A Technical Introduction to XML
Author's Note: It is somewhat remarkable to think that this article, which appeared initially in the Winter 1997 edition of the World Wide Web Journal was out of date by the time the final XML Recommendation was approved in February. And even as this update brings the article back into line with the final spec, a new series of recommendations are under development. When finished, these will bring namespaces, linking, schemas, stylesheets, and more to the table.
This introduction to XML presents the Extensible Markup Language at a reasonably technical level for anyone interested in learning more about structured documents. In addition to covering the XML 1.0 Specification, this article outlines related XML specifications, which are evolving. The article is organized in four main sections plus an appendix.
What is XML?
What Do XML Documents Look Like?
Pulling the Pieces Together
XML is a markup language for documents containing structured information.
Structured information contains both content (words, pictures, etc.) and some indication of what role that content plays (for example, content in a section heading has a different meaning from content in a footnote, which means something different than content in a figure caption or content in a database table, etc.). Almost all documents have some structure.
A markup language is a mechanism to identify structures in a document. The XML specification defines a standard way to add markup to documents.
The number of applications currently being developed that are based on, or make use of, XML documents is truly amazing (particularly when you consider that XML is not yet a year old)! For our purposes, the word "document" refers not only to traditional documents, like this one, but also to the myriad of other XML "data formats". These include vector graphics, e-commerce transactions, mathematical equations, object meta-data, server APIs, and a thousand other kinds of structured information.
No. In HTML, both the tag semantics and the tag set are fixed. An <h1> is always a first level heading and the tag <ati.product.code> is meaningless. The W3C, in conjunction with browser vendors and the WWW community, is constantly working to extend the definition of HTML to allow new tags to keep pace with changing technology and to bring variations in presentation (stylesheets) to the Web. However, these changes are always rigidly confined by what the browser vendors have implemented and by the fact that backward compatibility is paramount. And for people who want to disseminate information widely, features supported by only the latest releases of Netscape and Internet Explorer are not useful.
XML specifies neither semantics nor a tag set. In fact XML is really a meta-language for describing markup languages. In other words, XML provides a facility to define tags and the structural relationships between them. Since there's no predefined tag set, there can't be any preconceived semantics. All of the semantics of an XML document will either be defined by the applications that process them or by stylesheets.
No. Well, yes, sort of. XML is defined as an application profile of SGML. SGML is the Standard Generalized Markup Language defined by ISO 8879. SGML has been the standard, vendor-independent way to maintain repositories of structured documentation for more than a decade, but it is not well suited to serving documents over the web (for a number of technical reasons beyond the scope of this article). Defining XML as an application profile of SGML means that any fully conformant SGML system will be able to read XML documents. However, using and understanding XML documents does not require a system that is capable of understanding the full generality of SGML. XML is, roughly speaking, a restricted form of SGML.
For technical purists, it's important to note that there may also be subtle differences between documents as understood by XML systems and those same documents as understood by SGML systems. In particular, treatment of white space immediately adjacent to tags may be different.
In order to appreciate XML, it is important to understand why it was created. XML was created so that richly structured documents could be used over the web. The only viable alternatives, HTML and SGML, are not practical for this purpose.
HTML, as we've already discussed, comes bound with a set of semantics and does not provide arbitrary structure.
SGML provides arbitrary structure, but is too difficult to implement just for a web browser. Full SGML systems solve large, complex problems that justify their expense. Viewing structured documents sent over the web rarely carries such justification.
This is not to say that XML can be expected to completely replace SGML. While XML is being designed to deliver structured content over the web, some of the very features it lacks to make this practical, make SGML a more satisfactory solution for the creation and long-time storage of complex documents. In many organizations, filtering SGML to XML will be the standard procedure for web delivery.
The XML specification sets out the following goals for XML: [Section 1.1] (In this article, citations of the form [Section 1.1], these are references to the W3C Recommendation Extensible Markup Language (XML) 1.0. If you are interested in more technical detail about a particular topic, please consult the specification)
XML is defined by a number of related specifications:
As time goes on, additional requirements will be addressed by other specifications. Currently (Sep, 1998), namespaces (dealing with tags from multiple tag sets), a query language (finding out what's in a document or a collection of documents), and a schema language (describing the relationships between tags, DTDs in XML) are all being actively pursued.
For the most part, reading and understanding the XML specifications does not require extensive knowledge of SGML or any of the related technologies.
One topic that may be new is the use of EBNF to describe the syntax of XML. Please consult the discussion of EBNF in the appendix of this article for a detailed description of how this grammar works.
If you are conversant with HTML or SGML, XML documents will look familiar. A simple XML document is presented in Example 1.
Example 1. A Simple XML Document
<?xml version="1.0"?> <oldjoke> <burns>Say <quote>goodnight</quote>, Gracie.</burns> <allen><quote>Goodnight, Gracie.</quote></allen> <applause/> </oldjoke>
A few things may stand out to you:
XML documents are composed of markup and content. There are six kinds of markup that can occur in an XML document: elements, entity references, comments, processing instructions, marked sections, and document type declarations. The following sections introduce each of these markup concepts.
Elements are the most common form of markup. Delimited by angle brackets, most elements identify the nature of the content they surround. Some elements may be empty, as seen above, in which case they have no content. If an element is not empty, it begins with a start-tag, <element>, and ends with an end-tag, </element>.
Attributes are name-value pairs that occur inside start-tags after the element name. For example,
is a div element with the attribute class having the value preface. In XML, all attribute values must be quoted.
In order to introduce markup into a document, some characters have been reserved to identify the start of markup. The left angle bracket, < , for instance, identifies the beginning of an element start- or end-tag. In order to insert these characters into your document as content, there must be an alternative way to represent them. In XML, entities are used to represent these special characters. Entities are also used to refer to often repeated or varying text and to include the content of external files.
Every entity must have a unique name. Defining your own entity names is discussed in the section on entity declarations. In order to use an entity, you simply reference it by name. Entity references begin with the ampersand and end with a semicolon.
For example, the lt entity inserts a literal < into a document. So the string <element> can be represented in an XML document as <element>.
A special form of entity reference, called a character reference [Section 4.1], can be used to insert arbitrary Unicode characters into your document. This is a mechanism for inserting characters that cannot be typed directly on your keyboard.
Character references take one of two forms: decimal references, ℞, and hexadecimal references, ℞. Both of these refer to character number U+211E from Unicode (which is the standard Rx prescription symbol, in case you were wondering).
Comments begin with <!-- and end with -->. Comments can contain any data except the literal string --. You can place comments between markup anywhere in your document.
Comments are not part of the textual content of an XML document. An XML processor is not required to pass them along to an application.
Processing instructions (PIs) are an escape hatch to provide information to an application. Like comments, they are not textually part of the XML document, but the XML processor is required to pass them to an application.
Processing instructions have the form: <?name pidata?>. The name, called the PI target, identifies the PI to the application. Applications should process only the targets they recognize and ignore all other PIs. Any data that follows the PI target is optional, it is for the application that recognizes the target. The names used in PIs may be declared as notations in order to formally identify them.
PI names beginning with xml are reserved for XML standardization.
In a document, a CDATA section instructs the parser to ignore most markup characters.
Consider a source code listing in an XML document. It might contain characters that the XML parser would ordinarily recognize as markup (< and &, for example). In order to prevent this, a CDATA section can be used.
<![CDATA[ *p = &q; b = (i <= 3); ]]>
Between the start of the section, <![CDATA[ and the end of the section, ]]>, all character data is passed directly to the application, without interpretation. Elements, entity references, comments, and processing instructions are all unrecognized and the characters that comprise them are passed literally to the application.
The only string that cannot occur in a CDATA section is ]]>.
A large percentage of the XML specification deals with various sorts of declarations that are allowed in XML. If you have experience with SGML, you will recognize these declarations from SGML DTDs (Document Type Definitions). If you have never seen them before, their significance may not be immediately obvious.
One of the greatest strengths of XML is that it allows you to create your own tag names. But for any given application, it is probably not meaningful for tags to occur in a completely arbitrary order. Consider the old joke example introduced earlier. Would this be meaningful?
<gracie><quote><oldjoke>Goodnight, <applause/>Gracie</oldjoke></quote> <burns><gracie>Say <quote>goodnight</quote>, </gracie>Gracie.</burns></gracie>
It's so far outside the bounds of what we normally expect that it's nonsensical. It just doesn't mean anything.
However, from a strictly syntactic point of view, there's nothing wrong with that XML document. So, if the document is to have meaning, and certainly if you're writing a stylesheet or application to process it, there must be some constraint on the sequence and nesting of tags. Declarations are where these constraints can be expressed.
More generally, declarations allow a document to communicate meta-information to the parser about its content. Meta-information includes the allowed sequence and nesting of tags, attribute values and their types and defaults, the names of external files that may be referenced and whether or not they contain XML, the formats of some external (non-XML) data that may be referenced, and the entities that may be encountered.
There are four kinds of declarations in XML: element type declarations, attribute list declarations, entity declarations, and notation declarations.
Element type declarations [Section 3.2] identify the names of elements and the nature of their content. A typical element type declaration looks like this:
<!ELEMENT oldjoke (burns+, allen, applause?)>
This declaration identifies the element named oldjoke. Its content model follows the element name. The content model defines what an element may contain. In this case, an oldjoke must contain burns and allen and may contain applause. The commas between element names indicate that they must occur in succession. The plus after burns indicates that it may be repeated more than once but must occur at least once. The question mark after applause indicates that it is optional (it may be absent, or it may occur exactly once). A name with no punctuation, such as allen, must occur exactly once.
Declarations for burns, allen, applause and all other elements used in any content model must also be present for an XML processor to check the validity of a document.
In addition to element names, the special symbol #PCDATA is reserved to indicate character data. The moniker PCDATA stands for parseable character data .
Elements that contain only other elements are said to have element content [Section 3.2.1]. Elements that contain both other elements and #PCDATA are said to have mixed content [Section 3.2.2].
For example, the definition for burns might be
<!ELEMENT burns (#PCDATA | quote)*>
The vertical bar indicates an or relationship, the asterisk indicates that the content is optional (may occur zero or more times); therefore, by this definition, burns may contain zero or more characters and quote tags, mixed in any order. All mixed content models must have this form: #PCDATA must come first, all of the elements must be separated by vertical bars, and the entire group must be optional.
Two other content models are possible: EMPTY indicates that the element has no content (and consequently no end-tag), and ANY indicates that any content is allowed. The ANY content model is sometimes useful during document conversion, but should be avoided at almost any cost in a production environment because it disables all content checking in that element.
Here is a complete set of element declarations for Example 1:
Example 2. Element Declarations for Old Jokes
<!ELEMENT oldjoke (burns+, allen, applause?)> <!ELEMENT burns (#PCDATA | quote)*> <!ELEMENT allen (#PCDATA | quote)*> <!ELEMENT quote (#PCDATA)*> <!ELEMENT applause EMPTY>
Attribute list declarations [Section 3.3] identify which elements may have attributes, what attributes they may have, what values the attributes may hold, and what value is the default. A typical attribute list declaration looks like this:
<!ATTLIST oldjoke name ID #REQUIRED label CDATA #IMPLIED status ( funny | notfunny ) 'funny'>
In this example, the oldjoke element has three attributes: name, which is an ID and is required; label, which is a string (character data) and is not required; and status, which must be either funny or notfunny and defaults to funny, if no value is specified.
Each attribute in a declaration has three parts: a name, a type, and a default value.
You are free to select any name you wish, subject to some slight restrictions [Section 2.3, production 5], but names cannot be repeated on the same element.
There are six possible attribute types:
There are four possible default values:
The XML processer performs attribute value normalization [Section 3.3.3] on attribute values: character references are replaced by the referenced character, entity references are resolved (recursively), and whitespace is normalized.
Entity declarations [Section 4.2] allow you to associate a name with some other fragment of content. That construct can be a chunk of regular text, a chunk of the document type declaration, or a reference to an external file containing either text or binary data.
A few typical entity declarations are shown in Example 3.
Example 3. Typical Entity Declarations
<!ENTITY ATI "ArborText, Inc."> <!ENTITY boilerplate SYSTEM "/standard/legalnotice.xml"> <!ENTITY ATIlogo SYSTEM "/standard/logo.gif" NDATA GIF87A>
There are three kinds of entities:
Internal entities can include references to other internal entities, but it is an error for them to be recursive.
The XML specification predefines five internal entities:
<!ELEMENT burns (#PCDATA | quote)*> <!ELEMENT allen (#PCDATA | quote)*>
<!ENTITY % personcontent "#PCDATA | quote"> <!ELEMENT burns (%personcontent;)*> <!ELEMENT allen (%personcontent;)*>
Notation declarations [Section 4.7] identify specific types of external binary data. This information is passed to the processing application, which may make whatever use of it it wishes. A typical notation declaration is:
<!NOTATION GIF87A SYSTEM "GIF">
As we've seen, XML content can be processed without a document type declaration. However, there are some instances where the declaration is required:
In applications where a person composes or edits the data (as opposed to data that may be generated directly from a database, for example), a DTD is probably going to be required if any structure is to be guaranteed.
If present, the document type declaration must be the first thing in the document after optional processing instructions and comments [Section 2.8].
The document type declaration identifies the root element of the document and may contain additional declarations. All XML documents must have a single root element that contains all of the content of the document. Additional declarations may come from an external DTD, called the external subset, or be included directly in the document, the internal subset, or both:
<?XML version="1.0" standalone="no"?> <!DOCTYPE chapter SYSTEM "dbook.dtd" [ <!ENTITY %ulink.module "IGNORE"> <!ELEMENT ulink (#PCDATA)*> <!ATTLIST ulink xml:link CDATA #FIXED "SIMPLE" xml-attributes CDATA #FIXED "HREF URL" URL CDATA #REQUIRED> ]> <chapter>...</chapter>
This example references an external DTD, dbook.dtd, and includes element and attribute declarations for the ulink element in the internal subset. In this case, ulink is being given the semantics of a simple link from the XLink specification.
Note that declarations in the internal subset override declarations in the external subset. The XML processor reads the internal subset before the external subset and the first declaration takes precedence.
In order to determine if a document is valid, the XML processor must read the entire document type declaration (both internal and external subsets). But for some applications, validity may not be required, and it may be sufficient for the processor to read only the internal subset. In the example above, if validity is unimportant and the only reason to read the doctype declaration is to identify the semantics of ulink, reading the external subset is not necessary.
You can communicate this information in the standalone document declaration [Section 2.9]. The standalone document declaration, standalone="yes" or standalone="no" occurs in the XML declaration. A value of yes indicates that only internal declarations need to be processed. A value of no indicates that both the internal and external declarations must be processed.
In addition to markup, there are a few other issues to consider: white space handling, attribute value normalization, and the language in which the document is written.
White space handling [Section 2.10] is a subtle issue. Consider the following content fragment:
<oldjoke> <burns>Say <quote>goodnight</quote>, Gracie.</burns>
Is the white space (the new line between <oldjoke> and <burns> ) significant?
But how can you tell? You can only determine if white space is significant if you know the content model of the elements in question. In a nutshell, white space is significant in mixed content and is insignificant in element content.
The rule for XML processors is that they must pass all characters that are not markup through to the application. If the processor is a validating processor [Section 5.1], it must also inform the application about which whitespace characters are significant.
The special attribute xml:space may be used to indicate explicitly that white space is significant. On any element which includes the attribute specification xml:space='preserve', all white space within that element (and within subelements that do not explicitly reset xml:space ) is significant.
The only legal values for xml:space are preserve and default. The value default indicates that the default processing is desired. In a DTD, the xml:space attribute must be declared as an enumerated type with only those two values.
One last note about white space: in parsed text, XML processors are required to normalize all end-of-line markers to a single line feed character (&#A;) [Section 2.11]. This is rarely of interest to document authors, but it does eliminate a number of cross-platform portability issues.
The XML processer performs attribute value normalization [Section 3.3.3] on attribute values: character references are replaced by the referenced character, entity references are resolved (recursively), and whitespace is normalized.
Many document processing applications can benefit from information about the natural language in which a document is written, XML defines the attribute xml:lang [Section 2.12] to identify the language. Since the purpose of this attribute is to standardize information across applications, the XML specification also describes how languages are to be identified.
Given the preceding discussion of type declarations, it follows that some documents are valid and some are not. There are two categories of XML documents: well-formed and valid.
A document can only be well-formed [Section 2.1] if it obeys the syntax of XML. A document that includes sequences of markup characters that cannot be parsed or are invalid cannot be well-formed.
In addition, the document must meet all of the following conditions (understanding some of these conditions may require experience with SGML):
By definition, if a document is not well-formed, it is not XML. This means that there is no such thing as an XML document which is not well-formed, and XML processors are not required to do anything with such documents.
A well-formed document is valid only if it contains a proper document type declaration and if the document obeys the constraints of that declaration (element sequence and nesting is valid, required attributes are provided, attribute values are of the correct type, etc.). The XML specification identifies all of the criteria in detail.
The XPointer and XLink specifications, currently under development, introduce a standard linking model for XML. In consideration of space, and the fact that the XLink draft is still developing, what follows is survey of the features of XLink, rather than a detailed description of the specification.
In the parlance of XLink, a link expresses a relationship between resources. A resource is any location (an element, or its content, or some part of its content, for example) that is addressed in a link. The exact nature of the relationship between resources depends on both the application that processes the link and semantic information supplied.
Some highlights of XLink are:
Since XML does not have a fixed set of elements, the name of the linking element cannot be used to locate links. Instead, XML processors identify links by recognizing the xml:link attribute. Other attributes can be used to provide additional information to the XML processor. An attribute renaming facility exists to work around name collisions in existing applications.
Two of the attributes, show and actuate allow you to exert some control over the linking behavior. The show attribute determines whether the document linked-to is embeded in the current document, replaces the current document, or is displayed in a new window when the link is traversed. actuate determines how the link is traversed, either automatically or when selected by the user.
Some applications will require much finer control over linking behaviors. For those applications, standard places are provided where the additional semantics may be expressed.
A Simple Link strongly resembles an HTML <A> link:
<link xml:link="simple" href="locator">Link Text</link>
A Simple Link identifies a link between two resources, one of which is the content of the linking element itself. This is an in-line link.
The locator identifies the other resource. The locator may be a URL, a query, or an Extended Pointer.
Extended Links allow you to express relationships between more than two resources:
<elink xml:link="extended" role="annotation"> <locator xml:link="locator" href="text.loc">The Text</locator> <locator xml:link="locator" href="annot1.loc">Annotations </locator> <locator xml:link="locator" href="annot2.loc"> More Annotations</locator> <locator xml:link="locator" href="litcrit.loc"> Literary Criticism</locator> </elink>
This example shows how the relationships between a literary work, annotations, and literary criticism of that work might be expressed. Note that this link is separate from all of the resources involved.
Extended Links can be in-line, so that the content of the linking element (other than the locator elements), participates in the link as a resource, but that is not necessarily the case. The example above is an out-of-line link because it does not use its content as a resource.
Cross references with the XML ID/IDREF mechanism (which is similar to the #fragment mechanism in HTML) require that the document being linked-to has defined anchors where links are desired (and technically requires that both the ID and the IDREF occur in the same document). This may not always be the case and sometimes it is not possible to modify the document to which you wish to link.
XML XPointers borrow concepts from HyTime and the Text Encoding Initiative (TEI). XPointers offer a syntax that allows you to locate a resource by traversing the element tree of the document containing the resource.
locates the third child (whatever it may be) of the second oldjoke in the document.
XPointers can span regions of the tree. The XPointer
selects the second and third oldjoke s in the document.
In addition to selecting by elements, XPointers allow for selection by ID, attribute value, and string matching. In this article, the XPointer
selects the first occurance of the word "Here" in the What Do XML Documents Look Like? section of this article. The link can be established by an extended link without modifying the target document.
Note that an XPointer range can span a structurally invalid section of the document. The XLink specification does not specify how applications should deal with such ranges.
Out-of-line links introduce the possibility that an XML processor may need to process several files in order to correctly display the hypertext document.
Following the annotated text example above, assuming that the actual text is read only, the XML processor must load at least the text and the document that contains the extended link.
XLink defines Extended Link Groups for this purpose. Loading an Extended Link Group communicates which documents must be loaded to the XML processor. Extended Link Groups can be used recursively, and a steps attribute is provided to limit the depth of recursion.
Some documents, particularly compound documents pulled together with XLinks, are likely to be composed of elements from multiple tag sets. For example, a technical article might be written using one DTD, but include mathematical equations written in MathML and vector graphics written in a third DTD.
In order for a processing application to associate the correct semantics with an element, it must know which tag set the element comes from. XML solves this problem with namespaces. Namespaces in XML describes this system in detail.
The principle is to allow a colon-delimited prefix to be associated with some external semantic via a URI. Then use of that prefix identifies the element as having the semantics described by the URI. For example:
<bk:para>The fraction 3/4 can be expressed in MathML as: <ml:cn type="rational">3<ml:sep/>4</ml:cn>.</bk:para>
The para element in this example is explicitly identified as being in the namespace identified by the bk prefix, which must have been defined earlier in the document, and the cn and sep elements come from the ml namespace (presumably associated in some way with MathML).
HTML browsers are largely hard-coded. Although some browsers can base their formatting on Cascading Style Sheets (CSS), they still contain hard-coded conventions for documents which do not provide a stylesheet. A first level heading appears the way that it does largely because the browser recognizes the <h1> tag.
Again, since XML documents have no fixed tag set, this approach will not work. The presentation of an XML document is dependent on a stylesheet.
The standard stylesheet language for XML documents is the Extensible Style Language (XSL). At the time of this writing, the XSL effort is well underway, but many questions remain unanswered. The XSL Working Group produced its first Working Draft on 18 Aug 1998.
Other stylesheet languages, like Cascading Style Sheets, are likely to be supported as well.
In this article, most of the major features of the XML Language have been discussed, and some of the concepts behind XLink, Namespaces, and XSL have been described. Although some things have been left out in the interest of the big picture (such as character encoding issues), hopefully you now have enough background to pick up and read the XML Specifications without difficulty.
One of the most significant design improvements in XML is to make it easy to use with modern compiler tools. Part of this improvement involves making it possible to express the syntax of XML in Extended Backus-Naur Form (EBNF) [Section 6]. If you've never seen EBNF before, think of it this way:
Let's take a simple example that has nothing to do with XML (or the real rules of language):
 Word ::= Consonant Vowel+ Consonant  Consonant ::= [^aeiou]  Vowel ::= [aeiou]
Rule 1 states that a word is a consonant followed by one or more vowels followed by another consonant. Rule 2 states that a consonant is any letter other than a, e, i, o, or u. Rule 3 states that a vowel is any of the letters a, e, i, o, or u. (The exact syntax of the rules, the meaning of square brackets and other special symbols, is laid out in the XML specification.)
Using the above example, is this red a Word? Yes.
By the same analysis, reed , road , and xeaiioug are also words, but rate is not. There is no way to match Consonant Vowel Consonant Vowel using the EBNF above. XML is defined by an EBNF grammar of about 80 rules. Although the rules are more complex, the same sort of analysis allows an XML parser to determine that <greeting>Hello World</greeting> is a syntactically correct XML document while <greeting]Wrong Bracket!</greeting> is not.
In very general terms, that's all there is to it. You'll find all the details about EBNF in Compilers: Principles, Techniques, and Tools by Aho, Sethi, and Ullman or in any modern compiler text book.
While EBNF isn't an efficient way to represent syntax for human consumption, there are programs that can automatically turn EBNF into a parser. This makes it a particularly efficient way to represent the syntax for a language that will be parsed by a computer.
XML.com Copyright © 1998-2006 O'Reilly Media, Inc.