Home
Interests
Photos
Favorites

Java Feature: Developing Intelligent Web Applications with AJAX Part 2

The publicity that AJAX grabbed over the last half a year is based on closing the gap between the Web applications and the desktop applications, combining the "reach" and "rich." At the same time, the gap between the technological level of AJAX and what corporate developers expect in their modern arsenal is really astonishing. After all, AJAX is neither a tool nor a platform. There is no AJAX standards committee or community process in place. While software vendors are crafting proprietary development platforms on top of AJAX (which pretty much means "from scratch"), early adopters of AJAX are left on their own.

 In Part 1 (JDJ, Vol. 10, issue 9) we touched on the foundation of AJAX - the ability to establish script-to-server communication. This is what makes HTML pages dynamic and responsive. Does it mean we are ready to kick-off our own version of Yahoo mail? No, we are not. Here is why: AJAX is a mixed blessing. On one hand it enables us to create rich, desktop-class applications on the Web. On the other, if we compare "page-flipping" Web applications with the client/server or Swing ones, the development practices are not quite the same. What about management expectations? We'll need to get used to the fact that it takes time to build a rich UI. The more flexibility with more permutations the user is allowed - the more time it takes.

The answer, of course, is component libraries, frameworks, and industrial-strength tools. Leaving tools aside, this article concentrates on what is available for AJAX enthusiasts today. Addressing a need to build reusable business components, it focuses on the "hidden" object-oriented power of JavaScript. Also, by addressing a need to build custom-rich UI components, it illustrates a convenient way to encapsulate presentation logic in custom client-side HTML tags.

AJAX Language: Object-Oriented JavaScript
By definition, JavaScript is the language of classic AJAX. Unlike Java, JavaScript does not enforce the OO style of coding. That said, it's surprising how often it's overlooked that Java-Script fully supports all the major attributes of an OO language: inheritance, polymorphism, and encapsulation. Douglas Crockford even named Java Script "The World's Most Misunderstood Programming Language." Let's review the object-oriented side of JavaScript.

Data Types
In Java, a class defines a combination of data and its associated behaviors. While JavaScript reserves the class keyword, it does not support the same semantic as in conventional OOP languages.

It may sound strange but in JavaScript, functions are used as object definitions. By defining a function in the example below you, in fact, define a simple empty class - Calculator:

function Calculator() {}

A new instance is created the same way as in Java - by using the new operator:

var myCalculator = new Calculator();

The function not only defines a class, but also acts as a constructor. The operator new does the magic, instantiating an object of class Calculator and returning an object reference in contrast to merely calling the function.

Creating an empty class is nice but not really useful in real life. We are going to fill-in the class definition using a Java-Script prototype construct. JavaScript uses prototype to serve as a template for object creation. All prototype properties and methods are copied by reference into each object of a class, so they all have the same values. You can change the value of a prototype property in one object, and the new value overrides the default, copied from the prototype, but only in that one instance. The following statement will add a new property to the prototype of the Calculator object:

Calculator.prototype._prop = 0;

Since JavaScript does not provide a way to syntactically denote a class definition, we'll use the with statement to mark the class definition boundaries. This will also make the example code smaller as the with statement is allowed to perform a series of statements on a specified object without qualifying the attributes.

function Calculator() {};
with (Calculator) {
   prototype._prop = 0;
   prototype.setProp = function(p) {_prop = p};
   prototype.getProp = function() {return _prop};
}

So far we have defined and initialized the public _prop variable as well as provided getter and setter methods for it.

Need to define a static variable? Just think of the static variable as being a variable owned by the class. Because classes in JavaScript are represented by function objects, we just need to add a new property to the function:

Calculator.iCount = 0;

Now that the iCount variable is a property of the Calculator object, it will be shared between all instances of the class calculator.

function Calculator() {Calculator.iCount++;};

The above code will count all created instances of the class Calculator.

Encapsulation
Using "Calculator", as defined above, permits access to all the "class" data, increasing the risk of name collisions in inherited classes. We clearly need encapsulation to view objects as self-contained entities.

A standard language mechanism of data encapsulation is private variables. And a common JavaScript technique for emulating a private variable is to define a local variable in the constructor, so that this local variable is accessible exclusively via getter and setter - inner functions of the very same constructor. In the following example, the _prop variable is defined within the Calculator function and is not visible outside of the function scope. Two anonymous inner functions, assigned to setProp and getProp attributes, provide access to our "private" variable. Also, please note the use of this, quite similar to how it is used in Java:

function Calculator() {
   var _prop = 0;
   this.setProp = function (p){_prop = p};
   this.getProp = function() {return _prop};
};

What is often overlooked is the cost of such encapsulation in JavaScript. It can be tremendous, because inner function objects get repeatedly created for each instance of the "class".

Accordingly, since constructing objects based on the prototype is faster and consumes less memory, we cast our vote in favor of public variables wherever performance is critical. You can use naming conventions to avoid name collisions, for example, by prefixing public variables with the class name.

Inheritance
At first glance, JavaScript lacks support for the class hierarchy similar to what programmers of conventional object-oriented languages expect from the modern language. However, although JavaScript syntax does not support class inheritance as in Java, inheritance can still be implemented by copying an instance of a previously defined class into the prototype of the derived one.

Before we provide an illustration, we need to introduce a constructor property. JavaScript makes sure that every prototype contains constructor, which holds a reference to the constructor function. In other words, Calculator.prototype.constructor contains a reference to Calculator().

 

Now, the code below shows how to derive the class ArithmeticCalculator from the base class Calculator. "Line 1" results in borrowing all properties of the Calculator, while "Line 2" restores the value of the prototype, constructor back to ArithmeticCalculator:

function ArithmeticCalculator() { };
with (ArithmeticCalculator) {
   ArithmeticCalculator .prototype = new Calculator(); //Line 1
   prototype.constructor = ArithmeticCalculator; //Line 2
}

Even if the example above looks like a composition rather than inheritance, the JavaScript engine knows about the prototype chain. In particular, the instanceof operator will work correctly with both the base and derived classes. Assuming you create a new instance of a class ArithmeticCalculator:

var c = new ArithmeticCalculator;

expressions c instanceof Calculator and c instanceof ArithmeticCalculator will both evaluate to true.

Notice, that the constructor of the base class in the example above is called at the point when the ArithmeticCalculator prototype is initialized and not when an instance of the derived class is created. This could have unwanted side effects and you should consider creating a separate function for initialization purposes. As the constructor is not a member function, it can't be called through this reference directly. We will need to create a "Calculator" member function to be able to call super:

function Calculator(ops) { ...};
with (Calculator) {
   prototype.Calculator = Calculator;
}

Now we can write an inherited class that explicitly calls the constructor in the base class:

function ArithmeticCalculator(ops) {
   this.Calculator(ops);
};
with (ArithmeticCalculator) {
   ArithmeticCalculator .prototype = new Calculator;
   prototype.constructor = ArithmeticCalculator;

   prototype.ArithmeticCalculator = ArithmeticCalculator;
}

Polymorphism
JavaScript is a non-typed language where everything is an object. Accordingly, if there are two classes A and B, both defining method foo(), JavaScript will allow polymorphic invocation of foo() across instances of A and B even if there is no hierarchical relation (albeit implementational) whatsoever. From that perspective, JavaScript provides a wider polymorphism then Java. The flexibility, as usual, comes at a price. In this case, it is a price of delegating the type checking job to application code. Specifically, if there is a need to check that a reference indeed points to a desired base class, it can be done with the instanceof operator.

On the other hand, JavaScript doesn't check parameters in the function calls, which prevents from defining polymorphic functions with the same name and different parameters (and let the compiler choose the right signature). Instead, JavaScript provides an argument object - Java 5 style - within a function scope that allows you to implement a different behavior depending on the parameter's type and quantity.

Example
Listing 1 implements a calculator that calculates expressions in a reverse Polish notation. It illustrates the main techniques described in the articles and also shows the usage of the unique JavaScript features, such as accessing object properties as an array element for a dynamic function call.

To make Listing 1 work we also need to provide a piece of code that instantiates the calculator objects and calls the evaluate method:

var e = new ArithmeticCalcuator([2,2,5,"add","mul"]);
alert(e.evaluate());

AJAX Component Authoring
All AJAX component authoring solutions known today can be logically divided into two groups. The first group specifically targets the seamless integration with the HTML-based UI definition. The second group drops HTML as a UI definition language in favor of certain XML. In this article we illustrate one approach from the first group, an analog to JSP tags, albeit in the browser. These browser-specific component authoring extensions are called element behaviors in the IE case or extensible bindings in the case of the latest versions of Firefox, Mozilla, and Netscape 8.

Custom Tag Dialects
Internet Explorer, starting with version 5.5, enables the JavaScript authoring of custom, client-side HTML elements. Unlike JSP tags, these objects are not preprocessed into HTML on the server side. Rather, they're legitimate extensions of a standard HTML object model and everything, including control construction, happens dynamically on the client. Similarly, Gecko-engine based browsers can dynamically decorate any existing HTML element with a reusable functionality.

It's possible, therefore, to build a library of rich UI components with methods, events, and attributes that will have HTML syntax. Such components can be freely mixed with standard HTML. Internally, these components will communicate with application servers, AJAX style. In other words, it's possible (and relatively simple) to build your own AJAX object model.

The IE flavor of this approach is called HTC or HTML components; the Gecko version is called XBL - eXtensible Bindings Language. For the purposes of this article, we'll focus on IE.

Enter the HTML Components - HTC
HTC or HTML components are also called behaviors. In turn they are divided into attached behaviors that decorate any existing HTML element with a set of properties, events, and methods, and element behaviors that look like an extended set of custom HTML tags to the hosting page. Together, element and attached behaviors provide a simple solution for authoring both components and applications. Here we'll illustrate the most comprehensive case, element behaviors.

Data-Bound Checkbox Control
As an illustration of element behavior, we'll construct a custom data-bound checkbox. The rationale behind building such a control is that a standard HTML checkbox has several noticeable shortcomings:

  • It requires application code to map the value of the "checked" attribute into business domain values, such as "Y[es]"/"N[o]", "M[ale]"/"F[emale]", etc. The HTML checkbox uses "checked" contrary to many other HTML controls using "value".
  • It requires application code to maintain the state of the control (modified versus not modified). This is actually a common problem with all HTML controls.
  • It requires application code to create an associated label that should accept click and change the state of the checkbox accordingly.
  • The standard HTML checkbox doesn't support "validation" events to allow the canceling of a GUI action under certain application conditions.
To settle on a syntax, let's say that a sample usage of the control we are building could look the following way:

<checkbox id="cbx_1" value="N" labelonleft="true"
label="Show Details:" onValue="Y" offValue="N"/>

In addition, our control will support the cancelable event onItemChanging and the notification event onItemChanged.

 

Custom Tag Definition
Structurally, a custom tag is a file with an HTC extension that describes its properties, methods, and events between <PUBLIC:COMPONENT> and </PUBLIC:COMPONENT>.

To define a custom CHECKBOX tag, we create a file checkbox.htc as in the following snippet where the first line sets the tag name of the component:

<PUBLIC:COMPONENT NAME="cbx" tagName="CHECKBOX">
<PROPERTY NAME="value" GET="getValue" PUT="putValue" />
// Here we place all other properties of the component
<METHOD NAME="show" />
// Here we place all other methods of the components
<EVENT NAME="onItemChanging" ID="onItemChanging"/>
// Here we place all other events that component will fire to application
<ATTACH EVENT="oncontentready" HANDLER="constructor" />
// Here we place all other events that component handles itself
<SCRIPT >
// Here we place all methods, properties getters and setters and event handlers
</SCRIPT>
</PUBLIC:COMPONENT>

Custom Tag Use Case
While the contents of the HTC file matter a lot, the name of the file is irrelevant, although, ultimately, the URL to the HTC file needs to be specified using the IMPORT instruction. It has to be done before the corresponding custom tag is mentioned for the first time (on the page). Here is how the simplest possible page utilizing a custom checkbox might look, assuming the page and the HTC file are located in one folder:

<HTML xmlns:myns>
<?IMPORT namespace="myns" implementation="checkbox.htc" >
<BODY>
<myns:checkbox id='cbx_1' label='Hello'/>
</BODY>
</HTML>

Please notice how custom CHECKBOX has been mapped to a nondefault namespace "myns" in the opening HTML tag. The IMPORT instruction performs a synchronous load of the HTC into the browser's memory and also instructs the browser how to perform name resolution for the appropriate namespace (HTC to namespace association can be many-to-one).

Constructor of the Custom Tag
The best way to initialize HTC, once it's loaded, is to process the "oncontentready" event. Accordingly, we define a handler function, which for sheer clarity is called constructor:

<ATTACH EVENT="oncontentready" HANDLER="constructor" />

The logic of constructor() is simple: concatenate a regular HTML checkbox and HTML label in the order dependent on the value of property labelonleft (see property definition below):

function constructor() {
   // We will add an HTML checkbox and label to the element body
   // See Listing 2 for details
}

Defining Custom Tag Properties
To define property labelonleft, we add one more line to the <PUBLIC:COMPONENT> section:

<PROPERTY NAME="labelonleft" VALUE="true"/>

Please note that this property does not contain getter and/or setter methods. Properties onValue and offValue that provide the mapping of the checkbox status into a business value domain also don't need getters and setters:

<PROPERTY NAME="onValue" VALUE="true"/>
<PROPERTY NAME="offValue" VALUE="false" />

However, property checked is defined with both getter and setter:

<PROPERTY NAME="checked" GET="getChecked" PUT="putChecked" />

Accordingly, we have definitions for both methods in the <SCRIPT> section. As you can see, setter putChecked(), which is triggered every time checked is modified, sets the value property to one of two variants: onValue or OffValue. Please note that putChecked() will get triggered not only by the script on the checkbox-hosting page, but also by any assignment to checked done inside this very checkbox.htc.

var _value;
function putChecked( newValue ) {
   value = (newValue?onValue:offValue);
}
function getChecked(){
   return ( _value == onValue);
}

Defining Events for the Custom Tag
Let's look at the definition of onItemChanging and onItemChanged events and how these events are being fired and processed inside the setter for value property (see Listing 2).

Method putValue() has a couple of points of interest. First, it can be called during the parsing on the CHECKBOX tag, as long as the HTML value attribute is specified. That's why we have a separate logic branch for not constructed objects, to differentiate the process of construction from a reaction to a user click. Second, we illustrate here the creation and processing of the custom event onItemChanging, which allows the application to cancel (block) the action. Please note that both clicking as well as programmatic assignment to the value can get cancelled this way.

Event Canceling
To cancel the event, an application should intercept the event and set event.returnValue to false. The schematic snippet of code illustrating how the application would cancel the processing is presented below:

cbx_1::onItemChanging() {
. . . . .
if (canNotBeAllowed) {
   event.returnValue=false;
. . . . .
}

If the event is not cancelled, putValue() sets the internal, plain HTML checkbox's checked property as per the current value: if that is equal to onValue, the internal checkbox will get checked; if it is equal to offValue (there is no third option), unchecked (the full listing is presented in Listing 2).

HTML Internals of the CHECKBOX
The painting of our control is done by the helper functions addLabel() and addCheckBox() and is called from within a constructor(). These functions inject HTML inside the element's innerHTML (element is the analog for this in HTC parlor). The injected HTML, in the simplified form, looks like the following:

<LABEL for=cb_{uniqueID}>Show Details:</LABEL>
<INPUT id=cb_{uniqueID} type=checkbox />

where uniqueID is a unique (within a page) string literal generated by IE, which identifies the instance of the HTC.

 

Encapsulation ... Again
There is one shortcoming in our CHECKBOX. The way we built it, the HTML injected during the constructor() contributes to the DOM of the page that hosts the HTC. Also global JavaScript variables like_value fall into the global scope of the document they're included in. This is dangerous since we run into the possibility of name clashes: the most obvious case is more than one instance of the same control. In addition, it presents a possibility that our control may accidentally reference other objects with the same names and vice versa.

To put it simply, a special mechanism is required to enable a truly modular approach for object authoring. Fortunately, HTC technology supports an intelligent answer - viewLink.

The easiest way to declare a control as encapsulated is to put one extra declaration between opening and closing PUBLIC:COMPONENT tags:

<PUBLIC:DEFAULTS viewLinkContent/>

Instantly, the control becomes encapsulated; it has its own HTML document tree, atomic to the master document. Every instance of the object has its own set of instantiated values and only public methods and properties can be accessed by the outside code. In other words, the viewLink mechanism fully enables the design and implementation of sophisticated Web applications using a true OO component-based approach.

In particular, we can simplify our code by removing uniqueID suffixes from the definition of internal checkboxes and HTML labels, since we are not afraid of name clashes anymore. Accordingly, we may replace the line:

eval( 'cb_'+uniqueID).checked = ( _value == onValue );

with the

cb.checked = ( _value == onValue );

as well as change addCheckbox() and addLabel() accordingly.

Conclusion
Since the AJAX race has just started, there are no AJAX standards and no commonly accepted RAD tools you can rely on to build your applications. While software vendors have a long way to go to create the robust development platforms, AJAX enthusiasts can prepare by encapsulating reusable blocks of code as business components with a well-defined API.

Navigating in this direction, this article outlined the OO "powers" of the AJAX language - JavaScript. It also illustrated one of the available component-authoring strategies - client-side custom tags technology. While we presented only IE-specific custom tags, we also provide a downloadable example of extensible bindings example for the Mozilla browser. All article-related examples can be downloaded from www.ajaxmaker.com/JDJ/AJAX/partII.html.

 

2005 SYS-CON Media Inc.

Questions or problems regarding this web site should be directed to abeckman@outdoorssite.com.

Copyright 2008 Art Beckman. All rights reserved.

Last Modified: March 9, 2008