Saxon home page

Extensibility


This page describes how to extend the capability of Saxon XSLT Stylesheets

Contents
Writing extension functions
Writing extension instructions
Writing input filters
Writing output filters
Implementing a collating sequence
Implementing a numbering sequence
Adding an output encoding
Writing a URI Resolver for input files
Writing a URI Resolver for output files

Writing extension functions

An extension function is invoked using a name such as prefix:localname(). The prefix must be the prefix associated with a namespace declaration that is in scope.

Extension functions must be implemented in Java.

The command line option -TJ is useful for debugging the loading of Java extensions. It gives detailed information about the methods that are examined for a possible match.

Saxon supports the <saxon:script> element, based on the <xsl:script> element defined in the (now withdrawn) XSLT 1.1 working draft. This element defines a mapping between a namespace URI used in calls of extension functions, and a Java class that contains implementations of these functions. See saxon:script for details.

You can also use a short-cut technique of binding external Java classes, by making the class name part of the namespace URI. In this case, you don't need a <saxon:script> element.

With the short-cut technique, the URI for the namespace identifies the class where the external function will be found. The namespace URI must either be "java:" followed by the fully-qualified class name (for example xmlns:date="java:java.util.Date"), or a string containing a "/", in which the fully-qualified class name appears after the final "/". (for example xmlns:date="http://www.jclark.com/xt/java/java.util.Date"). The part of the URI before the final "/" is immaterial. The class must be on the classpath. For compatibility with previous releases, the format xmlns:date="java.util.Date" is also supported.

The Saxon namespace URI "http://saxon.sf.net/" is recognised as a special case, and causes the function to be loaded from the class net.sf.saxon.functions.Extensions. This class name can be specified explicitly if you prefer.

There are three cases to consider: static methods, constructors, and instance-level methods. In addition, a public field in a class is treated as if it were a zero-argument method, so public static fields can be accessed in the same way as public static methods, and public instance-level fields in the same way as instance-level methods.

Static methods can be called directly. The localname of the function must match the name of a public static method in this class. The names match if they contain the same characters, excluding hyphens and forcing any character that follows a hyphen to upper-case. For example the XPath function call to-string() matches the Java method toString(); but the function call can also be written as toString() if you prefer.

If there are several methods in the class that match the localname, and that have the correct number of arguments, then the system attempts to find the one that is the best fit to the types of the supplied arguments: for example if the call is f(1,2) then a method with two int arguments will be preferred to one with two String arguments. The rules for deciding between methods are quite complex. Essentially, for each candidate method, Saxon calculates the "distance" between the types of the supplied arguments and the Java class of the corresponding method in the method's signature, using a set of tables given below. For example, the distance between the XPath data type "boolean" and the Java class "boolean" is very small, while the distance between an XPath string and a Java boolean is much larger. If there is one candidate method where the distances of all arguments are less-than-or-equal-to the distances computed for other candidate methods, and the distance of at least one argument is smaller, then that method is chosen. If there are several methods with the same name and the correct number of arguments, but none is preferable to the others under these rules, an error is reported: the message indicates that there is more than one method that matches the function call.

For example:


<xsl:value-of select="math:sqrt($arg)"
   xmlns:math="java:java.lang.Math"/>

This will invoke the static method java.lang.Math#sqrt(), applying it to the value of the variable $arg, and copying the value of the square root of $arg to the result tree.

Similarly:


<xsl:value-of select="double:MAX_VALUE()"
   xmlns:double="java:java.lang.Double"/>

This will output the value of the static field java.lang.Double#MAX_VALUE.

Java constructors are called by using the function named new(). If there are several constructors, then again the system tries to find the one that is the best fit, according to the types of the supplied arguments. The result of calling new() is an XPath value whose type is denoted by the QName java:java.lang.Object, where the namespace prefix java represents the namespace URI http://saxon.sf.net/java-type. The only things that can be done with a Java Object are to assign it to a variable, to pass it to an extension function, and to convert it to a string, number, or boolean, using the rules given below. The name java:java.lang.Object has been chosen to allow finer-grained references to Java classes in the future, but at the moment the local name java.lang.Object is the only name recognized in this namespace.

Instance-level methods (that is, non-static methods) are called by supplying an extra first argument of type Java Object which is the object on which the method is to be invoked. A Java Object is usually created by calling an extension function (e.g. a constructor) that returns an object; it may also be passed to the style sheet as the value of a global parameter. Matching of method names is done as for static methods. If there are several methods in the class that match the localname, the system again tries to find the one that is the best fit, according to the types of the supplied arguments.

For example, the following stylesheet prints the date and time. This example is copied from the documentation of the xt product, and it works unchanged with Saxon, because Saxon does not care what the namespace URI for extension functions is, so long as it ends with the class name. (Extension functions are likely to be compatible between Saxon and xt provided they only use the data types string, number, and boolean).


<xsl:stylesheet
  version="1.0"
  xmlns:xsl="http://www.w3.org/1999/XSL/Transform"
  xmlns:date="http://www.jclark.com/xt/java/java.util.Date">

<xsl:template match="/">
  <html>
    <xsl:if test="function-available('date:to-string') and function-available('date:new')">
      <p><xsl:value-of select="date:to-string(date:new())"/></p>
    </xsl:if>
  </html>
</xsl:template>

</xsl:stylesheet>

A Java method called as an extension function may have an extra first argument of class net.sf.saxon.expr.XPathContext. This argument is not supplied by the calling XSLT code, but by Saxon itself. The XPathContext object provides methods to access many internal Saxon resources, the most useful being current() which returns the current Item. The XPathContext object is not available with constructors.

If any exceptions are thrown by the method, or if a matching method cannot be found, processing of the stylesheet will be abandoned. If the tracing option has been set (-T) on the command line, a full stack trace will be output. The exception will be wrapped in a TransformerException and passed to any user-specified ErrorListener object, so the ErrorListener can also produce extra diagnostics.

Conversion of arguments to the function

The following conversions are supported between the supplied value of the argument and the declared Java class of the argument. The mappings are given in order of preference; a class that appears earlier in the list has smaller "conversion distance" than one appearing later. These priorities are used to decide which method to call when the class has several methods of the same name. Simple classes (such as boolean) are acceptable wherever the corresponding wrapper class (Boolean) is allowed. Class names shown in italics are Saxon-specific classes.

If the supplied value is a singleton (a sequence of one item) then the type of that item is decisive. If it is a sequence of length zero or more than one, then the general rules for a sequence are applied, and the types of the items within the sequence are irrelevant.

Supplied valueRequired type
booleanBooleanValue, Boolean, (String, CharSequence), Byte, Character, Double, Float, Integer, Long, Short, Byte), Object
dateTimeDateTimeValue, Date, String, Object
dateDateValue, Date, (String, CharSequence), Object
decimalDecimalValue, BigDecimal, Double, Float, Long, Integer, Short, Character, Byte, Boolean, (String, CharSequence), Object
doubleDoubleValue, Double, Float, Long, Integer, Short, Character, Byte, Boolean, (String, CharSequence), Object
durationDurationValue, (String, CharSequence), Object
floatFloatValue, Float, Double, Long, Integer, Short, Character, Byte, Boolean, (String, CharSequence), Object
integerIntegerValue, Long, Double, Float, Integer, Short, Character, Byte, Boolean, (String, CharSequence), Object
stringStringValue, (String, CharSequence), Character, Double, Float, Integer, Long, Short, Boolean, Byte, Object
anyURIAnyURIValue, java.net.URI, java.net.URL, (String, CharSequence), Object
nodeSingletonNodeSet, NodeList, (Element, Attr, Document, DocumentFragment, Comment, Text, ProcessingInstruction, CharacterData), Node, Boolean, Byte, Character, Double, Float, Integer, Long, Short, (String, CharSequence), Object
sequenceSequenceIterator, SequenceValue, List, NodeList, NodeInfo, Node, (String, CharSequence), Boolean, Byte, Character, Double, Float, Integer, Long, Short, Object

Saxon first tries to select the appropriate method based on the static type of the arguments to the function call. If there is insufficient information statically, it tries again at run-time, based on the dynamic type of the arguments once evaluated. This means that the same function call may invoke different methods on different occasions.

Note that the XPath value is considered to be one of the singleton classes if it is produced by an expression that always produces a singleton. So the expression 1+2 will be an integer. An expression that can potentially produce a sequence of any cardinality is represented in the table above by the generic class "sequence". For example, ($a to $b) is represented as a sequence, except when it can be determined statically that $a and $b are equal, in which case it is represented as an integer.

These rules will probably be rationalized further in future releases.

A wrapped Java object may be converted to another data type as follows.

Result of an extension function

The result type of the method is converted to an XPath value as follows.

Note that Saxon's two principal tree structures both conform to the DOM Core Level 2 interface. However, they are read-only: any attempt to modify the tree causes an exception. Saxon's trees can only be built using the Saxon subclasses of the net.sf.saxon.tree.Builder class, and they cannot be modified in situ. (The simplest way for a Java application to build a Saxon tree is by using the net.sf.saxon.xpath.XPathEvaluator class.)

The system function function-available(String name) returns true if there appears to be a method available with the right name. It does not test whether this method has the appropriate number of arguments or whether the arguments are of appropriate types. If the function name is "new" it returns true so long as the class is not an abstract class or interface, and so long as it has at least one constructor.

There are a number of extension functions supplied with the Saxon product: for details, see extensions.html. The source code of these methods, which in most cases is extremely simple, can be used as an example for writing other user extension functions. It is found in class net.sf.saxon.functions.Extensions.

Writing extension instructions

Saxon implements the element extensibility feature defined in the XSLT standard. This feature allows you to define your own instruction types for use in the stylesheet. These instructions can be used anywhere within a content constructor, for example as a child of xsl:template, xsl:if, xsl:variable, or of a literal result element.

If a namespace prefix is to be used to denote extension elements, it must be declared in the extension-element-prefixes attribute on the xsl:stylesheet element, or the xsl:extension-element-prefixes attribute on any enclosing literal result element or extension element.

Note that Saxon itself provides a number of stylesheet elements beyond those defined in the XSLT specification, including saxon:assign, saxon:entity-ref, and saxon:while. To enable these, use the standard XSLT extension mechanism: define extension-element-prefixes="saxon" on the xsl:stylesheet element, or xsl:extension-element-prefixes="saxon" on any enclosing literal result element.

To invoke a user-defined set of extension elements, include the prefix in this attribute as described, and associate it with a namespace URI that ends in "/" followed by the fully qualified class name of a Java class that implements the net.sf.saxon.style.ExtensionElementFactory interface. This interface defines a single method, getExtensionClass(), which takes the local name of the element (i.e., the name without its namespace prefix) as a parameter, and returns the Java class used to implement this extension element (for example, "return SQLConnect.class"). The class returned must be a subclass of net.sf.saxon.style.StyleElement.

The best way to see how to implement an extension element is by looking at the example, for SQL extension elements, provided in package net.sf.saxon.sql, and at the sample stylesheet books-sql.xsl which uses these extension elements. The main methods a StyleElement class must provide are:

prepareAttributes() This is called while the stylesheet tree is still being built, so it should not attempt to navigate the tree. Its task is to validate the attributes of the stylesheet element and perform any preprocessing necessary. For example, if the attribute is an attribute value template, this includes creating an Expression that can subsequently be evaluated to get the AVT's value.
validate() This is called once the tree has been built, and its task is to check that the stylesheet element appears in the right context within the tree, e.g. that it is within a template
process() This is called to process a particular node in the source document, which can be accessed by reference to the XSLTContext object supplied as a parameter.
isInstruction() This should return true, to ensure that the element is allowed to appear within a template body.
mayContainTemplateBody() This should return true, to ensure that the element can contain instructions. Even if it can't contain anything else, extension elements should allow an xsl:fallback instruction to provide portability between processors

The StyleElement class has access to many services supplied either via its superclasses or via the XSLTContext object. For details, see the API documentation of the individual classes.

Any element whose prefix matches a namespace listed in the extension-element-prefixes attribute of an enclosing element is treated as an extension element. If no class can be instantiated for the element (for example, because no ExtensionElementFactory can be loaded, or because the ExtensionElementFactory doesn't recognise the local name), then fallback action is taken as follows. If the element has one or more xsl:fallback children, they are processed. Otherwise, an error is reported. When xsl:fallback is used in any other context, it and its children are ignored.

It is also possible to test whether an extension element is implemented by using the system function element-available(). This returns true if the namespace of the element identifies it as an extension element (or indeed as a standard XSLT instruction) and if a class can be instantiated to represent it. If the namespace is not that of an extension element, or if no class can be instantiated, it returns false.

Writing input filters

Saxon takes its input from a SAX2 Parser reading from an InputSource. A very useful technique is to interpose a filter between the parser and Saxon. The filter will typically be an instance of the SAX2 XMLFilter class.

See the TrAX examples for hints on using a Saxon Transformer as part of a chain of SAX Filters.

Note that Saxon relies on the application to supply a well-balanced sequence of SAX events; it doesn't need to be well-formed (the root node can have any number of element or text children), but if it isn't well-balanced, the consequences are unpredictable.

The -x option on the Saxon command line specifies the parser that Saxon will use to process the source files. This class must implement the SAX2 XMLReader interface, but it is not required to be a real XML parser; it can take the input from any kind of source file, so long as it presents it in the form of a stream of SAX events. When using the JAXP API, the equivalent to the -x option is to call transformerFactory.setAttribute( net.sf.saxon.FeatureKeys.SOURCE_PARSER_CLASS, 'com.example.package.Parser')

Writing output filters

The output of a Saxon stylesheet can be directed to a user-defined output filter. This filter can be defined either as a SAX2 ContentHandler, or as a subclass of the Saxon class net.sf.saxon.event.Emitter. The advantage of using an Emitter is that more information is available from the stylesheet, for example the attributes of the xsl:output element.

When a ContentHandler is used, Saxon will by default always supply a stream of events corresponding to a well-formed document. (The XSLT specification also allows the output to be an external general parsed entity, also known as a "well-balanced document".) If the result tree is not well-formed, Saxon will raise a dynamic error, unless the ContentHandler indicates that it is prepared to accept such a result tree, which it can do by implementing the marker interface net.sf.saxon.event.BalancedContentHandler. This interface has no methods.

As specified in the JAXP 1.1 interface, requests to disable or re-enable output escaping are also notified to the content handler by means of special processing instructions. The names of these processing instructions are defined by the constants PI_DISABLE_OUTPUT_ESCAPING and PI_ENABLE_OUTPUT_ESCAPING defined in class javax.xml.transform.Result.

If an Emitter is used, however, it will be informed of all events.

The Emitter or ContentHandler to be used is specified in the method attribute of the xsl:output element, as a fully-qualified class name; for example method="prefix:com.acme.xml.SaxonOutputFilter". The namespace prefix is ignored, but must be present to meet XSLT conformance rules.

See the documentation of class net.sf.saxon.output.Emitter for details of the methods available, or implementations such as HTMLEmitter and XMLEmitter and TEXTEmitter for the standard output formats supported by Saxon.

It can sometimes be useful to set up a chain of emitters working as a pipeline. To write a filter that participates in such a pipeline, the class ProxyEmitter is supplied. Use the class XMLIndenter, which handles XML indentation, as an example of how to write a ProxyEmitter.

Rather than writing an output filter in Java, Saxon also allows you to process the output through another XSLT stylesheet. To do this, simply name the next stylesheet in the saxon:next-in-chain attribute of xsl:output.

Any number of user-defined attributes may be defined on xsl:output. These attributes must have names in a non-null namespace, which must not be either the XSLT or the Saxon namespace. These attributes are interpreted as attribute value templates. The value of the attribute is inserted into the Properties object made available to the Emitter handling the output; they will be ignored by the standard output methods, but can supply arbitrary information to a user-defined output method. The name of the property will be the expanded name of the attribute in JAXP format, for example {http://my-namespace/uri}local-name, and the value will be the value as given, after evaluation as an attribute value template.

Implementing a collating sequence

It is possible to define a collating sequence for use by xsl:sort. This is controlled through the class attributes of the saxon:collation element. See saxon:collation for details.

Implementing a numbering sequence

It is possible to define a numbering sequence for use by xsl:number. This is controlled through the lang attribute of the xsl:number element. The feature is primarily intended to provide language-dependent numbering, but in fact it can be used to provide arbitrary numbering sequences: for example if you want to number items as "*", "†", "‡", "§", "¶" etc, you could implement a numbering class to do this and invoke it say with lang="x-footnote".

To implement a numberer for language X, you need to define a class net.sf.saxon.number.Numberer_X, for example net.sf.saxon.sort.Numberer_xfootnote. This must implement the interface Numberer. A (not very useful) Numberer is supplied for lang="de" as a specimen, and you can use this as a prototype to write your own. A numbering sequence is also supplied for lang="en", and this is used by default if no other can be loaded.

Note that any hyphens in the language name are ignored in forming the class name, but case is significant. For example if you specify lang="en-GB", the Numberer must be named net.sf.saxon.number.Numberer_enGB.

Adding an output encoding

If you want to use an output encoding that is not directly supported by Saxon (for a list of encodings that are supported, see conformance.html) you can do this by writing a Java class that implements the interface net.sf.saxon.charcode.PluggableCharacterSet. You need to supply two methods: inCharSet() which tests whether a particular Unicode character is present in the character set, and getEncodingName() which returns the name given to the encoding by your Java VM. The encoding must be supported by the Java VM. To use this encoding, specify the fully-qualified class name as the value of the encoding attribute in xsl:output.

Alternatively, it is possible to specify the CharacterSet class to be used for a named output encoding by setting the system property, e.g. -D"encoding.EUC-JP"="EUC_JP"; the value of the property should be the name of a class that implements the PluggableCharacterSet interface. This indicates the class to be used when the xsl:output element specifies encoding="EUC-JP".

Writing a URI Resolver for Input Files

Saxon allows you to write your own URIResolver to handle the URIs of input documents, as defined in the JAXP 1.1 specification. Such a URIResolver is used to process the URIs used in the xsl:include and xsl:import declarations as well as the document() function. It is also used to process the URIs supplied for the source document and the stylesheet on the command line. The URIResolver is called to process the supplied URI, and it returns a JAXP Source object, which Saxon uses as the source of the input. Note that the Source must be one of the implementations of Source that Saxon recognizes, you cannot write your own implementations of the JAXP 1.1 Source class.

Writing a URI Resolver for Output Files

Saxon also allows you to write an OutputURIResolver, which performs an analogous role for URIs specified in the href attribute of xsl:result-document. The OutputURIResolver is called when writing of the output document starts, at which point it must return a JAXP Result object to act as the output destination. It is called again when writing of an output document is complete.

You can nominate an OutputURIResolver by calling ((Controller)transformer).setOutputURIResolver(new UserOutputResolver()), or by calling factory.setAttribute("http://saxon.sf.net/feature/outputURIResolver", new UserOutputResolver()).


Michael H. Kay
22 June 2003