By late 1999 it was obvious that these concerns wouldn't be resolved in XML Schema, so the various experts chose instead to tackle the issues by designing their own schema languages. Away from the constraints of the W3C process, they tried quite different approaches. The first was a language called DSD (Document Structure Definition). In contrast to XML Schema, this approach defines a schema as a set of rules (such as if element name = X, then content is "this") rather than declarative patterns. This approach wasn't widely embraced, and work ended sometime in 2002 or 2003.

A more important approach came with Schematron, introduced in late 1999 and developed largely by Rick Jelliffe. Unlike the grammar-based approach of XML Schema (where the schema defines allowed structural patterns), Schematron defines required structures as a set of required tree patterns expressed using XPath expressions. When comfortable with a recursive tree pattern-based way of thinking, this proves to be a powerful approach to schema specification. Indeed, Schematron has become modestly popular and is being adopted as an ISO standard.

Introduced in early 2000, RELAX (Regular Language description for XML) is a grammar-based schema approach Murata Makoto developed. Unlike XML Schema, RELAX doesn't demand deterministic content models and has a much cleaner separation between grammatical rules and data types. RELAX simply incorporates the primitive types defined by XML Schema and doesn't define its own. Importantly, the RELAX design is based on a formal mathematical foundation (called hedge automata theory) that ensures all schemas defined in RELAX are provably valid. This also lets RELAX describe structures that XML Schema (with its deterministic content model constraint) can't.

TREX (Tree Regular Expressions for XML), developed by James Clark, followed shortly on the heels of RELAX. TREX uses simple, easy-to-understand regular expressions to define grammatical constraints/rules. This language is based on an approach taken in XDuce, a programming language developed in early 2000 and designed specifically for processing XML. Roughly put, Clark took the type system in XDuce, expressed it in XML syntax, and added the features necessary for a minimal schema language. This language focused on structure and ignored primitive types. On the structural side, however, TREX was similar to that of RELAX.

In late 2001, TREX and RELAX were merged under the auspices of the ISO to form a new language called RELAX-NG, or RNG for short. This synthesized language retained the benefits of both parents: a simple to understand syntax, clean separation between structural patterns, and primitive types; mathematically rigorous underpinnings; and the ability to model structures that simply aren't possible with XML Schema. Moreover, the simpler underlying model means that element and attribute models are treated almost identically, which greatly simplifies writing and understanding schemas. RNG has a wide following of avid supporters and is relatively well-supported by modern XML Schema tools.

Last but not least is Examplotron. Developed by Eric van der Vlist, it uses XML instance documents to provide examples of the allowed structure. In its purest form, Examplotron is limited to modeling documents that are like the provided example. However, Examplotron also lets designers include RELAX-NG expressions to add more complex rules into an Examplotron schema. Thus, it is often best to think of Examplotron as a short cut to defining simple RNG schemas.

Next month, we will begin to look at some of these other schema approaches in far more detail. Look for it on CPUmag.com.

by Ian Graham


(You can find a document that provides links to online information related to this article at www.utoronto.ca/ian/articles/feb05.)