The following code sample shows a very simple EventListener. It basically just sends all events to System.out (stdout) or System.err (stderr) depending on the event severity.

You can see that for every event the method processEvent of the EventListener will be called. Inside this method you can do whatever processing you would like including throwing a RuntimeException, if you want to abort the current processing run.

The code above also shows how you can turn an event into a human-readable, localized message that can be presented to a user. The EventFormatter class does this for you. It provides additional methods if you'd like to explicitly specify the locale.

It is possible to gather all events for a whole processing run so they can be evaluated afterwards. However, care should be taken about memory consumption since the events provide references to objects inside FOP which may themselves have references to other objects. So holding on to these objects may mean that whole object trees cannot be released!

To register the event listener with FOP, get the EventBroadcaster which is associated with the user agent (FOUserAgent) and add it there:

Please note that this is done separately for each processing run, i.e. for each new user agent.

Here's an additional example of an event listener:

By default, FOP continues processing even if an image wasn't found. If you have more strict requirements and want FOP to stop if an image is not available, you can do something like the following in the simplest case:

Increasing the event severity to FATAL will signal the event broadcaster to throw an exception and stop further processing. In the above case, all resource-related events will cause FOP to stop processing.

You can also customize the exception to throw (you can may throw a RuntimeException or subclass yourself) and/or which event to respond to:

This throws a RuntimeException with the FileNotFoundException as the cause. Further processing effectively stops in FOP. You can catch the exception in your code and react as you see necessary.

The basics are very simple. Just instantiate an Event object and fill it with the necessary parameters. Then pass it to the EventBroadcaster which distributes the events to the interested listeneners. Here's a code example:

The Event.paramsBuilder() is a fluent interface to help with the build-up of the parameters. You could just as well instantiate a Map (Map<String, Object>) and fill it with values.

To simplify event production, the event subsystem provides the EventProducer interface. You can create interfaces which extend EventProducer. These interfaces will contain one method per event to be generated. By contract, each event method must have as its first parameter a parameter named "source" (Type Object) which indicates the object that generated the event. After that come an arbitrary number of parameters of any type as needed by the event.

The event producer interface does not need to have any implementation. The implementation is produced at runtime by a dynamic proxy created by DefaultEventBroadcaster. The dynamic proxy creates Event instances for each method call against the event producer interface. Each parameter (except "source") is added to the event's parameter map.

To simplify the code needed to get an instance of the event producer interface it is suggested to create a public inner provider class inside the interface.

Here's an example of such an event producer interface:

To produce the same event as in the first example above, you'd use the following code:

Inside an invocation handler for a dynamic proxy, there's no information about the names of each parameter. The JVM doesn't provide it. The only thing you know is the interface and method name. In order to properly fill the Event's parameter map we need to know the parameter names. These are retrieved from an event object model. This is found in the org.apache.fop.events.model package. The data for the object model is retrieved from an XML representation of the event model that is loaded as a resource. The XML representation is generated using an Ant task at build time (ant resourcegen). The Ant task (found in src/codegen/java/org/apache/fop/tools/EventProducerCollectorTask.java) scans FOP's sources for descendants of the EventProducer interface and uses QDox to parse these interfaces.

The event model XML files are generated during build by the Ant task mentioned above when running the "resourcegen" task. So just run "ant resourcegen" if you receive a MissingResourceException at runtime indicating that "event-model.xml" is missing.

Primarily, the QDox-based collector task records the parameters' names and types. Furthermore, it extracts additional attributes embedded as Javadoc comments from the methods. At the moment, the only such attribute is "@event.severity" which indicates the default event severity (which can be changed by event listeners). The example event producer above shows the Javadocs for an event method.

There's one more information that is extracted from the event producer information for the event model: an optional primary exception. The first exception in the "throws" declaration of an event method is noted. It is used to throw an exception from the invocation handler if the event has an event severity of "FATAL" when all listeners have been called (listeners can update the event severity). Please note that an implementation of org.apache.fop.events.EventExceptionManager$ExceptionFactory has to be registered for the EventExceptionManager to be able to construct the exception from an event.

For a given application, there can be multiple event models active at the same time. In FOP, each renderer is considered to be a plug-in and provides its own specific event model. The individual event models are provided through an EventModelFactory. This interface is implemented for each event model and registered through the service provider mechanism (see the plug-ins section for details).

Four different levels of severity for events has been defined:

1. INFO: informational only
2. WARN: a Warning
3. ERROR: an error condition from which FOP can recover. FOP will continue processing.
4. FATAL: a fatal error which causes an exception in the end and FOP will stop processing.

Event listeners can choose to ignore certain events based on their event severity. Please note that you may recieve an event "twice" in a specific case: if there is a fatal error an event is generated and sent to the listeners. After that an exception is thrown with the same information and processing stops. If the fatal event is shown to the user and the following exception is equally presented to the user it may appear that the event is duplicated. Of course, the same information is just published through two different channels.

The event subsystem is extensible. There are a number of extension points:

• org.apache.fop.events.model.EventModelFactory: Provides an event model to the event subsystem.
• org.apache.fop.events.EventExceptionManager$ExceptionFactory: Creates exceptions for events, i.e. turns an event into a specific exception.

The names in bold above are used as filenames for the service provider files that are placed in the META-INF/services directory. That way, they are automatically detected. This is a mechanism defined by the JAR file specification.

One goal of the event subsystem was to have localized (translated) event messages. The EventFormatter class can be used to convert an event to a human-readable message. Each EventProducer can provide its own XML-based translation file. If there is none, a central translation file is used, called "EventFormatter.xml" (found in the same directory as the EventFormatter class).

The XML format used by the EventFormatter is the same as Apache Cocoon's catalog format. Here's an example:

The example (extracted from the RTF handler's event producer) has message templates for two event methods. The class used to do variable replacement in the templates is org.apache.fop.util.text.AdvancedMessageFormat which is more powerful than the MessageFormat classes provided by the Java class library (java.util.text package).

"locator" is a template that is reused by the other message templates by referencing it through "{{locator}}". This is some kind of include command.

Normal event parameters are accessed by name inside single curly braces, for example: "{node}". For objects, this format just uses the toString() method to turn the object into a string, unless there is an ObjectFormatter registered for that type (there's an example for org.xml.sax.Locator).

The single curly braces pattern supports additional features. For example, it is possible to do this: "{start,if,start,end}". "if" here is a special field modifier that evaluates "start" as a boolean and if that is true returns the text right after the second comma ("start"). Otherwise it returns the text after the third comma ("end"). The "equals" modifier is similar to "if" but it takes as an additional (comma-separated) parameter right after the "equals" modifier, a string that is compared to the value of the variable. An example: {severity,equals,EventSeverity:FATAL,,some text} (this adds "some text" if the severity is not FATAL).

Additional such modifiers can be added by implementing the AdvancedMessageFormat$Part and AdvancedMessageFormat$PartFactory interfaces.

Square braces can be used to specify optional template sections. The whole section will be omitted if any of the variables used within are unavailable. Pipe (|) characters can be used to specify alternative sub-templates (see "locator" above for an example).

Developers can also register a function (in the above example: {#gatherContextInfo}) to do more complex information rendering. These functions are implementations of the AdvancedMessageFormat$Function interface. Please take care that this is done in a locale-independent way as there is no locale information available, yet.