This paper shows how to use Rhino to reach beyond JavaScript into Java [1]. Scripting Java has many uses. It allows us to write powerful scripts quickly by making use of the many Java libraries available. We can test Java classes by writing scripts. We can also aid our Java development by using scripting for exploratory programming. Exploratory programming is the process of learning about what a library or API can do by writing quick programs that use it. As we will see, scripting makes this process easier.
Note that the ECMA standard doesn't cover communication with Java (or with any external object system for that matter). All the functionality covered in this chapter should thus be considered an extension.
Every piece of Java code is part of a class. Every Java class is part of a package. In JavaScript, however, scripts exist outside of any package hierarchy. How then, do we access classes in Java packages?
Rhino defines a top-level variable named Packages. The
properties of the Packages
variable are all the top-level Java packages, such as java and
com. For example,
we can access the value of the javapackage:
js> Packages.java [JavaPackage java]
As a handy shortcut, Rhino defines a top-level variable java
that is equivalent to Packages.java.
So the previous example could be even shorter:
js> java [JavaPackage java]
We can access Java classes simply by stepping down the package hierarchy:
js> java.io.File [JavaClass java.io.File]
If your scripts access a lot of different Java classes it
can get awkward to use the full package name of the class every time. Rhino
provides a top-level function importPackage that serves the same
purpose as Java's import declaration. For example, we could
import all of the classes in the java.iopackage and
access class java.io.File
using just the name File:
js> importPackage(java.io) js> File [JavaClass java.io.File]
Here importPackage(java.io) makes all the classes in
the java.io package (such as File)
available at the top level. It's equivalent in effect to the Java
declaration import java.io.*;.
It's important to note that Java imports java.lang.*
implicitly, while Rhino does not. The reason is that JavaScript has
its own top-level objects Boolean, Math,
Number, Object, and String that
are different from the classes by those names defined in the
java.lang package. Because of this conflict, it's a good
idea not to use importPackage on the
java.lang package.
One thing to be careful of is Rhino's handling of errors in
specifying Java package or class names. If java.MyClass
is accessed, Rhino attempts to load a class named
java.MyClass. If that load fails, it assumes that
java.MyClass is a package name, and no error is
reported:
js> java.MyClass [JavaPackage java.MyClass]
Only if you attempt to use this object as a class will an error be reported.
Now that we can access Java classes, the next logical step is to
create an object. This works just as in Java, with the use of the
new operator:
js> new java.util.Date() Thu Jan 24 16:18:17 EST 2002
If we store the new object in a JavaScript variable, we can then call methods on it:
js> f = new java.io.File("test.txt")
test.txt
js> f.exists()
true
js> f.getName()
test.txt
Static methods and fields can be accessed from the class object itself:
js> java.lang.Math.PI 3.141592653589793 js> java.lang.Math.cos(0) 1
In JavaScript, unlike Java, the method by itself is an object and can be evaluated as well as being called. If we just view the method object by itself we can see the various overloaded forms of the method:
js> f.listFiles
function listFiles() {/*
java.io.File[] listFiles()
java.io.File[] listFiles(java.io.FilenameFilter)
java.io.File[] listFiles(java.io.FileFilter)
*/}
This output shows that the File class defines three
overloaded methods listFiles: one that takes no
arguments, another with a FilenameFilter argument, and a
third with a FileFilter argument. All the methods return
an array of File objects. Being able to view the
parameters and return type of Java methods is particularly useful in
exploratory programming where we might be investigating a method and
are unsure of the parameter or return types.
Another useful feature for exploratory programming is the ability
to see all the methods and fields defined for an object. Using the
JavaScript for..in construct, we can print out all these
values:
js> for (i in f) { print(i) }
exists
parentFile
mkdir
toString
wait
[44 others]
Note that not only the methods of the File class are
listed, but also the methods inherited from the base class
java.lang.Object (like wait). This makes it
easier to work with objects in deeply nested inheritance hierarchies
since you can see all the methods that are available for that
object.
Rhino provides another convenience by allowing properties of
JavaBeans to be accessed directly by their property names. A JavaBean
property foo is defined by the methods
getFoo and setFoo. Additionally, a boolean
property of the same name can be defined by an isFoo
method [2]. For example, the following code
actually calls the File object's getName and
isDirectory methods.
js> f.name test.txt js> f.directory false
The process of choosing a method to call based upon the types of the arguments is called overload resolution. In Java, overload resolution is performed at compile time, while in Rhino it occurs at runtime. This difference is inevitable given JavaScript's use of dynamic typing as was discussed in Chapter 2: since the type of a variable is not known until runtime, only then can overload resolution occur.
As an example, consider the following Java class that defines a number of overloaded methods and calls them.
public class Overload {
public String f(Object o) { return "f(Object)"; }
public String f(String s) { return "f(String)"; }
public String f(int i) { return "f(int)"; }
public String g(String s, int i) { return "g(String,int)"; }
public String g(int i, String s) { return "g(int,String)"; }
public static void main(String[] args) {
Overload o = new Overload();
Object[] a = new Object[] { new Integer(3), "hi", Overload.class };
for (int i = 0; i != a.length; ++i)
System.out.println(o.f(a[i]));
}
}
When we compile and execute the program, it produces the output
f(Object) f(Object) f(Object)
However, if we write a similar script
var o = new Packages.Overload();
var a = [ 3, "hi", Packages.Overload ];
for (var i = 0; i != a.length; ++i)
print(o.f(a[i]));
and execute it, we get the output
f(int) f(String) f(Object)
Because Rhino selects an overloaded method at runtime, it calls the more specific type that matches the argument. Meanwhile Java selects the overloaded method purely on the type of the argument at compile time.
Although this has the benefit of selecting a method that may be a better match for each call, it does have an impact on performance since more work is done at each call. In practice this performance cost hasn't been noticeable in real applications.
Because overload resolution occurs at runtime, it can fail at
runtime. For example, if we call Overload's method
g with two integers we get an error because neither form
of the method is closer to the argument types than the other:
js> o.g(3,4) js:"<stdin>", line 2: The choice of Java method Overload.g matching JavaScript argument types (number,number) is ambiguous; candidate methods are: class java.lang.String g(java.lang.String,int) class java.lang.String g(int,java.lang.String)
A more precise definition of overloading semantics can be found at http://www.mozilla.org/js/liveconnect/lc3_method_overloading.html.
Now that we can access Java classes, create Java objects, and
access fields, methods, and properties of those objects, we have a
great deal of power at our fingertips. However, there are a few
instances where that is not enough: many APIs in Java work by
providing interfaces that clients must implement. One example of this
is the Thread class: its constructor takes a
Runnable that contains a single method run
that will be called when the new thread is started.
To address this need, Rhino provides the ability to create new Java
objects that implement interfaces. First we must define a JavaScript
object with function properties whose names match the methods required
by the Java interface. To implement a Runnable, we need
only define a single method run with no parameters. If
you remember from Chapter 3, it is possible to define a JavaScript
object with the {propertyName: value} notation. We can
use that syntax here in combination with a function expression to
define a JavaScript object with a run method:
js> obj = { run: function () { print("\nrunning"); } }
[object Object]
js> obj.run()
running
Now we can create an object implementing the Runnable interface
by constructing a Runnable:
js> r = new java.lang.Runnable(obj); [object JavaObject]
In Java it is not possible to use the new operator on
an interface because there is no implementation available. Here Rhino
gets the implementation from the JavaScript object
obj. Now that we have an object implementing
Runnable, we can create a Thread and run
it. The function we defined for run will be called on a
new thread.
js> t = new java.lang.Thread(r) Thread[Thread-2,5,main] js> t.start() js> running
The final js prompt and the output from the new thread
may appear in either order, depending on thread scheduling.
Behind the scenes, Rhino generates the bytecode for a new Java
class that implements Runnable and forwards all calls to
its run method over to an associated JavaScript
object. The object that implements this class is called a Java
adapter. Because the forwarding to JavaScript occurs at runtime,
it is possible to delay defining the methods implementing an interface
until they are called. While omitting a required method is bad
practice for programming in the large, it's useful for small scripts
and for exploratory programming.
JavaAdapter constructorIn the previous section we created Java adapters using the
new operator with Java interfaces. This approach has its
limitations: it's not possible to implement multiple interfaces, nor
can we extend non-abstract classes. For these reasons there is a
JavaAdapter constructor.
The syntax of the JavaAdapter constructor is
new JavaAdapter(javaIntfOrClass, [javaIntf, ..., javaIntf,] javascriptObject)
Here javaIntfOrClass is an interface to implement or a
class to extend and javaIntf are aditional interfaces to
implement. The javascriptObject is the JavaScript object
containing the methods that will be called from the Java adapter.
In practice there's little need to call the
JavaAdapter constructor directly. Most of the time the
previous syntaxes using the new operator will be
sufficient.
Often we need to implement an interface with only one method, like in the
previous Runnable example or when providing various event
listener implementations. To facilitate this Rhino allows to pass
JavaScript function when such interface is expected. The
function is called as the implementation of interface method.
Here is the simplified Runnable example:
js> t = java.lang.Thread(function () { print("\nrunning"); });
Thread[Thread-0,5,main]
js> t.start()
js>
running
Rhino also allows to use JavaScript function as implementation of Java
interface with more then method if all the methods has the same
signature. When calling the function, Rhino passes method's name as
the additional argument. Function can
use it to distinguish on behalf of which method it was called:
js> var frame = new Packages.javax.swing.JFrame();
js> frame.addWindowListener(function(event, methodName) {
if (methodName == "windowClosing") {
print("Calling System.exit()..."); java.lang.System.exit(0);
}
});
js> frame.setSize(100, 100);
js> frame.visible = true;
true
js> Calling System.exit()...
Rhino provides no special syntax for creating Java arrays. You
must use the java.lang.reflect.Array class for this
purpose. To create an array of five Java strings you would make the
following call:
js> a = java.lang.reflect.Array.newInstance(java.lang.String, 5); [Ljava.lang.String;@7ffe01
To create an array of primitive types, we must use the special TYPE
field defined in the associated object class in the
java.lang package. For example, to create an array of
bytes, we must use the special field
java.lang.Byte.TYPE:
js> a = java.lang.reflect.Array.newInstance(java.lang.Character.TYPE, 2); [C@7a84e4
The resulting value can then be used anywhere a Java array of that type is allowed.
js> a[0] = 104 104 js> a[1] = 105 105 js> new java.lang.String(a) hi
It's important to keep in mind that Java strings and JavaScript
strings are not the same. Java strings are instances
of the type java.lang.String and have all the methods
defined by that class. JavaScript strings have methods defined by
String.prototype. The most common stumbling block is
length, which is a method of Java strings and a dynamic
property of JavaScript strings:
js> javaString = new java.lang.String("Java")
Java
js> jsString = "JavaScript"
JavaScript
js> javaString.length()
4
js> jsString.length
10
Rhino provides some help in reducing the differences between the
two types. First, you can pass a JavaScript string to a Java method
that requires a Java string and Rhino will perform the conversion. We
actually saw this feature in action on the call to the
java.lang.String constructor in the preceding
example.
Rhino also makes the JavaScript methods available to Java strings if the java.lang.String class doesn't already define them. For example:
js> javaString.match(/a.*/) ava
[1] The ability to call Java from JavaScript was first implemented as part of a Netscape browser technology called LiveConnect. However, since that technology also encompassed communication with browser plugins, and since the way of calling JavaScript from Java in Rhino is entirely different, that term won't be used in this paper.
[2] For more information on JavaBeans, see Developing Java Beans by Robert Englander.