To E or Not To E: Elements vs. Attributes

You may have noticed in the Collection Markup Language example a few seeming inconsistencies such as:

<Published publisher="Harmony Books">1992</Published>

compared with

<Chart>
  <Peak>1</Peak> <!-- US implied -->
  <Peak country="UK">2</Peak>
</Chart>

for the Book and CD parent elements, respectively. Why didn't I make a Publisher element, or why didn't I make peakUS and peakUK attributes, instead of children elements of Chart? Would these have worked?

<Publication>
  <Published>1992</Published>
  <Publisher>Harmony Book</Publisher>
</Publication>

and

<Chart peakUS="1" peakUK="2" />

Believe it or not, while not visually equivalent, they are considered lexically equivalent. Either form would work just fine.

But which is better? Is it best to make almost everything elements, with attributes only used for details? Or is it better to create fewer elements and use attributes liberally? Ah, that is the $64,000 question! In fact, this question has been kicking around the SGML community since at least 1992. The consensus is that there is no one answer that applies to all situations. The decision depends on what you intend to do with the markup. Generally speaking, if my intention with Collection ML is to treat publisher as a relatively unimportant detail (not to be accessed often), whereas if I consider peak chart positions for any country listed to be something I'll want to access frequently, then my original definitions are appropriate. Some of the major differences between elements and attributes are:

• order of attributes cannot be controlled (element order can be specified in a content model)
• attributes cannot be hierarchical
• attributes are more like metadata than content
• attributes are also useful for enumerated values (name token groups)

For details, examples, and many other opinions, see Robin Cover's page on this subject, SGML/XML: Using Elements and Attributes. There is also a useful set of guidelines by G. Ken Holman posted to the newsgroup comp.xml.text in early 1999.

Writing Our Own ML without a Net: Examples
Doing It With XML, Part 1
Viewing It With IE5, Take 1