Developing with Apache Ant

Writing Your Own Task

It is very easy to write your own task:

  1. Create a Java class that extends org.apache.tools.ant.Task or another class that was designed to be extended.
  2. For each attribute, write a setter method. The setter method must be a public void method that takes a single argument. The name of the method must begin with set, followed by the attribute name, with the first character of the name in uppercase, and the rest in lowercase*. That is, to support an attribute named file you create a method setFile. Depending on the type of the argument, Ant will perform some conversions for you, see below.
  3. If your task shall contain other tasks as nested elements (like parallel), your class must implement the interface org.apache.tools.ant.TaskContainer. If you do so, your task can not support any other nested elements. See below.
  4. If the task should support character data (text nested between the start and end tags), write a public void addText(String) method. Note that Ant does not expand properties on the text it passes to the task.
  5. For each nested element, write a create, add or addConfigured method. A create method must be a public method that takes no arguments and returns an Object type. The name of the create method must begin with create, followed by the element name. An add (or addConfigured) method must be a public void method that takes a single argument of an Object type with a no-argument constructor. The name of the add (addConfigured) method must begin with add (addConfigured), followed by the element name. For a more complete discussion see below.
  6. Write a public void execute() method, with no arguments, that throws a BuildException. This method implements the task itself.

* Actually the case of the letters after the first one doesn't really matter to Ant, using all lower case is a good convention, though.

The Life-cycle of a Task

  1. The xml element that contains the tag corresponding to the task gets converted to an UnknownElement at parse time. This UnknownElement gets placed in a list within a target object, or recursively within another UnknownElement.
  2. When the target is executed, each UnknownElement is invoked using an perform() method. This instantiates the task. This means that tasks only gets instantiated at run time.
  3. The task gets references to its project and location inside the buildfile via its inherited project and location variables.
  4. If the user specified an id attribute to this task, the project registers a reference to this newly created task, at run time.
  5. The task gets a reference to the target it belongs to via its inherited target variable.
  6. init() is called at run time.
  7. All child elements of the XML element corresponding to this task are created via this task's createXXX() methods or instantiated and added to this task via its addXXX() methods, at run time. Child elements corresponding to addConfiguredXXX() are created at this point but the actual addConfigured method is not called.
  8. All attributes of this task get set via their corresponding setXXX() methods, at run time.
  9. The content character data sections inside the XML element corresponding to this task is added to the task via its addText() method, at run time.
  10. All attributes of all child elements get set via their corresponding setXXX() methods, at run time.
  11. If child elements of the XML element corresponding to this task have been created for addConfiguredXXX() methods, those methods get invoked now.
  12. execute() is called at run time. If target1 and target2 both depend on target3, then running ant target1 target2 will run all tasks in target3 twice.

Conversions Ant will perform for attributes

Ant will always expand properties before it passes the value of an attribute to the corresponding setter method. Since Ant 1.8, it is possible to extend Ant's property handling such that a non-string Object may be the result of the evaluation of a string containing a single property reference. These will be assigned directly via setter methods of matching type. Since it requires some beyond-the-basics intervention to enable this behavior, it may be a good idea to flag attributes intended to permit this usage paradigm.

The most common way to write an attribute setter is to use a java.lang.String argument. In this case Ant will pass the literal value (after property expansion) to your task. But there is more! If the argument of you setter method is

What happens if more than one setter method is present for a given attribute? A method taking a String argument will always lose against the more specific methods. If there are still more setters Ant could chose from, only one of them will be called, but we don't know which, this depends on the implementation of your Java virtual machine.

Supporting nested elements

Let's assume your task shall support nested elements with the name inner. First of all, you need a class that represents this nested element. Often you simply want to use one of Ant's classes like org.apache.tools.ant.types.FileSet to support nested fileset elements.

Attributes of the nested elements or nested child elements of them will be handled using the same mechanism used for tasks (i.e. setter methods for attributes, addText() for nested text and create/add/addConfigured methods for child elements).

Now you have a class NestedElement that is supposed to be used for your nested <inner> elements, you have three options:

  1. public NestedElement createInner()
  2. public void addInner(NestedElement anInner)
  3. public void addConfiguredInner(NestedElement anInner)

What is the difference?

Option 1 makes the task create the instance of NestedElement, there are no restrictions on the type. For the options 2 and 3, Ant has to create an instance of NestedInner before it can pass it to the task, this means, NestedInner must have a public no-arg constructor or a public one-arg constructor taking a Project class as a parameter. This is the only difference between options 1 and 2.

The difference between 2 and 3 is what Ant has done to the object before it passes it to the method. addInner() will receive an object directly after the constructor has been called, while addConfiguredInner() gets the object after the attributes and nested children for this new object have been handled.

What happens if you use more than one of the options? Only one of the methods will be called, but we don't know which, this depends on the implementation of your JVM.

Nested Types

If your task needs to nest an arbitrary type that has been defined using <typedef> you have two options.

  1. public void add(Type type)
  2. public void addConfigured(Type type)

The difference between 1 and 2 is the same as between 2 and 3 in the previous section.

For example suppose one wanted to handle objects object of type org.apache.tools.ant.taskdefs.condition.Condition, one may have a class:

public class MyTask extends Task {
    private List conditions = new ArrayList();
    public void add(Condition c) {
        conditions.add(c);
    }
    public void execute() {
     // iterator over the conditions
    }
}

One may define and use this class like this:

<taskdef name="mytask" classname="MyTask" classpath="classes"/>
<typedef name="condition.equals"
         classname="org.apache.tools.ant.taskdefs.conditions.Equals"/>
<mytask>
    <condition.equals arg1="${debug}" arg2="true"/>
</mytask>

A more complicated example follows:

public class Sample {
    public static class MyFileSelector implements FileSelector {
         public void setAttrA(int a) {}
         public void setAttrB(int b) {}
         public void add(Path path) {}
         public boolean isSelected(File basedir, String filename, File file) {
             return true;
         }
     }

    interface MyInterface {
        void setVerbose(boolean val);
    }

    public static class BuildPath extends Path {
        public BuildPath(Project project) {
            super(project);
        }

        public void add(MyInterface inter) {}
        public void setUrl(String url) {}
    }

    public static class XInterface implements MyInterface {
        public void setVerbose(boolean x) {}
        public void setCount(int c) {}
    }
}

This class defines a number of static classes that implement/extend Path, MyFileSelector and MyInterface. These may be defined and used as follows:

<typedef name="myfileselector" classname="Sample$MyFileSelector"
         classpath="classes" loaderref="classes"/>
<typedef name="buildpath" classname="Sample$BuildPath"
         classpath="classes" loaderref="classes"/>
<typedef name="xinterface" classname="Sample$XInterface"
         classpath="classes" loaderref="classes"/>

<copy todir="copy-classes">
   <fileset dir="classes">
      <myfileselector attra="10" attrB="-10">
         <buildpath path="." url="abc">
            <xinterface count="4"/>
         </buildpath>
      </myfileselector>
   </fileset>
</copy>

TaskContainer

The TaskContainer consists of a single method, addTask that basically is the same as an add method for nested elements. The task instances will be configured (their attributes and nested elements have been handled) when your task's execute method gets invoked, but not before that.

When we said execute would be called, we lied ;-). In fact, Ant will call the perform method in org.apache.tools.ant.Task, which in turn calls execute. This method makes sure that Build Events will be triggered. If you execute the task instances nested into your task, you should also invoke perform on these instances instead of execute.

Example

Let's write our own task, which prints a message on the System.out stream. The task has one attribute, called message.

package com.mydomain;

import org.apache.tools.ant.BuildException;
import org.apache.tools.ant.Task;

public class MyVeryOwnTask extends Task {
    private String msg;

    // The method executing the task
    public void execute() throws BuildException {
        System.out.println(msg);
    }

    // The setter for the "message" attribute
    public void setMessage(String msg) {
        this.msg = msg;
    }
}

It's really this simple ;-)

Adding your task to the system is rather simple too:

  1. Make sure the class that implements your task is in the classpath when starting Ant.
  2. Add a <taskdef> element to your project. This actually adds your task to the system.
  3. Use your task in the rest of the buildfile.

Example

<?xml version="1.0"?>

<project name="OwnTaskExample" default="main" basedir=".">
  <taskdef name="mytask" classname="com.mydomain.MyVeryOwnTask"/>

  <target name="main">
    <mytask message="Hello World! MyVeryOwnTask works!"/>
  </target>
</project>

Example 2

To use a task directly from the buildfile which created it, place the <taskdef> declaration inside a target after the compilation. Use the classpath attribute of <taskdef> to point to where the code has just been compiled.

<?xml version="1.0"?>

<project name="OwnTaskExample2" default="main" basedir=".">

  <target name="build" >
    <mkdir dir="build"/>
    <javac srcdir="source" destdir="build"/>
  </target>

  <target name="declare" depends="build">
    <taskdef name="mytask"
        classname="com.mydomain.MyVeryOwnTask"
        classpath="build"/>
  </target>

  <target name="main" depends="declare">
    <mytask message="Hello World! MyVeryOwnTask works!"/>
  </target>
</project>

Another way to add a task (more permanently) is to add the task name and implementing class name to the default.properties file in the org.apache.tools.ant.taskdefs package. Then you can use it as if it were a built-in task.

Build Events

Ant is capable of generating build events as it performs the tasks necessary to build a project. Listeners can be attached to Ant to receive these events. This capability could be used, for example, to connect Ant to a GUI or to integrate Ant with an IDE.

To use build events you need to create an ant Project object. You can then call the addBuildListener method to add your listener to the project. Your listener must implement the org.apache.tools.antBuildListener interface. The listener will receive BuildEvents for the following events

If the build file invokes another build file via <ant> or <subant> or uses <antcall>, you are creating a new Ant "project" that will send target and task level events of its own but never sends build started/finished events. Since Ant 1.6.2, BuildListener interface has an extension named SubBuildListener that will receive two new events for

If you are interested in those events, all you need to do is to implement the new interface instead of BuildListener (and register the listener, of course).

If you wish to attach a listener from the command line you may use the -listener option. For example:

ant -listener org.apache.tools.ant.XmlLogger

will run Ant with a listener that generates an XML representation of the build progress. This listener is included with Ant, as is the default listener, which generates the logging to standard output.

Note: A listener must not access System.out and System.err directly since output on these streams is redirected by Ant's core to the build event system. Accessing these streams can cause an infinite loop in Ant. Depending on the version of Ant, this will either cause the build to terminate or the JVM to run out of Stack space. A logger, also, may not access System.out and System.err directly. It must use the streams with which it has been configured.

Note: All methods of a BuildListener except for the "Build Started" and "Build Finished" events may occur on several threads simultaneously—for example while Ant is executing a <parallel> task.

Example

Writing an adapter to your favourite log library is very easy. Just implement the BuildListener interface, instantiate your logger and delegate the message to that instance.

When starting your build provide your adapter class and the log library to the build classpath and activate your logger via -listener option as described above.

public class MyLogAdapter implements BuildListener {

    private MyLogger getLogger() {
        final MyLogger log = MyLoggerFactory.getLogger(Project.class.getName());
        return log;
    }

    @Override
    public void buildStarted(final BuildEvent event) {
        final MyLogger log = getLogger();
        log.info("Build started.");
    }

    @Override
    public void buildFinished(final BuildEvent event) {
        final MyLogger logger = getLogger();
        MyLogLevelEnum loglevel = ... // map event.getPriority() to enum via Project.MSG_* constants
        boolean allOK = event.getException() == null;
        String logmessage = ... // create log message using data of the event and the message invoked
        logger.log(loglevel, logmessage);
    }

    // implement all methods in that way
}

Source code integration

The other way to extend Ant through Java is to make changes to existing tasks, which is positively encouraged. Both changes to the existing source and new tasks can be incorporated back into the Ant codebase, which benefits all users and spreads the maintenance load around.

Please consult the Getting Involved pages on the Apache web site for details on how to fetch the latest source and how to submit changes for reincorporation into the source tree.

Ant also has some task guidelines which provides some advice to people developing and testing tasks. Even if you intend to keep your tasks to yourself, you should still read this as it should be informative.