XML.com 
 Published on XML.com http://www.xml.com/pub/a/2000/08/holman/s1.html
See this if you're having trouble printing code examples

 

What is XSLT? (I)
By G. Ken Holman
August 16, 2000

The Context of XSL Transformations and the XML Path Language

Table of Contents

1. The context of XSL Transformations and the XML Path Language
 •1.1 The XML family of
   Recommendations
      ·1.1.1 Extensible Markup
       Language (XML)
      ·1.1.2 XML Path Language
       (XPath)
      ·1.1.3 Styling structured
       information
      ·1.1.4 Extensible Stylesheet
       Language (XSL)
      ·1.1.5 Extensible Stylesheet
        Language Transformations
       (XSLT)
      ·1.1.6 Namespaces
      ·1.1.7 Stylesheet association
   •1.2 Transformation data
    flows
      ·1.2.1 Transformation from
       XML to XML
      ·1.2.2 Transformation from
       XML to XSL formatting
       semantics
      ·1.2.3 Transformation from
       XML to non-XML
      ·1.2.4 Three-tiered
       architectures

This first chapter examines the context of two W3C Recommendations -- Extensible Stylesheet Language Transformations (XSLT) and XML Path Language (XPath) -- within the growing family of Recommendations related to the Extensible Markup Language (XML). Later we will look at detailed examples, but first let's focus on XSLT and XPath in the context of a few of the Recommendations in the XML family and examine how these two Recommendations work together to address separate and distinct functionality required when working with structured information technologies.

This chapter does not attempt to address all of the numerous XML-related Recommendations currently released or in development. Specifically, we will be looking at only the following as they relate to XSLT and XPath:

Extensible Markup Language (XML)

For years, applications and vendors have imposed their constraints on the way we can represent our information. Our data has been created, maintained, stored and archived according to the rules enforced by others. The advent of the Extensible Markup Language (XML) moves the control of our information out of the hands of others and into our own by providing two basic facilities.

XML describes rules for structuring our information using embedded markup of our own choice. We can take control of our information representation by creating and using a vocabulary we design of elements and attributes that makes sense for the way we do our business and use our data.

In addition, XML describes a language for formally declaring the vocabularies we use. This allows our tools to constrain the creation of an instance of our information, and allows our users to validate a properly created instance of information against our set of constraints.

Note 1:

An XML document is just an instance of well-formed XML. The two terms document and instance could be used interchangeably, but this reference material uses the term instance to help readers remember that XML isn't just for documents or documentation. With XML we describe a related set of information in a tree-like hierarchical fashion, and gain the benefits of having done so, whether the information captures an invoice-related transaction between computers, or the content of a user manual rendered on paper.

XML Path Language (XPath)

XPath is a string syntax for building addresses to the information found in an XML document. We use this language to specify the locations of document structures or data found in an XML document when processing that information using XSLT. XPath allows us from any location to address any other location or content.

Extensible Stylesheet Language Family (XSLT/XSL)

Two vocabularies specified in separate W3C Recommendations provide for the two distinct styling processes of transforming and rendering XML instances.

We can transform information using one vocabulary into an alternate form by using the Extensible Stylesheet Language Transformations (XSLT).

The Extensible Stylesheet Language (XSL) is a rendering vocabulary describing the semantics of formatting information for different media.

Namespaces

We use XML namespaces to distinguish information when mixing multiple vocabularies in a single instance. Without namespaces our processes would find the information ambiguous when identical names have been chosen by the designers of the vocabularies we use.

Stylesheet Association

We declare our choice of an associated stylesheet for an XML instance by embedding the construct described in the Stylesheet Association Recommendation. Recipients and applications can choose to respect or ignore this choice, but the declaration indicates that we have tied some process (typically rendering) to our data, which specifies how to consume or work with our information.

1.1  The XML family of Recommendations

Now let's look at the objectives of these selected Recommendations.

1.1.1  Extensible Markup Language (XML)

Historically, the ways we have expressed, created, stored and transmitted our electronic information have been constrained and controlled by the vendors we choose and the applications we run. Alternatively, we now can express our data in a structured fashion oriented around our perspective of the nature of the information itself rather than the nature of an application's choice of how to represent our information. With Extensible Markup Language (XML), we describe our information using embedded markup of elements, attributes and other constructs in a tree-like structure.

1.1.1.1  Structuring information

Contrasted to a file format where information identification relies on some proprietary hidden format, predetermined ordering, or some kind of explicit labeling, the tree-like hierarchical storage structure infers relationships by the scope of values encompassing the scopes of other values.

Though trees shape a number of areas of XML, both logically (markup) and physically (entities such as files or other resources), they are not the only means by which relationships are specified. For example, a quantum of information can arbitrarily point or refer to other information elsewhere through use of unique identifiers.

Two basic objectives of representing information hierarchically are satisfied by the XML Recommendation. It provides:

There is an implicit document model for an instance of well-formed XML defined by the mere presence of nested elements found in the information. There is no need to declare this model because the syntax rules governing well-formedness guarantee the information to be seen properly as a hierarchy. As with all hierarchies, there are family-tree-like relationships of parent, child, and sibling constructs relative to each construct found.

Consider the following well-formed XML instance purc.xml:

01  <?xml version="1.0"?>
02  <purchase id="p001">
03    <customer db="cust123"/>
04    <product db="prod345">
05      <amount>23.45</amount>
06    </product>
07  </purchase> 
Example 1-1: A well-formed XML purchase order instance.

Observe the content nesting (whitespace has been added only for illustrative purposes). The instance follows the lexical rules for XML markup and the hierarchical model is implicit by the nesting of elements. Pay particular attention to the markup on line 3 for the empty element named customer, with the attribute named db. It will be used later in examples throughout this chapter. The customer element is a child of the document element, which is named purchase.

Although the presence of an explicit formal document model is useful to an XML processor or to a system working with XML instances, that model has no impact on the implicit structural model and only minor influence on the interpretation of content found in the instance. This point holds true whether the model is expressed in a DTD or in some of the other Recommendations for structural and content schemata being developed.

Consider the following valid XML instance purcdtd.xml:

01  <?xml version="1.0"?>
02  <!DOCTYPE purchase [
03  <!ELEMENT purchase ( customer, product+ )>
04  <!ATTLIST purchase id ID #REQUIRED>
05  <!ELEMENT customer EMPTY>
06  <!ATTLIST customer db CDATA #REQUIRED>
07  <!ELEMENT product  ( amount )>
08  <!ATTLIST product  db CDATA #REQUIRED>
09  <!ELEMENT amount   ( #PCDATA )>
10  ]>
11  <purchase id="p001">
12    <customer db="cust123"/>
13    <product db="prod345">
14      <amount>23.45</amount>
15    </product>
16  </purchase>
Example 1-2: A valid XML purchase order instance

See how the information content is no different from the previous example, but in this case an explicit document model using XML 1.0 DTD syntax is included (it could have been included by reference to a separate resource). A processor can validate that the information content conforms not only to the lexical rules for XML (well-formedness) but also the syntax rules dictated by the supplied document model (validity).

Looking at the same customer element as before (now on line 12), the document model indicates on line 6 that the db attribute is, indeed, required: if the attribute is absent the XML processor can report syntactic model constraint violation even if the element is otherwise lexically well-formed. The document model can also provide additional information not evident without a document model (such as the information on line 4 that the id attribute for purchase is of XML type ID).

1.1.1.2  No built-in meanings or concepts

The area of semantics associated with XML instances is very gray. A document model is but one component used to help describe the semantics of the information found in an instance. While well-formed instances do not have a formal document model, often the names of the constructs used within the instances give hints to the associated semantics. Without a formalism yet available in our community to express semantics in a rigorous fashion, we users of XML do (or should!) capture the semantics of a given vocabulary in prose, whether or not the document model is formalized.

The XML 1.0 Recommendation only describes the behavior required of an XML processor acting on an XML stream, and how it must identify constituent data and provide that data to an application using the processor:

Since there are no formalized semantic description facilities in XML, any XML that is used is not tied to any one particular concept or application. There are no rendition or transformation rules or constructs defined in XML. The only purpose of XML is to unambiguously identify and deliver constituent components of data. There are no inherent meanings or semantics of any kind associated with element types defined in a document model. There are no defined controls for implying any rendering semantics.

Even the xml:space attribute allowing for the differentiation of whitespace found in a document is not an aspect of rendering but of information description. The author or modeler of an instance is indicating with this reserved attribute (termed "special" in XML 1.0) the nature of the information and how the whitespace found in the information is to be either preserved or handled by a processor in a default fashion.

Some new users of XML who have a background in a markup language such as HTML often assume a magical association of semantics with element types of the same names they have been exposed to in their prior work. In a web page, they can safely assume that the construct <p> will be interpreted as a paragraph or <em> as emphasized text. However, this interpretation is solely the purview of the designers of HTML and user agents attempting to conform to the World Wide Web Consortium (W3C)-published semantics. Nothing is imposed by any process when creating a new XML vocabulary that happens to use the same names. Applications using XML processors to access XML information must be instructed how to interpret and implement the desired semantics.

1.1.2  XML Path Language (XPath)

Assuming that we have structured our information using XML, how are we going to talk about (address) what is inside our documents? Locating information in an XML document is critical to both transforming it and to associating or relating it to other information. When we write stylesheets and use linking languages, we can address components of our information for a processor by our use of the XML Path Language, also called XPath:

1.1.2.1  Addressing structured information

The W3C working group responsible for stylesheets collaborated with the W3C working group responsible for the next generation of hyperlinking to produce XPath as a common base for addressing requirements shared by their respective Recommendations. Both groups extend the core XPath facilities to meet the needs they have in each of their domains: the stylesheet group uses XPath as the core of expressions in XSLT; the linking group uses XPath as the core of expressions in the XPointer Recommendation.

In order to address components you have to know the addressing scheme with which the components are arranged. The basis of addressing XML documents is an abstract data model of interlinked nodes arranged hierarchically echoing the tree-shape of the nested elements in an instance. Nodes of different types make up this hierarchy, each node representing the parsed result of a syntactic structure found in the bytes of the XML instance.

This abstraction insulates addressing from the multiple syntactic forms of given XML constructs, allowing us to focus on the information itself and not the syntax used to represent the information.

Note 2:

We see XML documents as a stream or string of bytes that follow the rules of the XML 1.0 Recommendation. Stylesheets do not regard instances in this fashion, and we have to change the way we think of our XML documents in order to successfully work with our information. This leap of understanding ranks high on the list of key aspects of stylesheet writing I needed to internalize before successfully using this technology.

We are given tools to work in the framework provided by the abstraction: a set of data types used to represent values found in the generalization, and a set of functions we use to manipulate and examine those values. The data types include strings, numbers, boolean values and sets of nodes of our information. The functions allow us to cast these values into other data type representations and to return massaged information according to our needs.

1.1.2.2  Addressing identifies a hierarchical position or positions

XPath defines common semantics and syntax for addressing XML-expressed information, and bases these primarily on the hierarchical position of components in the tree. This ordering is referred to as document order in XPath, while in other contexts this is often termed either parse order or depth-first order. Alternatively, we can access an arbitrary location in the tree based on points in the tree having unique identifiers.

We convey XPath addresses in a simple and compact non-XML syntax. This allows us to use an XPath expression as the value of an attribute in an XML vocabulary as in the following examples:

01  select="answer"
Example 1-3: A simple XPath expression in a select attribute

The above attribute value expresses all children named "answer" of the current focus element.

01  match="question|answer"
Example 1-4: An XPath expression in a match attribute

The above attribute value expresses a test of an element being in the union of the element types named "question" and "answer".

The XPath syntax looks a lot like addressing subdirectories in a file system or as part of a Universal Resource Identifier (URI). Multiple steps in a location path are separated by either one or two oblique "/" characters. Filters can be specified to further refine the nature of the components of our information being addressed.

01  select="question[3]/answer[1]"
Example 1-5: A multiple step XPath expression in a select attribute

The above example selects only the first "answer" child of the third "question" child of the focus element.

01  select="id('start')//question[@answer='y']"
Example 1-6: A more complex XPath expression in a select attribute

The above example uses an XPath address identifying some descendants of the element in the instance that has the unique identifier with the value "start". Those identified are the question elements whose answer attribute is equal to the string equal to the lower-case letter 'y'. The value returned is the set of nodes representing the elements meeting the conditions expressed by the address. The address is used in a select attribute, thus the XSLT processor is selecting all of the addressed elements for some kind of processing.

1.1.2.3  XPath is not a query language

It is important to remember that addressing information is only one aspect of querying information. Other aspects include query operators that massage intermediate results into a final result. While a few operators and functions are available in XSLT to use values identified in documents, these are oriented to string processing, not to complex operations required by some applications.

Note 3:

When query Recommendations are developed, I would hope that the addressing portion is based on XPath as a core, just as with XSLT.

This is a prose version of an excerpt from the book "Practical Transformation Using XSLT and XPath" (Eighth Edition ISBN 1-894049-05-5 at the time of this writing) published by Crane Softwrights Ltd., written by G. Ken Holman; this excerpt was edited by Stan Swaren, and reviewed by Dave Pawson.

The Context of XSL Transformations and the XML Path Language (cont'd)

Table of Contents

1. The context of XSL Transformations and the XML Path Language
 •1.1 The XML family of
   Recommendations
      ·1.1.1 Extensible Markup
       Language (XML)
      ·1.1.2 XML Path Language
       (XPath)
      ·1.1.3 Styling structured
       information
      ·1.1.4 Extensible Stylesheet
       Language (XSL)
      ·1.1.5 Extensible Stylesheet
        Language Transformations
       (XSLT)
      ·1.1.6 Namespaces
      ·1.1.7 Stylesheet association
   •1.2 Transformation data
    flows
      ·1.2.1 Transformation from
       XML to XML
      ·1.2.2 Transformation from
       XML to XSL formatting
       semantics
      ·1.2.3 Transformation from
       XML to non-XML
      ·1.2.4 Three-tiered
       architectures

1.1.3  Styling structured information

1.1.3.1  Styling is transforming and formatting information

Styling is the rendering of information into a form suitable for consumption by a target audience. Because the audience can change for a given set of information, we often need to apply different styling for that information in order to obtain dissimilar renderings in order to meet the needs of each audience. Perhaps some information needs to be rearranged to make more sense for the reader. Perhaps some information needs to be highlighted differently to bring focus to key content.

It is important when we think about styling information to remember that two distinct processes are involved, not just one. First, we must transform the information from the organization used when it was created into the organization needed for consumption. Second, when rendering we must express, whatever the target medium, the aspects of the appearance of the reorganized information.

Consider the flow of information as a streaming process where information is created upstream and processed or consumed downstream. Upstream, in the early stages, we should be expressing the information abstractly, thus preventing any early binding of concrete or final-form concepts. Midstream, or even downstream, we can exploit the information as long as it remains flexible and abstract. Late binding of the information to a final form can be based on the target use of the final product; by delaying this binding until late in the process, we preserve the original information for exploitation for other purposes along the way.

It is a common but misdirected practice to model information based on how you plan to use it downstream. It does not matter if your target is a presentation-oriented structure, for example, or a structure that is appropriate for another markup-based system. Modeling practice should focus on both the business reasons and inherent relationships existing in the semantics behind the information being described (as such the vocabularies are then content-oriented). For example, emphasized text is often confused with a particular format in which it is rendered. Where we could model information using a <b> element type for eventual rendering in a bold face, we would be better off modeling the information using an <emph> element type. In this way we capture the reason for marking up information (that it is emphasized from surrounding information), and we do not lock the downstream targets into only using a bold face for rendering.

Many times the midstream or downstream processes need only rearrange, re-label or synthesize the information for a target purpose and never apply any semantics of style for rendering purposes. Transformation tasks stand alone in such cases, meeting the processing needs without introducing rendering issues.

One caveat regarding modeling content-oriented information is that there are applications where the content-orientation is, indeed, presentation-oriented. Consider book publishing where the abstract content is based on presentational semantics. This is meaningful because there is no abstraction beyond the appearance or presentation of the content.

Consider the customer information in Example 1-1. A web user agent doesn't know how to render an element named <customer>. The HTML vocabulary used to render the customer information could be as follows:

01  <p>From: <i>(Customer Reference) <b>cust123</b></i>
02  </p>
Example 1-7: HTML rendering semantics markup for example

The rendering result would then be as follows, with the rendering user agent interpreting the markup for italics and boldface presentation semantics:

Figure 1-1: 
HTML rendering for example
Figure 1-1: HTML rendering for example

The above illustrates these two distinct styling steps: transforming the instance of the XML vocabulary into a new instance according to a vocabulary of rendering semantics; and formatting the instance of the rendering vocabulary in the user agent.

1.1.3.2  Two W3C Recommendations

In order to meet these two distinct processes in a detached (yet related) fashion, the W3C Working Group responsible for the Extensible Stylesheet Language (XSL) split the original drafts of their work into two separate Recommendations: one for transforming information and the other for rendering information.

The XSL Transformations (XSLT) 1.0 Recommendation describes a vocabulary recognized by an XSLT processor to transform information from an organization in the source file into a different organization suitable for continued downstream processing.

The Extensible Stylesheet Language (XSL) Working Draft describes a vocabulary recognized by a rendering agent to reify abstract expressions of format into a particular medium of presentation.

Both XSLT and XSL are endorsed by members of WSSSL, an association of researchers and developers passionate about the application of markup technologies in today's information technology infrastructure.

1.1.4  Extensible Stylesheet Language (XSL)

When we need to present our structured information in a given medium or different media, we all have common needs for how the result appears and way the result flows through that appearance. The XSL Working Draft describes the current work developing a vocabulary of formatting and flow semantics that can be expressed using an XML model of elements and attributes:

1.1.4.1  Formatting and flow semantics vocabulary

This hierarchical vocabulary captures formatting semantics for rendering textual and graphic information in different media. A rendering agent is responsible for interpreting an instance of the vocabulary for a given medium to reify a final result.

This is no different in concept and architecture than using HTML and Cascading Stylesheets (CSS) as a hierarchical vocabulary for rendering a set of information in a web browser. In essence, we are transforming our XML documents into their final display form by transforming instances of our XML vocabularies into instances of a particular rendering vocabulary.

This Working Draft normatively references XSLT as an integral component of XSL. A stylesheet could be written with both the transformation vocabulary and the formatting semantics vocabulary together; it would style an XML instance by rendering the results of transformation. This result need not be serialized in XML syntax; rather, an XSLT/XSL processor can utilize the result of transformation to create a rendered result by interpreting the abstract hierarchy of information without seeing syntax.

1.1.4.2  Target of transformation

When using a formatting semantics vocabulary as the rendering language, the objective for a stylesheet writer is to convert an XML instance of some arbitrary XML vocabulary into an instance of the formatting semantics vocabulary. The result of transformation cannot contain any user-defined vocabulary construct (for example, an address, customer identifier, or purchase order number construct) because the rendering agent would not know what to do with constructs labeled with these foreign, unknown identifiers.

Consider two examples: HTML for rendering in a web browser and XSL for rendering on screen, on paper or audibly. In both cases, the rendering agents only understand the vocabulary expressing their respective formatting semantics and wouldn't know what to do with alien element types defined by the user.

Just as with HTML, a stylesheet writer utilizing XSL for rendering must transform each and every user construct into a rendering construct to direct the rendering agent to produce the desired result. By learning and understanding the semantics behind the constructs of XSL formatting, the stylesheet writer can create an instance of the formatting vocabulary expressing the desired layout of the final result (e.g. area geometry, spacing, font metrics, etc.), with each piece of information in the result coming from either the source data or the stylesheet itself.

Consider once more the customer information in Example 1-1. An XSL rendering agent doesn't know how to render a marked up construct named <customer>. The XSL vocabulary used to render the customer information could be as follows:

01  <fo:block space-before.optimum="20pt" font-size="20pt">From:
02  <fo:inline-sequence font-style="italic">(Customer Reference)
03  <fo:inline-sequence font-weight="bold">cust123</fo:inline-sequence>
04  </fo:inline-sequence>
05  </fo:block>
Example 1-8: XSL rendering semantics markup for example

The rendering result when using the Portable Document Format (PDF) would then be as follows, with an intermediate PDF generation step interpreting the XSL markup for italics and boldface presentation semantics:

Figure 1-2: 
XSL rendering for example
Figure 1-2: XSL rendering for example

The above again illustrates the two distinctive styling steps: transforming the instance of the XML vocabulary into a new instance according to a vocabulary of rendering semantics; and formatting the instance of the rendering vocabulary in the user agent.

The rendering semantics of much of the XSL vocabulary are device independent, so we can use one set of constructs regardless of the rendering medium. It is the rendering agent's responsibility to interpret these constructs accordingly. In this way, the XSL semantics can be interpreted for print, display, aural or other presentations. There are, indeed, some specialized semantics we can use to influence rendering on particular media, though these are just icing on the cake.

1.1.5  Extensible Stylesheet Language Transformations (XSLT)

We all have needs to transform our structured information when it is not appropriately ordered for a purpose other than how it is created. The XSLT 1.0 Recommendation describes a transformation instruction vocabulary of constructs that can be expressed in an XML model of elements and attributes:

1.1.5.1  Transformation by example

We can characterize XSLT from other techniques for transmuting our information by regarding it simply as "Transformation by Example", differentiating many other techniques as "Transformation by Program Logic". This perspective focuses on the distinction that our obligation is not to tell an XSLT processor how to effect the changes we need, rather, we tell an XSLT processor what we want as an end result, and it is the processor's responsibility to do the dirty work.

The XSLT Recommendation gives us a vocabulary for specifying templates that function as "examples of the result". Based on how we instruct the XSLT processor to access the source of the data being transformed, the processor will incrementally build the result by adding the filled-in templates.

We write our stylesheets, or "transformation specifications", primarily with declarative constructs though we can employ procedural techniques if and when needed. We assert the desired behavior of the XSLT processor based on conditions found in our source. We supply examples of how each component of our result is formulated and indicate the conditions of the source that trigger which component is next added to our result. Alternatively we can selectively add components to the result on demand.

Consider once again the customer information in our example purchase order at Example 1-1. An example of the HTML vocabulary supplied to the XSLT processor to produce the markup in Example 1-7 would be:

01  <xsl:template match="customer">
02    <p><xsl:text>From: </xsl:text>
03      <i><xsl:text>(Customer Reference) </xsl:text>
04        <b><xsl:value-of select="@db"/></b></i></p>
05  </xsl:template>
Example 1-9: Example XSLT template rule for the HTML vocabulary

An example of XSL vocabulary supplied to the XSLT processor to produce the markup in Example 1-8 would be:

01  <xsl:template match="customer">
02    <fo:block space-before.optimum="20pt" font-size="20pt">
03      <xsl:text>From: </xsl:text>
04      <fo:inline-sequence font-style="italic">
05        <xsl:text>(Customer Reference) </xsl:text>
06        <fo:inline-sequence font-weight="bold">
07          <xsl:value-of select="@db"/>
08        </fo:inline-sequence></fo:inline-sequence></fo:block>
09  </xsl:template>
Example 1-10: Example XSLT template rule for the XSL vocabulary

Where XSLT is similar to other transmutation approaches is that we deal with our information as trees of abstract nodes. We don't deal with the raw syntax of our source data. Unlike these other approaches, however, the primary memory management and information manipulation (node traversal and node creation) is handled by the XSLT processor not by the stylesheet writer. This is a significant difference between XSLT and a transformation programming language or interface like the Document Object Model (DOM), where the programmer is responsible for handling the low-level manipulation of information constructs.

XSLT includes constructs which we use to identify and iterate over structures found in the source information. The information being transformed can be traversed in any order needed and as many times as required to produce the desired result. We can visit source information numerous times if the result of transformation requires that information to be present numerous times.

We users of XSLT don't have the burden of implementing numerous practical algorithms required to present information. The designers of XSLT have specified that such algorithms be implemented within the processor itself, and have enabled us to engage these algorithms declaratively. High-level functions such as sorting and counting are available to us on demand when we need them. Low-level functions such as memory-management, node manipulation and garbage collection are all integral to the XSLT processor.

This declarative nature of the stylesheet markup makes XSLT so very much more accessible to non-programmers than the imperative nature of procedurally-oriented transformation languages. Writing a stylesheet is as simple as using markup to declare the behavior of the XSLT processor, much like HTML is used to declare the behavior of the web browser to paint information on the screen.

The designers have also accommodated the programmer as well as the non-programmer in that there are procedural constructs specified. XSLT is (in theory) "Turing complete", thus any arbitrarily complex algorithm could (theoretically) be implemented using the constructs available. While there will always be a trade-off between extending the processor to implement something internally and writing an elaborate stylesheet to implement something portably, there is sufficient expressive power to implement some algorithmic business rules and semantic processing in the XSLT syntax.

In short, straightforward and common requirements can be satisfied in a straightforward fashion, while unconventional requirements can be satisfied to an extent as well with some programming-styled effort.

Note 4:

Theory aside, the necessarily verbose XSLT syntax dictated by its declarative nature and use of XML syntax makes the coding of some complex algorithms a bit awkward. I have implemented some very complex traversals and content generation with successful results, but with code that could be difficult to maintain (my own valiant, if not always satisfactory, documentation practices notwithstanding).

The designers of XSLT recognized the need to maintain large transformation specifications, and the desire to tap prior accomplishments when writing stylesheets so they have included a number of constructs supporting the management, maintenance and exploitation of existing stylesheets. Organizations can build libraries of stylesheet components for sharing among their colleagues. Stylesheet writers can tweak the results of a transformation by writing shell specifications that include or import other stylesheets known to solve problems they are addressing. Stylesheet fragments can be written for particular vocabulary fragments; these fragments can subsequently be used in concert, as part of an organization's strategy for common information description in numerous markup models.

1.1.5.2  Not intended for general purpose XML transformations

It is important to remember that XSLT was designed primarily for transforming XML vocabularies to the XSL formatting vocabulary. This doesn't preclude us from using XSLT for other transformation requirements, but it does influence the design of the language and it does constrain some of the functionality from being truly general purpose.

For this reason, the designers do not claim XSLT is a general purpose transformation language. However, it is still powerful enough for most downstream processing transformation needs, and XSLT stylesheets are often called XSLT transformation scripts because they can be used in many areas not at all related to stylesheet rendering. Consider an electronic commerce environment where transformation is not used for presentation purposes. In this case, the XSLT processor may transform a source instance, which is based on a particular vocabulary, and deliver the results to a legacy application that expects a different vocabulary as input. In other words, we can use XSLT in a non-rendering situation when it doesn't matter what syntax is utilized to represent the content; when only the parsed result of the syntax is material.

An example of using such a legacy vocabulary for the XSLT processor would be:

01  <xsl:template match ="customer">
02    <buyer><xsl:value-of select="@db"/></buyer>
03  </xsl:template>
Example 1-11: Example XSLT template rule for a legacy vocabulary

The transformation would then produce the following result acceptable to the legacy application:

01  <buyer>cust123</buyer>
Example 1-12: Example legacy vocabulary for customer information

The designers of XSLT have focused on the results of delivering parsed XML information to a rendering agent, or to some other application employing an XML processor as the means to access information in an XML instance. The information being delivered represents the parsed result of working with the entire XML instance and, if supplied, the XML document model. The actual markup within the source XML instance is not considered material to the application. All that counts is the result of having processed the XML instance to find the underlying content the actual markup represents.

By focusing on this parsed result for downstream applications, there is little or no regard in an XSLT stylesheet for the actual XML syntax constructs found within the source input documents, or for the actual XML syntax constructs utilized in the resulting output document. This prevents a stylesheet from being aware of such constructs or controlling how such constructs are used. Any transformation requirement that includes "original markup syntax preservation" would not be suited for XSLT transformations.

Note 5:

Is not being able to support "original markup syntax preservation" really a problem? That depends how you regard the original markup syntax used in an XML instance. XML allows you to use various markup techniques to meet identical information representation requirements. If you treat this as merely syntactic sugar for human involvement in the markup process, then it will not be important how information is specifically marked up once it is out of the hands of the human involved. If, however, you are working with transformations where such issues are more than just a sugar coating, and it is necessary to utilize particular constructs based on particular requirements of how the result "looks" in syntactic form, then XSLT will not provide the kind of control you will need.

1.1.5.3  Document model and vocabulary independent

While checking source documents for validity can be very useful for diagnostic purposes, all of the hierarchical relationships of content are based on what is found inside of the instance, not what is found in the document model. The behavior of the stylesheet is specified against the presence of markup in an instance as the implicit model, not against the allowed markup prescribed by any explicit model. Because of this, an XSLT stylesheet is independent of any Document Type Definition (DTD) or other explicit schema that may have been used to constrain the instance at other stages. This is very handy when working with well-formed XML that doesn't have an explicit document model.

If an explicit document model is supplied, certain information such as attribute types and defaulted values enhance the processor's knowledge of the information found in the input documents. Without this information, the processor can still perform stylesheet processing as long as the absence of the information does not influence the desired results.

Without a reliance on the document model for the instance, we can design a single stylesheet that can process instances of different models. When the models are very similar, much of the stylesheet operates the same way each time and the rest of the stylesheet only processes that which it finds in the sources.

It may be obvious but should be stated for completeness that a given source file can be processed with multiple stylesheets for different purposes. This means, though, that it is possible to successfully process a source file with a stylesheet designed for an entirely different vocabulary. The results will probably be totally inappropriate, but there is nothing inherent to an instance that ties it to a single stylesheet or a set of stylesheets. Stylesheet designers might well consider how their stylesheets could validate input; perhaps issuing error messages when unexpected content arrives. However, this is a matter of practice and not a constraint.

1.1.5.4  XML source and stylesheet

The input files to an XSLT processor are one or more stylesheet files and one or more source files. The initial inputs are a single stylesheet file and a single source file. Other stylesheet files are assimilated before the first source file is processed. The XML processor will then access other source files according to the first file's XML content. The XSLT processor may then access other source files at any time under stylesheet control.

All of the inputs must be well-formed (but not necessarily valid) XML documents. This precludes using an HTML file following non-XML lexical conventions, but does not rule out processing an Extensible Hypertext Markup Language (XHTML) file as an input. Many users of existing HTML files that are not XML compliant will need to manipulate or transform them; all that is needed to use XSLT for this is a preprocess to convert existing Standard Generalized Markup Language (SGML) markup conventions into XML markup conventions.

XHTML can be created from HTML using a handy free tool on the W3C site: http://www.w3.org/People/Raggett/tidy/. This tool corrects whatever improperly coded HTML it can and flags any that it cannot correct. When the output is configured to follow XML lexical conventions, the resulting file can be used as an input to the XSLT processor.

1.1.5.5  Validation unnecessary (but convenient)

That an XSLT processor need not incorporate a validating XML processor to do its job does not minimize the importance of source validation when developing a stylesheet. Often when working incrementally to develop a stylesheet by simultaneously working on the test source file and stylesheet algorithm, time can be lost by inadvertently introducing well-formed but invalid source content. Because there is no validation in the XSLT processor, all well-formed source will be processed without errors, producing a result based on the data found. The first reaction of the stylesheet writer is often that a problem has been introduced in the stylesheet logic, when in fact the stylesheet works fine for the intended source data. The real problem is that the source data being used isn't as intended.

Note 6:

Personally, I run a separate post-process source file validation after running the source file through a given stylesheet. While I am examining the results of stylesheet processing, the post process determines whether or not the well-formed file validates against the model to which I'm designing the stylesheet. When anomalies are seen I can check the validation for the possible source of a problem before diagnosing the stylesheet itself.

1.1.5.6  Multiple source files possible

The first source file fed to the XSLT processor defines the first abstract tree of nodes the stylesheet uses.

The stylesheet may access arbitrary other source files, or even itself as a source file, to supplement the information found in the primary file. The names of these supplementary resources can be hardwired into the stylesheet, passed to the stylesheet as a parameter, or the stylesheet can find them in the source files.

A separate node tree represents every resource accessed as a source file, each with its own scope of unique node identifiers and global values. When a given resource is identified more than once as a source file, the XSLT processor creates only a single representation for that resource. In this way a stylesheet is guaranteed to work unambiguously with source information.

1.1.5.7  Stylesheet supplements source

A given transformation result does not necessarily obtain all of its information from the source files. It is often (almost always) necessary to supplement the source with boilerplate or other hardwired information. The stylesheet can add any arbitrary information to the result tree as it builds the result tree from information found in the source trees.

A stylesheet can be the synthesis of the primary file and any number of supplemental files that are included or imported by the main file. This provides powerful mechanisms for sharing and exploiting fragments of stylesheets in different scenarios.

1.1.5.8  Extensible language design supplements processing

The "X" in XSLT stands for "Extensible" for a reason: the designers have built-in conforming techniques for accessing non-conforming facilities requested by a stylesheet writer that may or may not be available in the XSLT processor interpreting the stylesheet. A conforming processor may or may not support such extensions and is only obliged to accommodate error and fallback processing in such a way that a stylesheet writer can reconcile the behavior if needed.

An XSLT processor can implement extension instructions, functions, serialization conventions and sorting schemes that provide functionality beyond what is defined in XSLT 1.0, all accessed through standardized facilities.

A stylesheet writer must not rely on any extension facilities if the XSLT processor being used for the stylesheet is not known or is outside of the stylesheet writer's control. If an end-user base utilizes different brands of XSLT processors, and the stylesheet needs to be portable across all processors, only the standardized facilities can be used.

Standardized presence-testing and fallback facilities can be used by the stylesheet writer to accommodate the ability of a processor to act on extension facilities used in the stylesheet.

1.1.5.9  Abstract structure result

In the same way our stylesheets are insulated from the syntax of our source files, our stylesheets are insulated from the syntax of our result.

We do not focus on the syntax of the file to be produced by the XSLT processor; rather, we create a result tree of abstract nodes, which is similar to the tree of abstract nodes of our input information. Our examples of transformation (converted to nodes from our stylesheet) are added to the result hierarchy as nodes, not as syntax. Our objective as XSLT transformation writers is to create a result node tree that may or may not be serialized externally as markup syntax.

The XSLT processor is not obliged to externalize the result tree if the processor is integral to some process interpreting the result tree for other purposes. For example, an XSL rendering agent may embed an XSLT processor for interpreting the inputs to produce the intermediate hierarchy of XSL rendering vocabulary to be reified in a given medium. In such cases, serializing the intermediate tree in syntax is not material to the process of rendering (though having the option to serialize the hierarchy is a useful diagnostic tool).

The stylesheet writer has little or no control over the constructs chosen by the XSLT processor for serializing the result tree. There are some behaviors the stylesheet can request of the processor, though the processor is not obliged to respect the requests. The stylesheet can request a particular output method be used for the serialization and, if supported, the processor guarantees the final result complies with the lexical requirements of that method.

Note 7:

It is possible to coerce the XSLT processor to violate the lexical rules through certain stylesheet controls that I personally avoid using at all costs. For every XML and HTML instance construct (not including the document model syntax constructs) there are proper XSLT methodologies to follow, though not always as compact as coercing the processor.

The abstract nature of the node trees representing the input source and stylesheet instances and the hands-off nature of serializing the abstract result node tree are the primary reasons that source tree original markup syntax preservation cannot be supported.

The design of the language does, however, support the serialization of the result tree in such a way as not to require the XSLT processor to maintain the result tree in the abstract form. For example, the processor can instantly serialize the start of an element as soon as the element content of the result is defined. There is no need to maintain, nor is there any ability in the stylesheet to add to, the start of an element once the stylesheet begins supplying element content.

The XSLT 1.0 Recommendation defines three output methods for lexically reifying the abstract result tree as serialized syntax: XML conventions, HTML conventions, and simple text conventions. An XSLT processor can be extended to support custom serialization methods for specialized needs.

1.1.5.10  Result-tree-oriented objective

This result abstraction impacts how we design our stylesheets. We have to always remember that the result of transformation is created in result parse order, thus allowing the XSLT processor to immediately serialize the result without maintaining the result for later purposes.

The examples of transformation that we include in our stylesheet already represent examples of the nodes that we want added to the result tree, but we must ensure these examples are triggered to be added to the result tree in result parse order, otherwise we will not get the desired result.

We can peruse and traverse our source files in any predictable order we need to produce the result, but we can only produce the result tree once and then only in result tree parse order. It is often difficult to change traditional perspectives of transformation that focus on the source tree, yet we must look at XSLT transformations focused on the result tree.

The predictable orders we traverse the source trees are not restricted to only source tree parse order (also called document order). Information in the source trees can be ignored or selectively processed. The order of the result tree dictates the order in which we must access our source trees.

Note 8:

I personally found this required orientation difficult to internalize, having been focused on the creation of my source information long before addressing issues of transforming the sources to different results. Understanding this orientation is key to quickly producing results using XSLT.

It is not, however, an XSLT processor implementation constraint to serially produce the result tree. This is an important distinction in the language design that supports parallelism. An XSLT processor supporting parallelism can simultaneously produce portions of the result tree provided only that the end result is created as if it were produced serially.

1.1.6  Namespaces

To successfully use and distinguish element types in our instances as being from given vocabularies, the Namespaces in XML Recommendation gives us means to preface our element type names to make them unique. The Recommendation and the following widely-read discussion document describe the precepts for using this technique:

1.1.6.1  Vocabulary distinction

It would be unreasonable to mandate that all document models have mutually unique element type names. We design our document models with our own business requirements and our own naming conventions; so do other users. A W3C working group developing vocabularies has its own conventions and requirements; so do other committees. An XML-based application knowing that an instance is using element types from only a single vocabulary can easily distinguish all elements by the name, since each element type is declared in the model by its name.

But what happens when we need to create an XML instance that contains element types from more than one vocabulary? If all the element types are uniquely named then we could guess the vocabulary for a given element by its name. But if the same name is used in more than one vocabulary, we need a technique to avoid ambiguity. Using cryptically compressed or unmanageably elongated element type names to guarantee uniqueness would make XML difficult to use and would only delay the problem to the point that these weakened naming conventions would still eventually result in vocabulary collisions.

Note 9:

Enter the dreaded namespaces: a Recommendation undeserving of its sullied reputation. This is a powerful, yet very simple technique for disambiguating element type names in vocabularies. Perhaps the reputation spread from those unfamiliar with the requirements being satisfied. Perhaps concerns were spread by those who made assumptions about the values used in namespace declarations. As unjustified as it is, evoking namespaces unnecessarily (and unfortunately) strikes fear in many people. It is my goal to help the reader understand that not only are namespaces easy to define and easy to use, but that they are easy to understand and are not nearly as complex as others have believed.

The Namespaces in XML Recommendation describes a technique for exploiting the established uniqueness of Uniform Resource Identifier (URI) values under the purview of the Internet Engineering Task Force (IETF). We users of the Internet accept the authority of the registrar of Internet domain names to allot unique values to organizations, and it is in our best interest to not arrogate or usurp values allotted to others as our own. We can, therefore, assume a published URI value belongs to the owner of the domain used as the basis of the value. The value is not a Uniform Resource Locator (URL), which is a URI that identifies an actual addressed location on the Internet; rather, the URI is being used merely as a unique string value.

To set the stage for how these URI values are used, consider an example of two vocabularies that could easily be used together in an XML instance: the Scalable Vector Graphics (SVG) vocabulary and the Mathematical Markup Language (MathML). In SVG the <set> element type is used to scope a value for reference by descendent elements. In MathML the <set> element type defines a set in the mathematical sense of a collection.

Remembering that names in XML follow rigid lexical constraints, we pick out of thin air a prefix we use to distinguish each element type from their respective vocabulary. The prefix we choose is not mandated by any organization or any authority; in our instances we get to choose any prefix we wish. We should, however, make the prefix meaningful or we will obfuscate our information, so let's choose in this example to distinguish the two element types as <svg:set> and <math:set>. Note that making the prefix short is a common convention supporting human legibility, and using the colon ":" separating the prefix from the rest of the name is prescribed by the Namespaces in XML recommendation.

While we are talking about names, let's not forget that some Recommendations utilize the XML name lexical construct for other purposes, such as naming facilities that may be available to a processor. We get to use this namespace prefix we've chosen on these names to guarantee uniqueness, just as we have done on the names used to indicate element types.

1.1.6.2  URI value association

But having the prefix is not enough because we haven't yet guaranteed global identity or singularity by a short string of name characters; to do so we must associate the prefix with a globally unique URI before we use that prefix. Note that we are unable to use a URI directly as a prefix because the lexical constraints on a URI are looser than those of an XML name; the invalid XML name characters in a URI would cause an XML processor to balk.

We assert the association between a namespace prefix and a namespace URI by using a namespace declaration attribute as in the following examples:

As noted earlier, the prefix we choose is arbitrary and can be any lexically valid XML name. The prefix is discarded by the namespace-aware processor, and is immaterial to the application using the names; it is only a syntactic shortcut to get at the associated URI. The associated URI supplants the prefix in the internal representation of the name value and the application can distinguish the names by the new composite name that would have been illegal in XML syntax. There is no convention for documenting a namespace qualified name using its associated URI, but one way to perceive the uniqueness is to consider our example as it might be internally represented by an application:

The specification of a URI instead of a URL means that the namespace-aware processor will never look at the URI as a URL to accomplish its work. There never need be any resource available at the URI used in a namespace declaration. The URI is just a string and its value is used only as a string and the fact that there may or may not be any resource at the URL identified by the URI is immaterial to namespace processing. The URI does not identify the location of a schema, or a DTD or any file whatsoever when used by a namespace aware processor.

Note 10:

Perhaps some of the confusion regarding namespaces is rooted in the overloading of the namespace URI by some Recommendations. These Recommendations require that the URI represent a URL where a particular resource is located, fetched, and utilized to some purpose. This behavior is outside the scope of namespaces and is mandated solely by the Recommendations that require it.

Practice has, however, indicated an end-user-friendly convention regarding the URI used in namespace declarations. The W3C has placed a documentation file at every URL represented by a namespace URI. Requesting the resource at the URL returns an HTML document discussing the namespace being referenced, perhaps a few pointer documents to specifications or user help information, and any other piece of helpful information deemed suitable for the public consumption. This convention should help clear up many misperceptions about the URI being used to obtain some kind of machine-readable resource or schema, though it will not dispel the misperception that there needs to be some resource of some kind at the URL represented by a namespace URI.

So now a processor can unambiguously distinguish an element's type as being from a particular vocabulary by knowing the URI associated with the vocabulary. Our choice of prefix is arbitrary and of no relevance. The URI we have associated with the prefix used in a namespace-qualified XML name (often called a QName) informs the processor of the identity of the name. Our choice of prefix is used and then discarded by the processor, while the URI persists and is the basis of namespace-aware processing. We have achieved uniqueness and identity in our element type names and other XML names in a succinct legible fashion without violating the lexical naming rules of XML.

1.1.6.3  Namespaces in XSL and XSLT

Namespaces identify different constructs for the processors interpreting XSL formatting specifications and XSLT stylesheets.

An XSL rendering agent responsible for interpreting an XSL formatting specification will recognize those constructs identified with the http://www.w3.org/1999/XSL/Format namespace. Note that the year value used in this URI value is not used as a version indictor; rather, the W3C convention for assigning namespace URI values incorporates the year the value was assigned to the working group.

An XSLT processor responsible for interpreting an XSLT stylesheet recognizes instructions and named system properties using the http://www.w3.org/1999/XSL/Transform namespace. An XSLT processor will not recognize using an archaic value for working draft specifications of XSLT.

XSLT use namespace-qualified names to identify extensions that implement non-standardized facilities. A number of kinds of extensions can be defined in XSLT including functions, instructions, serialization methods, sort methods and system properties.

The XT XSLT processor written by James Clark is an example of a processor implementing extension facilities. XT uses the http://www.jclark.com/xt namespace to identify the extension constructs it implements. Remembering that this is a URI and not a URL, you will not find any kind of resource or file when using this value as a URL.

We also use our own namespaces in an XSLT stylesheet for two other purposes. We need to specify the namespaces of the elements and attributes of our result if the process interpreting the result relies on the vocabulary to be identified. Furthermore, our own non-default namespaces distinguish internal XSLT objects we include in our stylesheets. Each of these will be detailed later where such constructs are described.

1.1.7  Stylesheet association

When we wish to associate with our information one or more preferred or suitable stylesheet resources geared to process that information, the W3C stylesheet association Recommendation describes the syntax and semantics for a construct we can add to our XML documents:

1.1.7.1  Relating documents to their stylesheets

XML information in its authored form is often not organized in an appropriate ordering for consumption. A stylesheet association processing instruction is used at the start of an XML document to indicate to the recipient which stylesheet resources are to be used when reading the contents of that document.

The recipient is not obliged to use the resources referenced and can choose to examine the XML using any stylesheet or transformation process they desire by ignoring the preferences stated within. Some XML applications ignore the stylesheet association instruction entirely, while others choose to steadfastly respect the instruction without giving any control to the recipient. A flexible application will let the recipient choose how they wish to view the content of the document.

The designers of this specification adopted the same semantics of the <LINK> construct defined in the HTML 4.0 recommendation:

1.1.7.2  Ancillary markup

A processing instruction is ancillary to the XML document model constraining the creation and validation of an instance. Therefore, we do not have to model the presence of this construct when we design our document model. Any instance can have any number of stylesheet associations added into the document during or after creation, or even removed, without impacting on the XML content itself.

An application respecting this construct will process the document content with the stylesheet before delivering the content to the application logic. Two cases of this are the use of a stylesheet for rendering to a browser canvas and the use of a transformation script at the front end of an e-commerce application.

The following two examples illustrate stylesheet associations that, respectively, reference an XSL resource and a Cascading Stylesheet (CSS) resource:

01  <?xml-stylesheet href="fancy.xsl" type="text/xsl"?>
Example 1-13: Associating an XSL stylesheet
01  <?xml-stylesheet href="normal.css" type="text/css"?>
Example 1-14: Associating a CSS stylesheet

The following example naming the association for later reference and indicating that it is not the primary stylesheet resource is less typical, but is allowed for in the specification:

01  <?xml-stylesheet alternate="yes" title="small" 
02                   href="small.xsl" type="text/xsl"?>
Example 1-15: Alternative stylesheet association

A URL that does not include a reference to another resource, but rather is defined exclusively by a local named reference, specifies a stylesheet resource that is located inside the XML document being processed, as in the following example:

01  <?xml-stylesheet href="#style1" type="text/xsl"?>
Example 1-16: Associating an internal stylesheet

The Recommendation designers expect additional schemes for linking stylesheets and other processing scripts to XML documents to be defined in future specifications.

Note 11:

Embedding stylesheet association information in an XML document and using the XML processing instruction to do so are both considered stopgap measures by the W3C. This Recommendation cautions readers that no precedents are set by employing these makeshift techniques and that urgency dictated their choice. Indeed, there is some question as to the appropriateness of tying processing to data so tightly, and we will see what considered approaches become available to us in the future.

This is a prose version of an excerpt from the book "Practical Transformation Using XSLT and XPath" (Eighth Edition ISBN 1-894049-05-5 at the time of this writing) published by Crane Softwrights Ltd., written by G. Ken Holman; this excerpt was edited by Stan Swaren, and reviewed by Dave Pawson.

The Context of XSL Transformations and the XML Path Language (cont'd)

1.2  Transformation data flows

Here we look at the interactions between some of the Recommendations we focus on by examining how our information flows through processes engaging or supporting the technologies.

1.2.1  Transformation from XML to XML

As we will see when looking at the data model, the normative behavior of XSLT is to transform an XML source into an abstract hierarchical result. We can request that result to be serialized into an XML file, thus we achieve XML results from XML sources:

Figure 1-3: Transformation from XML to XML
Figure 1-3: Transformation from XML to XML

An XSLT stylesheet can be applied to more than one XML document, each stylesheet producing a possibly (usually) different result. Nothing in XSLT inherently ties the stylesheet to a single instance, though the stylesheet writer can employ techniques to abort processing based on processing undesirable input.

An XML document can have more than one XSLT stylesheet applied, each stylesheet producing a possible (usually) different result. Even when stylesheet association indicates an author's preference for a stylesheet to use for processing, tools should provide the facility to override the preference with the reader's preference for a stylesheet. Nothing in XML prevents more than a single stylesheet to be applied.

Note 12:

In all cases in this chapter the depictions show the normative result of the XSLT processor's as the dotted triangle attached to the process rectangle. This serves to remind the reader that the serialization of the result into an XML file is a separate task, one that is the responsibility of the XSLT processor and not the stylesheet writer.

In all diagrams, the left-pointing triangle represents a hierarchically-marked up document such as an XML or HTML document. This convention stems from considering the apex of the hierarchy at the left, with the sub-elements nesting within each other towards the lowest leaves of the hierarchy at the right of the triangle.

Processes are depicted in rectangles, while arbitrary data files of some binary or text form are depicted in parallelograms. Other symbols representing screen display, print and auditory output are drawn with (hopefully) obvious shapes.

1.2.2  Transformation from XML to XSL formatting semantics

When the result tree is specified to utilize the XSL formatting vocabulary, the normative behavior of an XSL processor incorporating an XSLT processor is to interpret the result tree. This interpretation reifies the semantics expressed in the constructs of the result tree to some medium, be it pixels on a screen, dots on paper, sound through a synthesis device, or another medium that makes sense for presentation.

Figure 1-4: Transformation from XML to XSL Formatting Semantics
Figure 1-4: Transformation from XML to XSL Formatting Semantics

Without employing extension techniques or supplemental documentation, the stylesheets used in this scenario contain only the transformation vocabulary and the resulting formatting vocabulary. There are no other element types from other vocabularies in the result, including from the source vocabulary. For example, rendering processors would not inherently know what to do with an element of type custnbr representing a customer number; it is the stylesheet writer's responsibility to transform the information into information recognized by the rendering agent.

There is no obligation for the rendering processor to serialize the result tree created during transformation. The feature of serializing the result tree to XML syntax is, however, quite useful as a diagnostic tool, revealing to us what we really asked to be rendered instead of what we thought we were asking to be rendered when we saw incorrect results. There may also be performance considerations of taking the reified result tree in XML syntax and rendering it in other media without incurring the overhead of performing the transformation repeatedly.

1.2.3  Transformation from XML to non-XML

An XSLT processor may choose to recognize the stylesheet writer's desire to serialize a non-XML representation of the result tree:

Figure 1-5: Transformation from XML to Aware Non-XML
Figure 1-5: Transformation from XML to Aware Non-XML

The XSLT Recommendation documents two non-XML tree serialization methods that can be requested by the stylesheet writer. When the processor offers serialization, it is only obliged to reify the result using XML lexical and syntax rules, and may support producing output following either HTML lexical and syntax rules or simple text.

1.2.3.1  HTML lexical and syntactic conventions

Internet web browsers are specific examples of the generic HTML user agent. User agents are typically designed for instances of HTML following the precursor to XML: the Standard Generalized Markup Language (SGML) lexical conventions. Certain aspects of the HTML document model also dictate syntactic shortcuts available when working with SGML.

While some more recently developed user agents will accept XML lexical conventions, thus accepting Extensible Hypertext Markup Language (XHTML) output from an XSLT processor, older user agents will not. Some of these user agents will not accept XML lexical conventions for empty elements, while some require SGML syntax minimization techniques to compress certain attribute specifications.

Additionally, user agents recognize a number of general entity references as built-in characters supporting accented letters, the non-breaking space, and other characters from public entity sets defined or used by the designers of HTML. An XSLT processor recognizes the use of these characters in the result tree and serializes them using the assumed built-in general entities.

1.2.3.2  Text lexical conventions

An XSLT processor can be asked to serialize only the #PCDATA content of the entire result tree, resulting in a file of simple text without employing any markup techniques. All text is represented by the characters' individual values, even those characters sensitive to XML interpretation.

Note 13:

I use the text method often for synthesizing MSDOS batch files. By walking through my XML source I generate commands to act on resources identified therein, thus producing an executable batch file tailored to the information.

1.2.3.3  Arbitrary binary and custom lexical conventions

Many of our legacy systems or existing applications expect information to follow custom lexical conventions according to arbitrary rules. Often, this format is raw binary not following textual lexical patterns. We are usually obliged to write custom programs and transformation applications to convert our XML information to these non-standardized formats due to their binary or non-structured natures.

XSLT can play a role even here where the target format is neither structured, nor text, nor in any format anticipated by the designers of the Recommendation. We do have a responsibility to fill in a critical piece of the formula described below, but we can leverage this single effort in the equation to allow us and our colleagues to continue to use W3C Recommendations with our XML data.

Not using XSLT to produce custom output

Consider first the scenario without using XSLT where we must write individual XML-aware applications to accommodate our information vocabularies. For each of our vocabularies we need separate programs to convert to the common custom format required by the application. This incurs programming resources to accommodate any and every change to our vocabularies in order to meet the new inputs to satisfy the same custom output.

Figure 1-6: 
Accommodating multiple inputs with different XML vocabularies
Figure 1-6: Accommodating multiple inputs with different XML vocabularies

Using XSLT to produce custom output

If, however, we focus on the custom output instead of focusing on our vocabulary inputs, we can leverage a single investment in programming across all of our vocabularies. Moreover, by being independent of the vocabulary used in the source, we can accommodate any of our or others' vocabularies we may have to deal with in the future.

The approach involves us creating our own custom markup language based on a critical analysis of the target custom format to distill the semantics of how information is represented in the resulting file. These semantics can be expressed using an XML vocabulary whose elements and attributes engage the features and functions of the resulting format. We must not be thinking of our source XML vocabularies, rather, our focus is entirely on the semantics of what exactly makes up our target custom format. Let's refer to this custom format's XML vocabulary we divine from our analysis as the Custom Vocabulary Markup Language (CVML).

Using our programming resources we can then write a single transformation application responsible for interpreting XML instances of CVML to produce a file following the custom format. This transformation application could be written using the Document Object Model (DOM) as a basis for tree-oriented access to the information. Alternatively, a SAX-based application can interpret the instances to produce the outputs if the nature of CVML lends itself to that orientation. The key is that regardless of how instances of CVML are created, the interpretation of CVML markup to produce an output file never changes. Our one CVML Instance Interpreter application can produce any custom format output file expressible in the CVML semantics.

Getting back to our own or others' XML vocabularies, we have now reduced the problem to XML instance transformation. Our objective is simplified to produce XML instances of CVML from instances of our many input XML vocabularies. This is a classical XSLT situation and we need only write XSLT stylesheets combining the XSLT instructions with CVML as the result vocabulary. Our investment in XSLT for our colleagues is leveraged by the CVML Instance Interpreter so that they can now take their XML and use stylesheets to produce the binary or custom lexical format.

Figure 1-7: Transformation from XML to an arbitrary format
Figure 1-7: Transformation from XML to an arbitrary format

This approach separates the awareness of the lexical and syntactic requirements of the custom output format from the numerous stylesheets we write for all of our possible input XML vocabularies. Our colleagues use XSLT just as they would with HTML or XSL as a result vocabulary. They leverage the single investment in producing the custom format by using the CVML Interpreter to serialize the results of their transformations to produce the files designed for other applications. This, in turn, leverages the investment in learning and using XSLT in the organization.

Taking this two steps further

First, the "X" in XSLT represents the word "extensible" and result tree serialization is one of the areas where we can extend an XSLT processor's functionality. This allows us to implement non-standard vendor-specific or application-specific output serialization methods and engage these facilities in a standard manner. As with all extension mechanisms in XSLT, the trigger is the use of an XML namespace recognized by the XSLT processor implementing the extension:

01  xmlns:prefix="processor-recognized-URI"
02  <xsl:output method="prefix:serialization-method-name"/>
Example 1-17: Using namespaces to specify an extension serialization method
Comment:
- the namespace declaration attribute on line 1 must be somewhere in the element or the ancestry of the instruction on line 2

Using the same semantics described for the outboard CVML Interpreter program depicted in Figure 1-7, this translation facility can be incorporated into the XSLT processor itself as an inboard extension. The code itself may be directly portable based on the nature of how the outboard program is written. Such an extended processor would directly emit the custom format without reifying the intermediate structure (though this would be convenient for diagnostic purposes):

Figure 1-8: Built-in Transformation from XML to Arbitrary Non-XML
Figure 1-8: Built-in Transformation from XML to Arbitrary Non-XML

The XT XSLT processor implements an extension serialization method named NXML for "non-XML":

01  xmlns:prefix="http://www.jclark.com/xt"
02  <xsl:output method="prefix:nxml"/>
Example 1-18: Using the XT namespace to specify the NXML extension serialization method
Comment:
- the namespace declaration attribute on line 1 must be somewhere in the element or the ancestry of the instruction on line 2

Second, this extensibility opens up the opportunity to use an XSLT processor as a front-end to any application that can be modified to access the result tree. The intermediate result tree of CVML is not serialized externally; rather, it is fed directly to the application and the application interprets the internal representation of the content that would have been serialized to a custom format. Time is saved by not serializing the result tree and having the application parse the reified file back into a memory representation; performance is enhanced by the application directly accessing the result of transformation.

When generalized, a vendor's non-XML-based application can use this approach to accommodate arbitrary customers' XML vocabularies merely by writing W3C conforming XSLT stylesheets as the "interpretation specification". Some XSLT processors can build a DOM representation of result tree or deliver the result tree as Simple API for XML (SAX) events, thus giving an application developer standardized interfaces to the transformed information expressed using the application's custom semantics vocabulary. The developer's programming is then complete and the vendor accommodates each customer vocabulary with an appropriate stylesheet for translation to the application semantics.

1.2.4  Three-tiered architectures

A three-tiered architecture can meet technical and business objectives by delivering structured information to web browsers by using XSLT on the host, or on the user agent, or even on both.

Considering technical issues first, the server can distribute the processing load to XML/XSLT-aware user agents by delivering a combination of the stylesheet and the source information to be transformed on the recipient's platform. Alternatively, the server can perform the transformations centrally to accommodate those user agents supporting only HTML or HTML/CSS vocabularies:

Figure 1-9: Server-side Transformation Architecture
Figure 1-9: Server-side Transformation Architecture

There may be good business reasons to selectively deliver richly-marked-up XML to the user agent or to arbitrarily transform XML to HTML on the server regardless of the user agent capabilities. Even if it is technically possible to send semantically-rich information in XML, protecting your intellectual property by hiding the richness behind the security of a "semantic firewall" must be considered. Perhaps there are revenue opportunities by only delivering a richly marked-up rendition of your information to your customers. Perhaps you could even scale the richness to differing levels of utility for customers who value your information with different granularity or specificity, while preserving the most detailed normative version of the data away of view.

Lastly, there are no restrictions to using two XSLT processes: one on the server to translate our organization's rich markup into an arbitrary delivery-oriented markup. This delivery markup, in turn, is translated using XSLT on the user agent for consumption by the operator. This approach can reduce bandwidth utilization and increase distributed processing without sacrificing privacy.

Note 14:

There is no consensus in our XML community that semantic firewalls are a "good thing". Peers of mine preach that the World Wide Web must always be a semantic web with rich markup processed in a distributed fashion among user agents being de rigueur. Personally, I do not subscribe to this point of view. We have the flexibility to weigh the technical and business perspectives of our customers' needs for our information, our own infrastructure and processing capabilities, and our own commercial and privacy concerns. We can choose to "dumb down" our information for consumption, and the installed base of user agents supporting presentation-oriented semantic-less HTML can be the perfect delivery vehicle to protect these concerns of ours.

This is a prose version of an excerpt from the book "Practical Transformation Using XSLT and XPath" (Eighth Edition ISBN 1-894049-05-5 at the time of this writing) published by Crane Softwrights Ltd., written by G. Ken Holman; this excerpt was edited by Stan Swaren, and reviewed by Dave Pawson.

XML.com Copyright © 1998-2006 O'Reilly Media, Inc.