8014646: Update the Java interop documentation in the Java Scripting Programmer's Guide
authorattila
Wed, 15 May 2013 14:54:28 +0200
changeset 17747 57c9166de06e
parent 17746 bc27c77a3568
child 17748 e460be905b08
8014646: Update the Java interop documentation in the Java Scripting Programmer's Guide Reviewed-by: jlaskey, hannesw, lagergren
nashorn/docs/JavaScriptingProgrammersGuide.html
--- a/nashorn/docs/JavaScriptingProgrammersGuide.html	Wed May 15 10:28:18 2013 +0200
+++ b/nashorn/docs/JavaScriptingProgrammersGuide.html	Wed May 15 14:54:28 2013 +0200
@@ -71,9 +71,20 @@
 Arrays</a></span></li>
 <li><span><a href="#jsimplement">Implementing Java
 Interfaces</a></span></li>
-<li><span><a href="#jsextend">Extending Java classes
+<li><span><a href="#jsextendabstract">Extending Abstract Java Classes
+</a></span></li>
+<li><span><a href="#jsextendconcrete">Extending Concrete Java Classes
+</a></span></li>
+<li><span><a href="#jsimplementmultiple">Implementing Multiple Java Interfaces
+</a></span></li>
+<li><span><a href="#classBoundImplementations">Class-Bound Implementations
 </a></span></li>
 <li><span><a href="#jsoverload">Overload Resolution</a></span></li>
+<li><span><a href="#dataTypeMapping">Mapping of Data Types Between Java
+and JavaScript</a></span></li>
+
+
+
 </ul>
 </li>
 <li><span><a href="#engineimpl">Implementing Your Own Script
@@ -466,10 +477,10 @@
 </code>
 </pre> 
 
-Note that the name of the type is always a string for a fully qualified name. You can use any of these types to create new instances, e.g.:
+Note that the name of the type is always a string for a fully qualified name. You can use any of these expressions to create new instances, e.g.:
 
 <pre><code>
- var anArrayList = new Java.type("java.util.ArrayList")
+ var anArrayList = new (Java.type("java.util.ArrayList"))
 </code></pre> 
 
 or
@@ -496,6 +507,37 @@
 <p>
 You can access both static and non-static inner classes. If you want to create an instance of a non-static inner class, remember to pass an instance of its outer class as the first argument to the constructor.
 </p>
+<p>
+In addition to creating new instances, the type objects returned from <code>Java.type</code> calls can also be used to access the
+static fields and methods of the classes:
+<pre><code>
+ var File = Java.type("java.io.File")
+ File.createTempFile("nashorn", ".tmp")
+</code></pre> 
+<p>
+Methods with names of the form <code>isXxx()</code>, <code>getXxx()</code>, and <code>setXxx()</code> can also be used as properties, for both instances and statics.
+</p>
+<p>
+A type object returned from <code>Java.type</code> is distinct from a <code>java.lang.Class</code> object. You can obtain one from the other using properties <code>class</code> and <code>static</code> on them.
+<pre><code>
+ var ArrayList = Java.type("java.util.ArrayList")
+ var a = new ArrayList
+
+ // All of the following print true:
+ print("Type acts as target of instanceof: " + (a instanceof ArrayList))
+ print("Class doesn't act as target of instanceof: " + !(a instanceof a.getClass()))
+ print("Type is not same as instance's getClass(): " + (a.getClass() !== ArrayList))
+ print("Type's `class` property is same as instance getClass(): " + (a.getClass() === ArrayList.class))
+ print("Type is same as instance getClass()'s `static` property: " + (a.getClass().static === ArrayList))
+</code></pre> 
+<p>
+You can think of the type object as similar to the class names as used in Java source code: you use them as the
+arguments to the <code>new</code> and <code>instanceof</code> operators and as the namespace for the static fields
+and methods, but they are different than the runtime <code>Class</code> objects returned by <code>getClass()</code> calls.
+Syntactically and semantically, this separation produces code that is most similar to Java code, where a distinction
+between compile-time class expressions and runtime class objects also exists. (Also, Java can't have the equivalent of <code>static</code>
+property on a <code>Class</code> object since compile-time class expressions are never reified as objects).
+</p>
 <hr>
 <a name="jsimport" id="jsimport"></a>
 <h3>Importing Java Packages, Classes</h3>
@@ -558,10 +600,7 @@
 <a name="jsarrays" id="jsarrays"></a>
 <h3>Creating, Converting and Using Java Arrays</h3>
 <p>
-Array element access or length access is
-the same as in Java. Also, a script array can be used when a Java
-method expects a Java array (auto conversion). So in most cases we
-don't have to create Java arrays explicitly.</p>
+Array element access or length access is the same as in Java.</p>
 <pre><code>
 // <a href="source/javaarray.js">javaarray.js</a>
 
@@ -587,7 +626,11 @@
  print(javaIntArray[2]) // prints 0, as boolean false was converted to number 0 as per ECMAScript ToNumber conversion
 </code></pre>
 <p>
-Given a Java array or Collection, <code>Java.toJavaScriptArray</code> returns a JavaScript array with a shallow copy of its contents. Note that in most cases, you can use Java arrays and lists natively in Nashorn; in cases where for some reason you need to have an actual JavaScript native array (e.g. to work with the array comprehensions functions), you will want to use this method.i
+You can use either a string or a type object returned from <code>Java.type()</code> to specify the component type of the array. 
+You can also omit the array type, in which case a <code>Object[]</code> will be created.
+</p>
+<p>
+Given a Java array or Collection, <code>Java.toJavaScriptArray</code> returns a JavaScript array with a shallow copy of its contents. Note that in most cases, you can use Java arrays and lists natively in Nashorn; in cases where for some reason you need to have an actual JavaScript native array (e.g. to work with the array comprehensions functions), you will want to use this method.
 </p>
 <pre><code>
 var File = Java.type("java.io.File");
@@ -597,7 +640,7 @@
 </code></pre>
 <hr>
 <a name="jsimplement" id="jsimplement"></a>
-<h3>Implementing Java Interfaces</h3>
+<h3>Implementing Java interfaces</h3>
 <p>A Java interface can be implemented in JavaScript by using a
 Java anonymous class-like syntax:</p>
 <pre><code>
@@ -631,8 +674,8 @@
 </code>
 </pre>
 <hr>
-<a name="jsextend" id="jsextend"></a>
-<h3>Extending Java classes</h3>
+<a name="jsextendabstract" id="jsextendabstract"></a>
+<h3>Extending Abstract Java Classes</h3>
 <p>
 If a Java class is abstract, you can instantiate an anonymous subclass of it using an argument list that is applicable to any of its public or protected constructors, but inserting a JavaScript object with functions properties that provide JavaScript implementations of the abstract methods. If method names are overloaded, the JavaScript function will provide implementation for all overloads. E.g.:
 </p>
@@ -671,6 +714,9 @@
 
 Here, <code>Timer.schedule()</code> expects a <code>TimerTask</code> as its argument, so Nashorn creates an instance of a TimerTask subclass and uses the passed function to implement its only abstract method, run(). In this usage though, you can't use non-default constructors; the type must be either an interface, or must have a protected or public no-arg constructor.
 
+<hr>
+<a name="jsextendconcrete" id="jsextendconcrete"></a>
+<h3>Extending Concrete Java Classes</h3>
 <p>
 To extend a concrete Java class, you have to use <code>Java.extend</code> function.
 <code>Java.extend</code> returns a type object for a subclass of the specified Java class (or implementation of the specified interface) that acts as a script-to-Java adapter for it.  
@@ -695,26 +741,178 @@
 printSizeInvokedArrayList.size();
 printAddInvokedArrayList.add(33, 33);
 </code></pre>
+<p>
+The reason you must use <code>Java.extend()</code> with concrete classes is that with concrete classes, there can be a 
+syntactic ambiguity if you just invoke their constructor. Consider this example:
+</p>
+<pre><code>
+var t = new java.lang.Thread({ run: function() { print("Hello!") } })
+</code></pre>
+<p>
+If we allowed subclassing of concrete classes with constructor syntax, Nashorn couldn't tell if you're creating a new 
+<code>Thread</code> and passing it a <code>Runnable</code> at this point, or you are subclassing <code>Thread</code> and
+passing it a new implementation for its own <code>run()</code> method.
+</p>
+<hr>
+<a name="jsimplementmultiple" id="jsimplementmultiple"></a>
+<h3>Implementing Multiple Interfaces</h3>
+<p>
+<code>Java.extend</code> can in fact take a list of multiple types. At most one of the types can be a class, and the rest must
+be interfaces (the class doesn't have to be the first in the list). You will get back an object that extends the class and 
+implements all the interfaces. (Obviously, if you only specify interfaces and no class, the object will extend <code>java.lang.Object</code>).
+<hr>
+<a name="classBoundImplementations" id="classBoundImplementations"></a>
+<h3>Class-Bound Implementations</h3>
+<p>
+The methods shown so far for extending Java classes and implementing interfaces &ndash; passing an implementation JavaScript object 
+or function to a constructor, or using <code>Java.extend</code> with <code>new</code> &ndash; all produce classes that take an
+extra JavaScript object parameter in their constructors that specifies the implementation. The implementation is therefore always bound
+to the actual instance being created with <code>new</code>, and not to the whole class. This has some advantages, for example in the
+memory footprint of the runtime, as Nashorn can just create a single "universal adapter" for every combination of types being implemented.
+In reality, the below code shows that different instantiations of, say, <code>Runnable</code> have the same class regardless of them having
+different JavaScript implementation objects:
+</p>
+<pre><code>
+var Runnable = java.lang.Runnable;
+var r1 = new Runnable(function() { print("I'm runnable 1!") })
+var r2 = new Runnable(function() { print("I'm runnable 2!") })
+r1.run()
+r2.run()
+print("We share the same class: " + (r1.class === r2.class))
+</code></pre>
+<p>
+prints:
+</p>
+<pre><code>
+I'm runnable 1!
+I'm runnable 2!
+We share the same class: true
+</code></pre>
+<p>
+Sometimes, however, you'll want to extend a Java class or implement an interface with implementation bound to the class, not to
+its instances. Such a need arises, for example, when you need to pass the class for instantiation to an external API; prime example
+of this is the JavaFX framework where you need to pass an Application class to the FX API and let it instantiate it.
+</p>
+<p>
+Fortunately, there's a solution for that: <code>Java.extend()</code> &ndash; aside from being able to take any number of type parameters
+denoting a class to extend and interfaces to implement &ndash; can also take one last argument that has to be a JavaScript object
+that serves as the implementation for the methods. In this case, <code>Java.extend()</code> will create a class that has the same
+constructors as the original class had, as they don't need to take an an extra implementation object parameter. The example below
+shows how you can create class-bound implementations, and shows that in this case, the implementation classes for different invocations
+are indeed different:
+</p>
+<pre><code>
+var RunnableImpl1 = Java.extend(java.lang.Runnable, function() { print("I'm runnable 1!") })
+var RunnableImpl2 = Java.extend(java.lang.Runnable, function() { print("I'm runnable 2!") })
+var r1 = new RunnableImpl1()
+var r2 = new RunnableImpl2()
+r1.run()
+r2.run()
+print("We share the same class: " + (r1.class === r2.class))
+</code></pre>
+<p>
+prints:
+</p>
+<pre><code>
+I'm runnable 1!
+I'm runnable 2!
+We share the same class: false
+</code></pre>
+<p>
+As you can see, the major difference here is that we moved the implementation object into the invocation of <code>Java.extend</code>
+from the constructor invocations &ndash; indeed the constructor invocations now don't even need to take an extra parameter! Since
+the implementations are bound to a class, the two classes obviously can't be the same, and we indeed see that the two runnables no
+longer share the same class &ndash; every invocation of <code>Java.extend()</code> with a class-specific implementation object triggers
+the creation of a new Java adapter class.
+</p>
+<p>
+Finally, the adapter classes with class-bound implementations can <i>still</i> take an additional constructor parameter to further
+override the behavior on a per-instance basis. Thus, you can even combine the two approaches: you can provide part of the implementation
+in a class-based JavaScript implementation object passed to <code>Java.extend</code>, and part in another object passed to the constructor.
+Whatever functions are provided by the constructor-passed object will override the functions in the class-bound object.
+</p>
+<pre><code>
+var RunnableImpl = Java.extend(java.lang.Runnable, function() { print("I'm runnable 1!") })
+var r1 = new RunnableImpl()
+var r2 = new RunnableImpl(function() { print("I'm runnable 2!") })
+r1.run()
+r2.run()
+print("We share the same class: " + (r1.class === r2.class))
+</code></pre>
+<p>
+prints:
+</p>
+<pre><code>
+I'm runnable 1!
+I'm runnable 2!
+We share the same class: true
+</code></pre>
 <hr>
 <a name="jsoverload" id="jsoverload"></a>
 <h3>Overload Resolution</h3>
 <p>Java methods can be overloaded by argument types. In Java,
 overload resolution occurs at compile time (performed by javac).
-When calling Java methods from a script, the script
-interpreter/compiler needs to select the appropriate method. With
-the JavaScript engine, you do not need to do anything special - the
-correct Java method overload variant is selected based on the
-argument types. But, sometimes you may want (or have) to explicitly
-select a particular overload variant.</p>
+When calling Java methods from Nashorn, the appropriate method will be
+selected based on the argument types at invocation time. You do not need
+to do anything special &ndash; the correct Java method overload variant 
+is selected based automatically. You still have the option of explicitly
+specifying a particular overload variant. Reasons for this include 
+either running into a genuine ambiguity with actual argument types, or 
+rarely reasons of performance &ndash; if you specify the actual overload
+then the engine doesn't have to perform resolution during invocation.
+Individual overloads of a Java methods are exposed as special properties
+with the name of the method followed with its signature in parentheses. 
+You can invoke them like this:</p>
 <pre><code>
 // <a href="source/overload.js">overload.js</a>
 
 var out = java.lang.System.out;
 
 // select a particular print function 
-out["println(java.lang.Object)"]("hello");
+out["println(Object)"]("hello");
 </code>
 </pre>
+<p>
+Note that you normally don't even have to use qualified class names in 
+the signatures as long as the unqualified name of the type is sufficient
+for uniquely identifying the signature. In practice this means that only
+in the extremely unlikely case that two overloads only differ in 
+parameter types that have identical unqualified names but come from 
+different packages would you need to use the fully qualified name of the
+class.
+</p>
+<hr>
+<a name="dataTypeMapping" id="dataTypeMapping"></a>
+<h3>Mapping of Data Types Between Java and JavaScript</h3>
+<p>
+We have previously shown some of the data type mappings between Java and JavaScript.
+We saw that arrays need to be explicitly converted. We have also shown that JavaScript functions
+are automatically converted to SAM types when passed as parameters to Java methods. Most other
+conversions work as you would expect.
+</p>
+<p>
+Every JavaScript object is also a <code>java.util.Map</code> so APIs receiving maps will receive them directly.
+</p>
+<p>
+When numbers are passed to a Java API, they will be converted to the expected target numeric type, either boxed or
+primitive, but if the target type is less specific, say <code>Number</code> or <code>Object</code>, you can only
+count on them being a <code>Number</code>, and have to test specifically for whether it's a boxed <code>Double</code>,
+<code>Integer</code>, <code>Long</code>, etc. &ndash; it can be any of these due to internal optimizations. Also, you 
+can pass any JavaScript value to a Java API expecting either a boxed or primitive number; the JavaScript specification's
+<code>ToNumber</code> conversion algorithm will be applied to the value.
+</p>
+<p>
+In a similar vein, if a Java method expects a <code>String</code> or a <code>Boolean</code>, the values will be
+converted using all conversions allowed by the JavaScript specification's <code>ToString</code> and <code>ToBoolean</code>
+conversions.
+</p>
+<p>
+Finally, a word of caution about strings. Due to internal performance optimizations of string operations, JavaScript strings are
+not always necessarily of type <code>java.lang.String</code>, but they will always be of type <code>java.lang.CharSequence</code>.
+If you pass them to a Java method that expects a <code>java.lang.String</code> parameter, then you will naturally receive a Java
+String, but if the signature of your method is more generic, i.e. it receives a <code>java.lang.Object</code> parameter, you can 
+end up with an object of private engine implementation class that implements <code>CharSequence</code> but is not a Java String.
+</p>
 <hr>
 <a name="engineimpl" id="engineimpl"></a>
 <h2>Implementing Your Own Script Engine</h2>