logolineright
bottomhttp://xml.apache.org/http://www.apache.org/http://www.w3.org/
join
Overview
separator
Compiler design
separator
Whitespace
xsl:sort
Keys
Comment design
separator
lang()
Unparsed entities
separator
If design
Choose|When|Otherwise design
Include|Import design
Variable|Param design
separator
Runtime
separator
Internal DOM
Namespaces
separator
Translet & TrAX
XPath Predicates
Xsltc Iterators
Xsltc Native API
Xsltc TrAX API
Performance Hints
separator
Credits
close
Contents
 

This document describes the design and overall architecture of XSLTC's runtime environment. This does not include the internal DOM and the DOM iterators, which are all covered in separate documents.


Runtime overview
 

This figure shows the main components of XSLTC's runtime environment:

runtime_design.gif

Figure 1: Runtime environment overview

The various steps these components have to go through to transform a document are:

  • instanciate a parser and hand it the input document
  • build an internal DOM from the parser's SAX events
  • instanciate the translet object
  • pass control to the translet object
  • receive output events from the translet
  • format the output document

This process can be initiated either through XSLTC's native API or through the implementation of the JAXP/TrAX API.


The compiled translet
 

A translet is always a subclass of AbstractTranslet. As well as having access to the public/protected methods in this class, the translet is compiled with these methods:

    public void transform(DOM, NodeIterator, TransletOutputHandler);

This method is passed a DOMImpl object. Depending on whether the stylesheet had any calls to the document() function this method will either generate a DOMAdapter object (when only one XML document is used as input) or a MultiDOM object (when there are more than one XML input documents). This DOM object is passed on to the topLevel() method.

When the topLevel() method returns, we initiate the output document by calling startDocument() on the supplied output handler object. We then call applyTemplates() to get the actual output contents, before we close the output document by calling endDocument() on the output handler.

    public void topLevel(DOM, NodeIterator, TransletOutputHandler);

This method handles all of these top-level elements:

  • <xsl:output>
  • <xsl:decimal-format>
  • <xsl:key>
  • <xsl:param> (for global parameters)
  • <xsl:variable> (for global variables)
    public void applyTemplates(DOM, NodeIterator, TransletOutputHandler);

This is the method that produces the actual output. Its central element is a big switch() statement that is used to trigger the code that represent the available templates for the various node in the input document. See the chapter on the main program loop for details on this method.

    public void <init>();

The translet's constructor initializes a table of all the elements we want to search for in the XML input document. This table is called the namesArray, and maps each element name to an unique integer value, know as the elements "translet-type". The DOMAdapter, which acts as a mediator between the DOM and the translet, will map these element identifier to the element identifiers used internally in the DOM. See the section on extern/internal type mapping and the internal DOM design document for details on this.

The constructor also initializes any DecimalFormatSymbol objects that are used to format numbers before passing them to the output post-processor. The output processor uses thes symbols to format decimal numbers in the output.

    public boolean stripSpace(int nodeType);

This method is only present if any <xsl:strip-space> or <xsl:preserve-space> elements are present in the stylesheet. If that is the case, the translet implements the StripWhitespaceFilter interface by containing this method.


External/internal type mapping
 

This is the very core of XSL transformations: Read carefully!!!

Every node in the input XML document(s) is assigned a type by the DOM builder class. This type is a unique integer value which represents the element, so that for instance all <bob> elements in the input document will be given type 7 and can be referred to by that integer. These types can be used for lookups in the namesArray table to get the actual element name (in this case "bob"). The type identifiers used in the DOM are referred to as external types or DOM types, as they are types known only outside of the translet.

Similarly the translet assignes types to all element and attribute names that are referenced in the stylesheet. This type assignment is done at compile-time, while the DOM builder assigns the external types at runtime. The element type identifiers used by the translet are referred to as internal types or translet types.

It is not very probable that there will be a one-to-one mapping between internal and external types. There will most often be elements in the DOM (ie. the input document) that are not mentioned in the stylesheet, and there could be elements in the stylesheet that do not match any elements in the DOM. Here is an example:

      <?xml version="1.0"?>
      <xsl:stylesheet version="1.0" xmlns:xsl="blahblahblah">

      <xsl:template match="/">
        <xsl:for-each select="//B">
          <xsl:apply-templates select="." />
        </xsl:for-each>
        <xsl:for-each select="C">
          <xsl:apply-templates select="." />
        </xsl:for-each>
        <xsl:for-each select="A/B">
          <xsl:apply-templates select="." />
        </xsl:for-each>
      </xsl:template>

    </xsl:stylesheet>

In this stylesheet we are looking for elements <B>, <C> and <A>. For this example we can assume that these element types will be assigned the values 0, 1 and 2. Now, lets say we are transforming this XML document:

      <?xml version="1.0"?>

      <A>
        The crocodile cried:
        <F>foo</F>
        <B>bar</B>
        <B>baz</B>
      </A>

This XML document has the elements <A>, <B> and <F>, which we assume are assigned the types 7, 8 and 9 respectively (the numbers below that are assigned for specific element types, such as the root node, text nodes,etc). This causes a mismatch between the type used for <B> in the translet and the type used for <B> in the DOM. The DOMAdapter class (which mediates between the DOM and the translet) has been given two tables for convertint between the two types; mapping for mapping from internal to external types, and reverseMapping for the other way around.

The translet contains a String[] array called namesArray. This array contains all the element and attribute names that were referenced in the stylesheet. In our example, this array would contain these string (in this specific order): "B", "C" and "A". This array is passed as one of the parameters to the DOM adapter constructor (the other parameter is the DOM itself). The DOM adapter passes this table on to the DOM. The DOM generates a hashtable that maps its known element names to the types the translet knows. The DOM does this by going through the namesArray from the translet sequentially, looks up each name in the hashtable, and is then able to map the internal type to an external type. The result is then passed back to the DOM adapter.

External types that are not interesting for the translet (such as the type for <F> elements in the example above) are mapped to a generic "ELEMENT" type (integer value 3), and are more or less ignored by the translet. Uninterresting attributes are similarly mapped to internal type "ATTRIBUTE" (integer value 4).

It is important that we separate the DOM from the translet. In several cases we want the DOM as a structure completely independent from the translet - even though the DOM is a structure internal to XSLTC. One such case is when transformations are offered by a servlet as a web service. Any DOM that is built should potentially be stored in a cache and made available for simultaneous access by several translet/servlet couples.

runtime_type_mapping.gif

Figure 2: Two translets accessing a single dom using different type mappings


Main program loop
 

The main body of the translet is the applyTemplates() method. This method goes through these steps:

  • Get the next node from the node iterator
  • Get the internal type of this node. The DOMAdapter object holds the internal/external type mapping table, and it will supply the translet with the internal type of the current node.
  • Execute a switch statement on the internal node type. There will be one "case" label for each recognised node type - this includes the first 7 internal node types.

The root node will have internal type 0 and will cause any initial literal elements to be output. Text nodes will have internal node type 1 and will simply be dumped to the output handler. Unrecognized elements will have internal node type 3 and will be given the default treatment (a new iterator is created for the node's children, and this iterator is passed with a recursive call to applyTemplates()). Unrecognised attribute nodes (type 4) will be handled like text nodes. This makes up the default (built in) templates of any stylesheet. Then, we add one "case"for each node type that is matched by any pattern in the stylesheet. The switch() statement in applyTemplates will thereby look something like this:

        public void applyTemplates(DOM dom, NodeIterator,
                                   TransletOutputHandler handler) {

            // get nodes from iterator
            while ((node = iterator.next()) != END) {
                // get internal node type
                switch(DOM.getType(node)) {

                case 0: // root
                    outputPreable(handler);
                    break;
                case 1: // text
                    DOM.characters(node,handler);
                    break;
                case 3: // unrecognised element
                    NodeIterator newIterator = DOM.getChildren(node);
                    applyTemplates(DOM,newIterator,handler);
                    break;
                case 4: // unrecognised attribute
                    DOM.characters(node,handler);
                    break;
                case 7: // elements of type <B>
                    someCompiledCode();
                    break;
                case 8: // elements of type <C>
                    otherCompiledCode();
                    break;
                default:
                    break;
                }
            }
        }

Each recognised element will have its own piece of compiled code.

Note that each "case" will not lead directly to a single template. There may be several templates that match node type 7 (say <B>). In the sample stylesheet in the previous chapter we have to templates that would match a node <B>. We have one match="//B" (match just any <B> element) and one match="A/B" (match a <B> element that is a child of a <A> element). In this case we would have to compile code that first gets the type of the current node's parent, and then compared this type with the type for <A>. If there was no match we will have executed the first <xsl:for-each> element, but if there was a match we will have executed the last one. Consequentally, the compiler will generate the following code (well, it will look like this anyway):

        switch(DOM.getType(node)) {
          :
          :
        case 7: // elements of type <B>
            int parent = DOM.getParent(node);
            if (DOM.getType(parent) == 9) // type 9 = elements <A>
                someCompiledCode();
            else
                evenOtherCompiledCode();
            break;
          :
          :
        }

We could do the same for namespaces, that is, assign a numeric value to every namespace that is references in the stylesheet, and use an "if" statement for each namespace that needs to be checked for each type. Lets say we had a stylesheet like this:

      <?xml version="1.0"?>
      <xsl:stylesheet version="1.0" xmlns:xsl="blahblahblah">

      <xsl:template match="/"
          xmlns:foo="http://foo.com/spec"
          xmlns:bar="http://bar.net/ref">
        <xsl:for-each select="foo:A">
          <xsl:apply-templates select="." />
        </xsl:for-each>
        <xsl:for-each select="bar:A">
          <xsl:apply-templates select="." />
        </xsl:for-each>
      </xsl:template>

    </xsl:stylesheet>

And a stylesheet like this:

      <?xml version="1.0"?>

      <DOC xmlns:foo="http://foo.com/spec"
           xmlns:bar="http://bar.net/ref">
        <foo:A>In foo namespace</foo:A>
        <bar:A>In bar namespace</bar:A>
      </DOC>

We could still keep the same type for all <A> elements regardless of what namespace they are in, and use the same "if" structure within the switch() statement above. The other option is to assign different types to <foo:A> and <bar:A> elements. The latter is the option we chose, and it is described in detail in the namespace design document.


Runtime library
 

The runtime library offers basic functionality to the translet at runtime. It is analoguous to UNIX's libc. The whole runtime library is contained in a single class file:

    org.apache.xalan.xsltc.runtime.BasisLibrary

This class contains a large set of static methods that are invoked by the translet. These methods are largely independent from eachother, and they implement the following:

  • simple XPath functions that do not require a lot of code compiled into the translet class
  • functions for formatting decimal numbers to strings
  • functions for comparing nodes, node-sets and strings - used by equality expressions, predicates and other
  • functions for generating localised error messages

The runtime library is a central part of XSLTC. But, as metioned earlier, the functions within the library are rarely related, so there is no real overall design/architecture. The only common attribute of many of the methods in the library is that all static methods that implement an XPath function and with a capital F.


Output handler
 

The translet passes its output to an output post-processor before the final result is handed to the client application over a standard SAX interface. The interface between the translet and the output handler is very similar to a SAX interface, but it has a few non-standard additions. This interface is described in this file:

    org.apache.xalan.xsltc.TransletOutputHandler

This interface is implemented by:

    org.apache.xalan.xsltc.runtime.TextOutput

This class, despite its name, handles all types of output (XML, HTML and TEXT). Our initial idea was to have a base class implementing the TransletOutputHandler interface, and then have one subclass for each of the output types. This proved very difficult, as the output type is not always known until after the transformation has started and some elements have been output. But, this is an area where a change like that has the potential to increase performance significantly. Output handling has a lot to do with analyzing string contents, and by narrowing down the number of string comparisons and string updates one can acomplish a lot.

The main tasks of the output handler are:

  • determine the output type based on the output generated by the translet (not always necessary)
  • generate SAX events for the client application
  • insert the necessary namespace declarations in the output
  • escape special characters in the output
  • insert <DOCTYPE> and <META> elements in HTML output

There is a very clear link between the output handler and the org.apache.xalan.xsltc.compiler.Output class that handles the <xsl:output> element. The Output class stores many output settings and parameters in the translet class file and the translet passes these on to the output handler.



dot
Copyright © 2004 The Apache Software Foundation. All Rights Reserved.