internals.html   [plain text]

<?xml version="1.0" encoding="ISO-8859-1"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "">
<html xmlns=""><head><meta http-equiv="Content-Type" content="text/html; charset=ISO-8859-1" /><style type="text/css">
TD {font-family: Verdana,Arial,Helvetica}
BODY {font-family: Verdana,Arial,Helvetica; margin-top: 2em; margin-left: 0em; margin-right: 0em}
H1 {font-family: Verdana,Arial,Helvetica}
H2 {font-family: Verdana,Arial,Helvetica}
H3 {font-family: Verdana,Arial,Helvetica}
A:link, A:visited, A:active { text-decoration: underline }
    </style><title>Library internals</title></head><body bgcolor="#8b7765" text="#000000" link="#a06060" vlink="#000000"><table border="0" width="100%" cellpadding="5" cellspacing="0" align="center"><tr><td width="120"><a href=""><img src="epatents.png" alt="Action against software patents" /></a></td><td width="180"><a href=""><img src="gnome2.png" alt="GNOME2 Logo" /></a><a href=""><img src="w3c.png" alt="W3C logo" /></a><a href=""><img src="redhat.gif" alt="Red Hat Logo" /></a><div align="left"><a href=""><img src="Libxslt-Logo-180x168.gif" alt="Made with Libxslt Logo" /></a></div></td><td><table border="0" width="90%" cellpadding="2" cellspacing="0" align="center" bgcolor="#000000"><tr><td><table width="100%" border="0" cellspacing="1" cellpadding="3" bgcolor="#fffacd"><tr><td align="center"><h1>The XSLT C library for GNOME</h1><h2>Library internals</h2></td></tr></table></td></tr></table></td></tr></table><table border="0" cellpadding="4" cellspacing="0" width="100%" align="center"><tr><td bgcolor="#8b7765"><table border="0" cellspacing="0" cellpadding="2" width="100%"><tr><td valign="top" width="200" bgcolor="#8b7765"><table border="0" cellspacing="0" cellpadding="1" width="100%" bgcolor="#000000"><tr><td><table width="100%" border="0" cellspacing="1" cellpadding="3"><tr><td colspan="1" bgcolor="#eecfa1" align="center"><center><b>Main Menu</b></center></td></tr><tr><td bgcolor="#fffacd"><form action="search.php" enctype="application/x-www-form-urlencoded" method="get"><input name="query" type="text" size="20" value="" /><input name="submit" type="submit" value="Search ..." /></form><ul><li><a href="index.html">Home</a></li><li><a href="intro.html">Introduction</a></li><li><a href="docs.html">Documentation</a></li><li><a href="bugs.html">Reporting bugs and getting help</a></li><li><a href="help.html">How to help</a></li><li><a href="downloads.html">Downloads</a></li><li><a href="FAQ.html">FAQ</a></li><li><a href="news.html">News</a></li><li><a href="xsltproc2.html">The xsltproc tool</a></li><li><a href="docbook.html">DocBook</a></li><li><a href="API.html">The programming API</a></li><li><a href="python.html">Python and bindings</a></li><li><a href="internals.html">Library internals</a></li><li><a href="extensions.html">Writing extensions</a></li><li><a href="contribs.html">Contributions</a></li><li><a href="EXSLT/index.html" style="font-weight:bold">libexslt</a></li><li><a href="xslt.html">flat page</a>, <a href="site.xsl">stylesheet</a></li><li><a href="html/index.html" style="font-weight:bold">API Menu</a></li><li><a href="ChangeLog.html">ChangeLog</a></li></ul></td></tr></table><table width="100%" border="0" cellspacing="1" cellpadding="3"><tr><td colspan="1" bgcolor="#eecfa1" align="center"><center><b>Related links</b></center></td></tr><tr><td bgcolor="#fffacd"><ul><li><a href="tutorial/libxslttutorial.html">Tutorial</a>,
          <a href="tutorial2/libxslt_pipes.html">Tutorial2</a></li><li><a href="xsltproc.html">Man page for xsltproc</a></li><li><a href="">Mail archive</a></li><li><a href="">XML libxml2</a></li><li><a href="">FTP</a></li><li><a href="">Windows binaries</a></li><li><a href="">Solaris binaries</a></li><li><a href="">MacOsX binaries</a></li><li><a href="">Bug Tracker</a></li><li><a href="">lxml Python bindings</a></li><li><a href="">Perl XSLT bindings</a></li><li><a href="">XSLT with PHP</a></li><li><a href="">Apache module</a></li><li><a href="">Pascal bindings</a></li><li><a href="">Xsldbg Debugger</a></li></ul></td></tr></table><table width="100%" border="0" cellspacing="1" cellpadding="3"><tr><td colspan="1" bgcolor="#eecfa1" align="center"><center><b>API Indexes</b></center></td></tr><tr><td bgcolor="#fffacd"><ul><li><a href="APIchunk0.html">Alphabetic</a></li><li><a href="APIconstructors.html">Constructors</a></li><li><a href="APIfunctions.html">Functions/Types</a></li><li><a href="APIfiles.html">Modules</a></li><li><a href="APIsymbols.html">Symbols</a></li></ul></td></tr></table></td></tr></table></td><td valign="top" bgcolor="#8b7765"><table border="0" cellspacing="0" cellpadding="1" width="100%"><tr><td><table border="0" cellspacing="0" cellpadding="1" width="100%" bgcolor="#000000"><tr><td><table border="0" cellpadding="3" cellspacing="1" width="100%"><tr><td bgcolor="#fffacd"><h3>Table  of contents</h3><ul><li><a href="internals.html#Introducti">Introduction</a></li>
  <li><a href="internals.html#Basics">Basics</a></li>
  <li><a href="internals.html#Keep">Keep it simple stupid</a></li>
  <li><a href="internals.html#libxml">The libxml nodes</a></li>
  <li><a href="internals.html#XSLT">The XSLT processing steps</a></li>
  <li><a href="internals.html#XSLT1">The XSLT stylesheet compilation</a></li>
  <li><a href="internals.html#XSLT2">The XSLT template compilation</a></li>
  <li><a href="internals.html#processing">The processing itself</a></li>
  <li><a href="internals.html#XPath">XPath expressions compilation</a></li>
  <li><a href="internals.html#XPath1">XPath interpretation</a></li>
  <li><a href="internals.html#Descriptio">Description of XPath
  <li><a href="internals.html#XPath3">XPath functions</a></li>
  <li><a href="internals.html#stack">The variables stack frame</a></li>
  <li><a href="internals.html#Extension">Extension support</a></li>
  <li><a href="internals.html#Futher">Further reading</a></li>
  <li><a href="internals.html#TODOs">TODOs</a></li>
  <li><a href="internals.html#Thanks">Thanks</a></li>
</ul><h3><a name="Introducti2" id="Introducti2">Introduction</a></h3><p>This document describes the processing of <a href="">libxslt</a>, the <a href="">XSLT</a> C library developed for the <a href="">GNOME</a> project.</p><p>Note: this documentation is by definition incomplete and I am not good at
spelling, grammar, so patches and suggestions are <a href="">really welcome</a>.</p><h3><a name="Basics1" id="Basics1">Basics</a></h3><p>XSLT is a transformation language. It takes an input document and a
stylesheet document and generates an output document:</p><p align="center"><img src="processing.gif" alt="the XSLT processing model" /></p><p>Libxslt is written in C. It relies on <a href="">libxml</a>, the XML C library for GNOME, for
the following operations:</p><ul><li>parsing files</li>
  <li>building the in-memory DOM structure associated with the documents
  <li>the XPath implementation</li>
  <li>serializing back the result document to XML and HTML. (Text is handled
</ul><h3><a name="Keep1" id="Keep1">Keep it simple stupid</a></h3><p>Libxslt is not very specialized. It is built under the assumption that all
nodes from the source and output document can fit in the virtual memory of
the system. There is a big trade-off there. It is fine for reasonably sized
documents but may not be suitable for large sets of data. The gain is that it
can be used in a relatively versatile way. The input or output may never be
serialized, but the size of documents it can handle are limited by the size
of the memory available.</p><p>More specialized memory handling approaches are possible, like building
the input tree from a serialization progressively as it is consumed,
factoring repetitive patterns, or even on-the-fly generation of the output as
the input is parsed but it is possible only for a limited subset of the
stylesheets. In general the implementation of libxslt follows the following
pattern:</p><ul><li>KISS (keep it simple stupid)</li>
  <li>when there is a clear bottleneck optimize on top of this simple
    framework and refine only as much as is needed to reach the expected
</ul><p>The result is not that bad, clearly one can do a better job but more
specialized too. Most optimization like building the tree on-demand would
need serious changes to the libxml XPath framework. An easy step would be to
serialize the output directly (or call a set of SAX-like output handler to
keep this a flexible interface) and hence avoid the memory consumption of the
result.</p><h3><a name="libxml" id="libxml">The libxml nodes</a></h3><p>DOM-like trees, as used and generated by libxml and libxslt, are
relatively complex. Most node types follow the given structure except a few
variations depending on the node type:</p><p align="center"><img src="node.gif" alt="description of a libxml node" /></p><p>Nodes carry a <strong>name</strong> and the node <strong>type</strong>
indicates the kind of node it represents, the most common ones are:</p><ul><li>document nodes</li>
  <li>element nodes</li>
  <li>text nodes</li>
</ul><p>For the XSLT processing, entity nodes should not be generated (i.e. they
should be replaced by their content). Most nodes also contains the following
"navigation" information:</p><ul><li>the containing <strong>doc</strong>ument</li>
  <li>the <strong>parent</strong> node</li>
  <li>the first <strong>children</strong> node</li>
  <li>the <strong>last</strong> children node</li>
  <li>the <strong>prev</strong>ious sibling</li>
  <li>the following sibling (<strong>next</strong>)</li>
</ul><p>Elements nodes carries the list of attributes in the properties, an
attribute itself holds the navigation pointers and the children list (the
attribute value is not represented as a simple string to allow usage of
entities references).</p><p>The <strong>ns</strong> points to the namespace declaration for the
namespace associated to the node, <strong>nsDef</strong> is the linked list
of namespace declaration present on element nodes.</p><p>Most nodes also carry an <strong>_private</strong> pointer which can be
used by the application to hold specific data on this node.</p><h3><a name="XSLT" id="XSLT">The XSLT processing steps</a></h3><p>There are a few steps which are clearly decoupled at the interface
level:</p><ol><li>parse the stylesheet and generate a DOM tree</li>
  <li>take the stylesheet tree and build a compiled version of it (the
    compilation phase)</li>
  <li>take the input and generate a DOM tree</li>
  <li>process the stylesheet against the input tree and generate an output
  <li>serialize the output tree</li>
</ol><p>A few things should be noted here:</p><ul><li>the steps 1/ 3/ and 5/ are optional:  the DOM representing the
    stylesheet and input can be created by other means, not just by parsing
    serialized XML documents, and similarly the result tree DOM can be
    made available to other processeswithout being serialized.
  </li><li>the stylesheet obtained at 2/ can be reused by multiple processing 4/
    (and this should also work in threaded programs)</li>
  <li>the tree provided in 2/ should never be freed using xmlFreeDoc, but by
    freeing the stylesheet.</li>
  <li>the input tree created in step 3/ is not modified except the
    _private field which may be used for labelling keys if used by the
    stylesheet. It's not modified at all in step 4/ to allow parallel
    processing using a shared precompiled stylesheet.</li>
</ul><h3><a name="XSLT1" id="XSLT1">The XSLT stylesheet compilation</a></h3><p>This is the second step described. It takes a stylesheet tree, and
"compiles" it. This associates to each node a structure stored in the
_private field and containing information computed in the stylesheet:</p><p align="center"><img src="stylesheet.gif" alt="a compiled XSLT stylesheet" /></p><p>One xsltStylesheet structure is generated per document parsed for the
stylesheet. XSLT documents allow includes and imports of other documents,
imports are stored in the <strong>imports</strong> list (hence keeping the
tree hierarchy of includes which is very important for a proper XSLT
processing model) and includes are stored in the <strong>doclist</strong>
list. An imported stylesheet has a parent link to allow browsing of the
tree.</p><p>The DOM tree associated to the document is stored in <strong>doc</strong>.
It is preprocessed to remove ignorable empty nodes and all the nodes in the
XSLT namespace are subject to precomputing. This usually consist of
extracting all the context information from the context tree (attributes,
namespaces, XPath expressions), and storing them in an xsltStylePreComp
structure associated to the <strong>_private</strong> field of the node.</p><p>A couple of notable exceptions to this are XSLT template nodes (more on
this later) and attribute value templates. If they are actually templates,
the value cannot be computed at compilation time. (Some preprocessing could
be done like isolation and preparsing of the XPath subexpressions but it's
not done, yet.)</p><p>The xsltStylePreComp structure also allows storing of the precompiled form
of an XPath expression that can be associated to an XSLT element (more on
this later).</p><h3><a name="XSLT2" id="XSLT2">The XSLT template compilation</a></h3><p>A proper handling of templates lookup is one of the keys of fast XSLT
processing. (Given a node in the source document this is the process of
finding which templates should be applied to this node.) Libxslt follows the
hint suggested in the <a href="">5.2
Patterns</a> section of the XSLT Recommendation, i.e. it doesn't evaluate it
as an XPath expression but tokenizes it and compiles it as a set of rules to
be evaluated on a candidate node. There usually is an indication of the node
name in the last step of this evaluation and this is used as a key check for
the match. As a result libxslt builds a relatively more complex set of
structures for the templates:</p><p align="center"><img src="templates.gif" alt="The templates related structure" /></p><p>Let's describe a bit more closely what is built. First the xsltStylesheet
structure holds a pointer to the template hash table. All the XSLT patterns
compiled in this stylesheet are indexed by the value of the the target
element (or attribute, pi ...) name, so when a element or an attribute "foo"
needs to be processed the lookup is done using the name as a key.</p><p>Each of the patterns is compiled into an xsltCompMatch
(i.e. an ''XSLT compiled match') structure. It holds
the set of rules based on the tokenization of the pattern stored in reverse
order (matching is easier this way). </p><p>The xsltCompMatch are then stored in the hash table, the clash list is
itself sorted by priority of the template to implement "naturally" the XSLT
priority rules.</p><p>Associated to the compiled pattern is the xsltTemplate itself containing
the information required for the processing of the pattern including, of
course, a pointer to the list of elements used for building the pattern
result.</p><p>Last but not least a number of patterns do not fit in the hash table
because they are not associated to a name, this is the case for patterns
applying to the root, any element, any attributes, text nodes, pi nodes, keys
etc. Those are stored independently in the stylesheet structure as separate
linked lists of xsltCompMatch.</p><h3><a name="processing" id="processing">The processing itself</a></h3><p>The processing is defined by the XSLT specification (the basis of the
algorithm is explained in <a href="">the Introduction</a>
section). Basically it works by taking the root of the input document
as the cureent node and applying the following algorithm:</p><ol><li>Finding the template applying to current node.
    This is a lookup in the template hash table, walking the hash list until
    the node satisfies all the steps of the pattern, then checking the
    appropriate global template(s) (i.e.  templates applying to a node type)
    to see if there isn't a higher priority rule to apply</li>
  <li>If there is no template, apply the default rule (recurse on the
    children as the current node)</li>
  <li>else walk the content list of the selected templates, for each of them:
    <ul><li>if the node is in the XSLT namespace then the node has a _private
        field pointing to the preprocessed values, jump to the specific
      <li>if the node is in an extension namespace, look up the associated
      <li>otherwise copy the node.</li>
    </ul><p>The closure is usually done through the XSLT
    <strong>apply-templates</strong>construct, which invokes this process
      recursively starting at step 1, to find the appropriate template
      for the nodes selected by the 'select' attribute of the apply-templates
      instruction (default: the children of the node currently being
</ol><p>Note that large parts of the input tree may not be processed by a given
stylesheet and that conversely some may be processed multiple times.
(This often is the case when a Table of Contents is built).</p><p>The module <code>transform.c</code> is the one implementing most of this
logic. <strong>xsltApplyStylesheet()</strong> is the entry point, it
allocates an xsltTransformContext containing the following:</p><ul><li>a pointer to the stylesheet being processed</li>
  <li>a stack of templates</li>
  <li>a stack of variables and parameters</li>
  <li>an XPath context</li>
  <li>the template mode</li>
  <li>current document</li>
  <li>current input node</li>
  <li>current selected node list</li>
  <li>the current insertion points in the output document</li>
  <li>a couple of hash tables for extension elements and functions</li>
</ul><p>Then a new document gets allocated (HTML or XML depending on the type of
output), the user parameters and global variables and parameters are
evaluated. Then <strong>xsltProcessOneNode()</strong> which implements the
1-2-3 algorithm is called on the docuemnt node of the input. Step 1/ is
implemented by calling <strong>xsltGetTemplate()</strong>, step 2/ is
implemented by <strong>xsltDefaultProcessOneNode()</strong> and step 3/ is
implemented by <strong>xsltApplyOneTemplate()</strong>.</p><h3><a name="XPath" id="XPath">XPath expression compilation</a></h3><p>The XPath support is actually implemented in the libxml module (where it
is reused by the XPointer implementation). XPath is a relatively classic
expression language. The only uncommon feature is that it is working on XML
trees and hence has specific syntax and types to handle them.</p><p>XPath expressions are compiled using <strong>xmlXPathCompile()</strong>.
It will take an expression string in input and generate a structure
containing the parsed expression tree, for example the expression:</p><pre>/doc/chapter[title='Introduction']</pre><p>will be compiled as</p><pre>Compiled Expression : 10 elements
    COLLECT  'child' 'name' 'node' chapter
      COLLECT  'child' 'name' 'node' doc
          EQUAL =
            COLLECT  'child' 'name' 'node' title
            ELEM Object is a string : Introduction
              COLLECT  'child' 'name' 'node' title
                NODE</pre><p>This can be tested using the  <code>testXPath</code>  command (in the
libxml codebase) using the <code>--tree</code> option.</p><p>Again, the KISS approach is used. No optimization is done. This could be
an interesting thing to add. <a href=";l=132%2ct=gr%2c+p=saxon">Michael
Kay describes</a> a lot of possible and interesting optimizations done in
Saxon which would be possible at this level. I'm unsure they would provide
much gain since the expressions tends to be relatively simple in general and
stylesheets are still hand generated. Optimizations at the interpretation
sounds likely to be more efficient.</p><h3><a name="XPath1" id="XPath1">XPath interpretation</a></h3><p>The interpreter is implemented by <strong>xmlXPathCompiledEval()</strong>
which is the front-end to <strong>xmlXPathCompOpEval()</strong> the function
implementing the evaluation of the expression tree. This evaluation follows
the KISS approach again. It's recursive and calls
<strong>xmlXPathNodeCollectAndTest()</strong> to collect a set of nodes when
evaluating a <code>COLLECT</code> node.</p><p>An evaluation is done within the framework of an XPath context stored in
an <strong>xmlXPathContext</strong> structure, in the framework of a
transformation the context is maintained within the XSLT context. Its content
follows the requirements from the XPath specification:</p><ul><li>the current document</li>
  <li>the current node</li>
  <li>a hash table of defined variables (but not used by XSLT,
      which uses its own stack frame for variables, described below)</li>
  <li>a hash table of defined functions</li>
  <li>the proximity position (the place of the node in the current node
  <li>the context size (the size of the current node list)</li>
  <li>the array of namespace declarations in scope (there also is a namespace
    hash table but it is not used in the XSLT transformation).</li>
</ul><p>For the purpose of XSLT an <strong>extra</strong> pointer has been added
allowing to retrieve the XSLT transformation context. When an XPath
evaluation is about to be performed, an XPath parser context is allocated
containing an XPath object stack (this is actually an XPath evaluation
context, this is a relic of the time where there was no separate parsing and
evaluation phase in the XPath implementation). Here is an overview of the set
of contexts associated to an XPath evaluation within an XSLT
transformation:</p><p align="center"><img src="contexts.gif" alt="The set of contexts associated " /></p><p>Clearly this is a bit too complex and confusing and should be refactored
at the next set of binary incompatible releases of libxml. For example the
xmlXPathCtxt has a lot of unused parts and should probably be merged with
xmlXPathParserCtxt.</p><h3><a name="Descriptio" id="Descriptio">Description of XPath Objects</a></h3><p>An XPath expression manipulates XPath objects. XPath defines the default
types boolean, numbers, strings and node sets. XSLT adds the result tree
fragment type which is basically an unmodifiable node set.</p><p>Implementation-wise, libxml follows again a KISS approach, the
xmlXPathObject is a structure containing a type description and the various
possibilities. (Using an enum could have gained some bytes.) In the case of
node sets (or result tree fragments), it points to a separate xmlNodeSet
object which contains the list of pointers to the document nodes:</p><p align="center"><img src="object.gif" alt="An Node set object pointing to " /></p><p>The <a href="">XPath API</a> (and
its <a href="">'internal'
part</a>) includes a number of functions to create, copy, compare, convert or
free XPath objects.</p><h3><a name="XPath3" id="XPath3">XPath functions</a></h3><p>All the XPath functions available to the interpreter are registered in the
function hash table linked from the XPath context. They all share the same
signature:</p><pre>void xmlXPathFunc (xmlXPathParserContextPtr ctxt, int nargs);</pre><p>The first argument is the XPath interpretation context, holding the
interpretation stack. The second argument defines the number of objects
passed on the stack for the function to consume (last argument is on top of
the stack).</p><p>Basically an XPath function does the following:</p><ul><li>check <code>nargs</code> for proper handling of errors or functions
    with variable numbers of parameters</li>
  <li>pop the parameters from the stack using <code>obj =
  <li>do the function specific computation</li>
  <li>push the result parameter on the stack using <code>valuePush(ctxt,
  <li>free up the input parameters with
</ul><p>Sometime the work can be done directly by modifying in-situ the top object
on the stack <code>ctxt-&gt;value</code>.</p><h3><a name="stack" id="stack">The XSLT variables stack frame</a></h3><p>Not to be confused with XPath object stack, this stack holds the XSLT
variables and parameters as they are defined through the recursive calls of
call-template, apply-templates and default templates. This is used to define
the scope of variables being called.</p><p>This part seems to be one needing most work , first it is
done in a very inefficient way since the location of the variables and
parameters within the stylesheet tree is still done at run time (it really
should be done statically at compile time), and I am still unsure that my
understanding of the template variables and parameter scope is actually
right.</p><p>This part of the documentation is still to be written once this part of
the code will be stable. <span style="background-color: #FF0000">TODO</span></p><h3><a name="Extension" id="Extension">Extension support</a></h3><p>There is a separate document explaining <a href="extensions.html">how the
extension support works</a>.</p><h3><a name="Futher" id="Futher">Further reading</a></h3><p>Michael Kay wrote <a href=";l=132%2ct=gr%2c+p=saxon">a
really interesting article on Saxon internals</a> and the work he did on
performance issues. I wish I had read it before starting libxslt design (I
would probably have avoided a few mistakes and progressed faster). A lot of
the ideas in his papers should be implemented or at least tried in
libxslt.</p><p>The <a href="">libxml documentation</a>, especially <a href="">the I/O interfaces</a> and the <a href="">memory management</a>.</p><h3><a name="TODOs" id="TODOs">TODOs</a></h3><p>redesign the XSLT stack frame handling. Far too much work is done at
execution time. Similarly for the attribute value templates handling, at
least the embedded subexpressions ought to be precompiled.</p><p>Allow output to be saved to a SAX like output (this notion of SAX like API
for output should be added directly to libxml).</p><p>Implement and test some of the optimization explained by Michael Kay
especially:</p><ul><li>static slot allocation on the stack frame</li>
  <li>specific boolean interpretation of an XPath expression</li>
  <li>some of the sorting optimization</li>
  <li>Lazy evaluation of location path. (this may require more changes but
    sounds really interesting. XT does this too.)</li>
  <li>Optimization of an expression tree (This could be done as a completely
    independent module.)</li>
</ul><p></p><p>Error reporting, there is a lot of case where the XSLT specification
specify that a given construct is an error are not checked adequately by
libxslt. Basically one should do a complete pass on the XSLT spec again and
add all tests to the stylesheet compilation. Using the DTD provided in the
appendix and making direct checks using the libxml validation API sounds a
good idea too (though one should take care of not raising errors for
elements/attributes in different namespaces).</p><p>Double check all the places where the stylesheet compiled form might be
modified at run time (extra removal of blanks nodes, hint on the
xsltCompMatch).</p><h3><a name="Thanks" id="Thanks">Thanks:</a></h3><p>Thanks to <a href="">Michael Sperberg-McQueen</a> for
   various fixes and clarifications on this document!</p><p></p><p><a href="bugs.html">Daniel Veillard</a></p></td></tr></table></td></tr></table></td></tr></table></td></tr></table></td></tr></table></body></html>