Each of these tasks maps to an aspect of DTD syntax or to a DTD operator, which together form the building blocks of a DTD. In my next column, I will address attributes, entities, and some of the more advanced aspects of DTDs. This month, we'll look at the element-specific tasks and bring them to bear on a hypothetical résumé.

Preconstruction Analysis
Once you've analyzed a representative subset of the documents in your document set, diced them up into a suitable level of structured detail, defined your markup strategy as well as your tags and attributes, and applied them to several documents in your set (see my previous tutorial [XML-J, Vol. 2, issue 6] for more information), you're ready to begin creating your DTD, which will no doubt be an iterative process.

You can create a DTD in one of two ways: (1) write it from scratch, by hand, as I later demonstrate, or, (2) you can make use of a DTD-creation tool. XML Authority (free trial version available from TIBCO Extensibility at www.extensibility.com/solutions/trial.htm) simplifies DTD construction. And it checks for errors, too, a few of which you'll probably make in your first DTD. It also has a help section on best practices.

Declaring Elements and Their Contents
Let's say we want all the résumés in a set of them to contain the root element résumé. To declare an element, you use the <!ELEMENT> declaration. Thus, to declare the résumé element in the DTD, write:

<!ELEMENT resume ANY>

This statement declares that the element résumé may appear with, as the DTD keyword ANY indicates, any combination of text and child nodes. The syntax of the <!ELEMENT> declaration is this:

<!ELEMENT elementName {rule}>

where {rule} may be replaced either by a DTD keyword in all capitals like ANY or EMPTY, by a parentheses-enclosed rule with one or more child elements separated either by commas indicating that their appearance is required in the order specified, or by a vertical bar, indicating choice. Don't panic; I'll show you examples of all this later.

In our example with the résumé element above, the syntactic slot for the rule filled by the keyword ANY is used to describe the legal content - a content model - of the résumé element. In this case it can be anything from child elements to text.

However, using the ANY does little to constrain our data, defying the purpose of our DTD, which is to make explicit the permissible relationships and associations among the data in a set of XML documents. Thus, in the rule slot where I've used the keyword ANY, better constraints can and should be put in place. Instead of simply saying that the resume element can take any combination of child elements and text, let's state exactly what children it may contain.

Specifying Children
After analyzing and marking up a couple of résumés in the document set, I've settled on a firm, high-level structure for the content for all the résumés I'm collecting for publication on a Web site. The high-level structure of each resume will be the same and must conform to a set pattern: every resume will have a root element called résumé that will contain the following child elements in the following order: name, contactInfo, experience, and education. Thus, in an XML document, I will structure the résumé's high-level data like this:

<resume>
<name>...</name>
<contactInfo>...</contactInfo>
<experience>...</experience>
<education>...</education>
</resume>

To make this grammar explicit, I specify the child elements and their ordering with the following multiple sequence:

<!ELEMENT resume (name, contactInfo, experience, education) >

This DTD rule says that the résumé element contains the children enclosed in parentheses. The commas in the rule stipulate that the elements must appear in the order in which they are listed. No other child elements or text are permitted directly under the résumé node. Such a rule, when related to the XML document in which the elements occur, is called a content model.

Occurrence Operators
A set of DTD operators, called occurrence operators, allows authors to generalize about the optionality and frequency with which elements may occur. If no occurrence operator is used, the default is exactly one occurrence. Table 1 summarizes the occurrence operators and the rule each one implements.

Now let's say we want the <experience> element in our resume to have the following structure:

<experience>
<position></position>
<company></company>
<location></location>
<task></task>
<note></note>
</experience>

<position> and <company> are both mandatory elements and must appear in each résumé in our set once and only once; <location>, however, may appear once or not at all but not multiple times. Meantime, <task> must appear at least once but may be reused as often as needed, while <note> is a fully optional element to add any additional information that isn't captured in the other elements: it may be used once, more than once, or not at all. To instill these rules about occurrence into your content model for the element <experience>, append the occurrence operator as a suffix to the appropriate element, as the following example shows:

<!ELEMENT experience (position, company, location?, task+, note*) >

Since neither <position> nor <company> are followed by an occurrence operator, they receive the default setting. They must appear in the XML document exactly one time.

Nesting with Parentheses
You can also use parentheses to nest elements within a rule and then apply any of the occurrence operators to all the elements within the nested set. For example:

<!ELEMENT location ((street, suite)?, city, (state, zip)?)>

This rule says that the (street, suite) sequence is optional but may not be repeated (note, though, that if the street element is used, so must the suite element). Ditto with the (state, zip) sequence. The city element, however, must appear exactly once, and if the other elements are used, they must be positioned according to the order dictated by the comma-separated list.

By the way, you can also nest a choice of elements by placing them in parentheses and separating them by vertical bars instead of commas. For instance, the following rule says that the name element must contain a first name, optionally followed either by a choice of a middle name, a middle initial, or by a nickname, followed by a required last name.

<!ELEMENT name (firstName, (
(middleName | middleInitial)? |
(nickName)? )*, lastName)>

Yes, this can get complex quickly. Since I don't want to bore you to death with needless complications, I'll stop with all this nesting balderdash and instead point you to a reference that explains in more detail how to use parentheses to form complex rules: Chapter 3, "Document Type Definitions," in XML in a Nutshell, by Elliotte Rusty Harold and W. Scott Means (O'Reilly).

Mixing Content
The guts of many narrative-oriented XML documents, such as software manuals, essays, or magazine articles, use mixed content - a choice of either text or elements or a combination of both. For instance, our sample résumé has an as-yet undefined element called <task>, the contents of which aim to describe what the résumé's author did in any given job. Here's an example of the kind of description that would go inside our <task> tag, with additional markup:

<task>Served as consultant for editing, electronic production, and desktop publishing of one financial newsletter, called <cite>Securities Today</cite>, and the organization, design, and launch of two others. <paragraph>All three became <emphasis>highly successful</emphasis>publications.</paragraph></task>

Within the task element, then, the résumé's author could have any combination of text and the <cite>, <paragraph>, and <emphasis> tags. Though not used in the task description, let's also make a line-break element, named <br> as in HTML, available.

To declare a rule that allows mixed content like this, you must first declare the text and then list the other elements, all separated by vertical bars to indicate choice and marked as optional and repeatable with the asterisk occurrence operator. Here's the rule:

<!ELEMENT task (#PCDATA | paragraph | cite | emphasis | br )*>

#PCDATA? What in the world is this? PCDATA is XML's name for standard text (though I'm oversimplifying a bit). PCDATA stands for parsed character data, which includes regular text characters, except <, &, or the sequence ]]>. PCDATA also includes general entities, which I'll discuss in my next column.

Because a DTD must declare the content model for each element used in the XML document, our DTD must include declarations for the paragraph, cite, emphasis, and <br> elements. With the exception of the <paragraph> and <br> elements, they contain only PCDATA, which, used alone, excludes other element tags. Thus:

<!ELEMENT paragraph (#PCDATA | cite | emphasis)>
<!ELEMENT cite (#PCDATA )>
<!ELEMENT emphasis (#PCDATA )>

Finally, empty elements, like my HTML-like <br> element in the rule for <task> above, may be declared using the EMPTY keyword:

<!ELEMENT br EMPTY>.

This ensures that no content - whether other elements or parsed character data - may be placed within it.

Declaring mixed content can get tricky. The key is to remember that if it includes PCDATA, then PCDATA must be declared first. And all the items in the rule must be separated by the vertical bar indicating choice and the rule must be marked as optional and repeatable with an asterisk.

For more of the nasty little details about mixed content, see the chapter in XML in a Nutshell on DTDs. If you want to dive straight into a full course on DTDs, David Megginson's book, Structuring XML Documents (Prentice Hall), explains all the nuances of building industrial-strength DTDs. For a concise introduction to DTDs, see Robert Eckstein's XML Pocket Reference (O'Reilly); I suggest you read its short but potent section on DTDs several times.

My next column will offer the second installment in DTD construction, picking up where this one left off. It'll cover attributes, entities, and the more advanced aspects of working with DTDs. Meantime, begin writing a DTD for your résumé by declaring its elements and the rules associated with them.

Don't worry about attributes just yet. Remember, though, to declare the content for every element, as I've done in the preceding examples for the child elements of <task> (but did not do yet for many of the other elements). In my next tutorial, we'll combine all the elements and attributes into a complete DTD.