This package exposes a kind of XML processing pipeline, based on sending SAX events, which can be used as components of application architectures. Pipelines are used to convey streams of processing events from a producer to one or more consumers, and to let each consumer control the data seen by later consumers.

There is a PipelineFactory class which accepts a syntax describing how to construct some simple pipelines. Strings describing such pipelines can be used in command line tools (see the DoParse class) and in other places that it is useful to let processing be easily reconfigured. Pipelines can of course be constructed programmatically, providing access to options that the factory won't.

Web applications are supported by making it easy for servlets (or non-Java web application components) to be part of a pipeline. They can originate XML (or XHTML) data through an InputSource or in response to XML messages sent from clients using CallFilter pipeline stages. Such facilities are available using the simple syntax for pipeline construction.

Programming Models

Pipelines should be simple to understand.

• XML content, typically entire documents, is pushed through consumers by producers.
• Pipelines are basically about consuming SAX2 callback events, where the events encapsulate XML infoset-level data.
  • Pipelines are constructed by taking one or more consumer stages and combining them to produce a composite consumer.
  • A pipeline is presumed to have pending tasks and state from the beginning of its ContentHandler.startDocument() callback until it's returned from its ContentHandler.doneDocument() callback.
  • Pipelines may have multiple output stages ("fan-out") or multiple input stages ("fan-in") when appropriate.
  • Pipelines may be long-lived, but need not be.
• There is flexibility about event production.
  • SAX2 XMLReader objects are producers, which provide a high level "pull" model: documents (text or DOM) are parsed, and the parser pushes individual events through the pipeline.
  • Events can be pushed directly to event consumer components by application modules, if they invoke SAX2 callbacks directly. That is, application modules use the XML Infoset as exposed through SAX2 event callbacks.
• Multiple producer threads may concurrently access a pipeline, if they coordinate appropriately.
• Pipeline processing is not the only framework applications will use.

Producers: XMLReader or Custom

Many producers will be SAX2 XMLReader objects, and will read (pull) data which is then written (pushed) as events. Typically these will parse XML text (acquired from org.xml.sax.helpers.XMLReaderFactory) or a DOM tree (using a DomParser) These may be bound to event consumer using a convenience routine, EventFilter.bind(). Once bound, these producers may be given additional documents to sent through its pipeline.

In other cases, you will write producers yourself. For example, some data structures might know how to write themselves out using one or more XML models, expressed as sequences of SAX2 event callbacks. An application module might itself be a producer, issuing startDocument and endDocument events and then asking those data structures to write themselves out to a given EventConsumer, or walking data structures (such as JDBC query results) and applying its own conversion rules. WAP format XML (WBMXL) can be directly converted to producer output.

SAX2 introduced an "XMLFilter" interface, which is a kind of XMLReader. It is most useful in conjunction with its XMLFilterImpl helper class; see the EventFilter javadoc for information contrasting that XMLFilterImpl approach with the relevant parts of this pipeline framework. Briefly, such XMLFilterImpl children can be either producers or consumers, and are more limited in configuration flexibility. In this framework, the focus of filters is on the EventConsumer side; see the section on pipe fitting below.

Consume to Standard or Custom Data Representations

Many consumers will be used to create standard representations of XML data. The TextConsumer takes its events and writes them as text for a single XML document, using an internal XMLWriter. The DomConsumer takes its events and uses them to create and populate a DOM Document.

In other cases, you will write consumers yourself. For example, you might use a particular unmarshaling filter to produce objects that fit your application's requirements, instead of using DOM. Such consumers work at the level of XML data models, rather than with specific representations such as XML text or a DOM tree. You could convert your output directly to WAP format data (WBXML).

Pipe Fitting

Pipelines are composite event consumers, with each stage having the opportunity to transform the data before delivering it to any subsequent stages.

The PipelineFactory class provides access to much of this functionality through a simple syntax. See the table in that class's javadoc describing a number of standard components. Direct API calls are still needed for many of the most interesting pipeline configurations, including ones leveraging actual or logical concurrency.

Four basic types of pipe fitting are directly supported. These may be used to construct complex pipeline networks.

• TeeConsumer objects split event flow so it goes to two two different consumers, one before the other. This is a basic form of event fan-out; you can use this class to copy events to any number of output pipelines.
• Clients can call remote components through HTTP or HTTPS using the CallFilter component, and Servlets can implement such components by extending the XmlServlet component. Java is not required on either end, and transport protocols other than HTTP may also be used.
• EventFilter objects selectively provide handling for callbacks, and can pass unhandled ones to a subsequent stage. They are often subclassed, since much of the basic filtering machinery is already in place in the base class.
• Applications can merge two event flows by just using the same consumer in each one. If multiple threads are in use, synchronization needs to be addressed by the appropriate application level policy.

Note that filters can be as complex as XSLT transforms available) on input data, or as simple as removing simple syntax data such as ignorable whitespace, comments, and CDATA delimiters. Some simple "built-in" filters are part of this package.

Coding Conventions: Filter and Terminus Stages

If you follow these coding conventions, your classes may be used directly (give the full class name) in pipeline descriptions as understood by the PipelineFactory. There are four constructors the factory may try to use; in order of decreasing numbers of parameters, these are:

• Filters that need a single String setup parameter should have a public constructor with two parameters: that string, then the EventConsumer holding the "next" consumer to get events.
• Filters that don't need setup parameters should have a public constructor that accepts a single EventConsumer holding the "next" consumer to get events when they are done.
• Terminus stages may have a public constructor taking a single paramter: the string value of that parameter.
• Terminus stages may have a public no-parameters constructor.

Of course, classes may support more than one such usage convention; if they do, they can automatically be used in multiple modes. If you try to use a terminus class as a filter, and that terminus has a constructor with the appropriate number of arguments, it is automatically wrapped in a "tee" filter.

Debugging Tip: "Tee" Joints can Snapshot Data

It can sometimes be hard to see what's happening, when something goes wrong. Easily fixed: just snapshot the data. Then you can find out where things start to go wrong.

If you're using pipeline descriptors so that they're easily administered, just stick a write ( filename ) filter into the pipeline at an appropriate point.

Inside your programs, you can do the same thing directly: perhaps by saving a Writer (perhaps a StringWriter) in a variable, using that to create a TextConsumer, and making that the first part of a tee -- splicing that into your pipeline at a convenient location.

You can also use a DomConsumer to buffer the data, but remember that DOM doesn't save all the information that XML provides, so that DOM snapshots are relatively low fidelity. They also are substantially more expensive in terms of memory than a StringWriter holding similar data.

Debugging Tip: Non-XML Producers

Producers in pipelines don't need to start from XML data structures, such as text in XML syntax (likely coming from some XMLReader that parses XML) or a DOM representation (perhaps with a DomParser).

One common type of event producer will instead make direct calls to SAX event handlers returned from an EventConsumer. For example, making ContentHandler.startElement calls and matching ContentHandler.endElement calls.

Applications making such calls can catch certain common "syntax errors" by using a WellFormednessFilter. That filter will detect (and report) erroneous input data such as mismatched document, element, or CDATA start/end calls. Use such a filter near the head of the pipeline that your producer feeds, at least while debugging, to help ensure that you're providing legal XML Infoset data.

You can also arrange to validate data on the fly. For DTD validation, you can configure a ValidationConsumer to work as a filter, using any DTD you choose. Other validation schemes can be handled with other validation filters.