figure 1 shows what this looks like from a high level when the frameworks are combined.





Figure 1. Overview of framework architecture with Struts, Spring, and Hibernate.

Application Layer

Most non-trivial web applications can be divided into at least four layers of responsibility. These layers are the presentation, persistence, business, and domain model layers. Each layer has a distinct responsibility in the application and should not mix functionality with other layers. Each application layer should be isolated from other layers but allow an interface for communication between them. Let's start by inspecting each of these layers and discuss what these layers should provide and what they should not provide.

At one end of a typical web application is the presentation layer. Many Java developers understand what Struts provides. However, too often, coupled code such as business logic is placed into an org.apache.struts.Action. So, let's agree on what a framework like Struts should provide. Here is what Struts is responsible for:

- Managing requests and responses for a user.

- Providing a controller to delegate calls to business logic and other upstream processes.

- Handling exceptions from other tiers that throw exceptions to a Struts Action.

- Assembling a model that can be presented in a view.

- Performing UI validation.

Here are some items that are often coded using Struts but should not be associated with the presentation layer:

- Direct communication with the database, such as JDBC calls.

- Business logic and validation related to your application.

- Transaction management.

Introducing this type of code in the presentation layer leads to type coupling and cumbersome maintenance.

The Persistence Layer

At the other end of a typical web application is the persistence layer. This is usually where things get out of control fast. Developers underestimate the challenges in building their own persistence frameworks. A custom, in-house persistence layer not only requires a great amount of development time, but also often lacks functionality and becomes unmanageable. There are several open source object-to-relational mapping (ORM) frameworks that solve much of this problem. In particular, the Hibernate framework allows object-to-relational persistence and query service for Java. Hibernate has a medium learning curve for Java developers who are already familiar with SQL and the JDBC API. Hibernate persistent objects are based on plain-old Java objects and Java collections. Furthermore, using Hibernate does not interfere with your IDE. The following list contains the type of code that you would write inside a persistence framework:

- Querying relational information into objects. Hibernate does this through an OO query language called HQL, or by using an expressive criteria API. HQL is very similar to SQL except you use objects instead of tables and fields instead of columns. There are some new specific HQL language elements to learn; however, they are easy to understand and well documented. HQL is a natural language to use for querying objects that require a small learning curve.

- Saving, updating, and deleting information stored in a database.

- Advanced object-to-relational mapping frameworks like Hibernate have support for most major SQL databases, and they support parent/child relationships, transactions, inheritance, and polymorphism.

Here are some items that should be avoided in the persistence layer:

- Business logic should be in a higher layer of your application. Only data access operations should be permitted.

- You should not have persistence logic coupled with your presentation logic. Avoid logic in presentation components such as JSPs or servlet-based classes that communicate with data access directly. By isolating persistence logic into its own layer, the application becomes flexible to change without affecting code in other layers. For example, Hibernate could be replaced with another persistence framework or API without modification to the code in any other layer.

The Business Layer

The middle component of a typical web application is the business or service layer. This service layer is often the most ignored layer from a coding perspective. It is not uncommon to find this type of code scattered around in the UI layer or in the persistence layer. This is not the correct place because it leads to tightly coupled applications and code that can be hard to maintain over time. Fortunately, several frameworks exist that address these issues. Two of the most popular frameworks in this space are Spring and PicoContainer. These are referred to as microcontainers that have a very small footprint and determine how you wire your objects together. Both of these frameworks work on a simple concept of dependency injection (also known as inversion of control). This article will focus on Spring's use of setter injection through bean properties for named configuration parameters. Spring also allows a sophisticated form of constructor injection as an alternative to setter injection as well. The objects are wired together by a simple XML file that contains references to objects such as the transaction management handler, object factories, service objects that contain business logic, and data access objects (DAO).

The way Spring uses these concepts will be made clearer with examples later in this article. The business layer should be responsible for the following:

- Handling application business logic and business validation

- Managing transactions

- Allowing interfaces for interaction with other layers

- Managing dependencies between business level objects

- Adding flexibility between the presentation and the persistence layer so they do not directly communicate with each other

- Exposing a context to the business layer from the presentation layer to obtain business services

- Managing implementations from the business logic to the persistence layer

The Domain Model Layer

Finally, since we are addressing non-trivial, web-based applications we need a set of objects that can move between the different layers. The domain object layer consists of objects that represent real-world business objects such as an Order, OrderLineItem, Product, and so on. This layer allows developers to stop building and maintaining unnecessary data transfer objects, or DTOs, to match their domain objects. For example, Hibernate allows you to read database information into an object graph of domain objects, so that you can present it to your UI layer in a disconnected manner. Those objects can be updated and sent back across to the persistence layer and updated within the database. Furthermore, you do not have to transform objects into DTOs, which can get lost in translation as they are moved between different application layers. This model allows Java developers to work with objects naturally in an OO fashion without additional coding.

First, we will create our domain objects since they will interoperate with each layer. These objects will allow us to define what should be persisted, what business logic should be provided, and what type of presentation interface should be designed. Next, we will configure the persistence layer and define object-to-relational mappings with Hibernate for our domain objects. Then we will define and configure our business objects. After we have these components we can discuss wiring these layers using Spring. Finally, we will provide a presentation layer that knows how to communicate with the business service layer and knows how to handle exceptions that arise from other layers.

Conclusion

This article covers a lot of ground in terms of technology and architecture. The main concept to take away is how to better separate your application, user interface, persistence logic, and any other application layer you require. Doing this will decouple your code, allow new code components to be added, and make your application more maintainable in the future. The technologies covered here address specific problems well. However, by using this type of architecture you can replace application layers with other technologies.