Structured document formats like DITA, DocBook, and Solbook are characterized by deeply nested tags and a multitude of schema constraints. Unstructured tagging languages like HTML, on the other hand, are wide open. In one meeting, every reason we came up with that made them seem necessary, was answered by a convincing counter argument. "Reuse" would seem to be the most important reason. And maybe there are some compelling cases. But maybe all-out reuse isn't needed. Maybe we really only need a very restricted form that solves those cases. In at least the case of version-dimension reuse, variable substitution and conditional metadata seem to be a darn good idea. And in at least the case of table and list tags, nesting seems to be a requirement. So it's clearly not the case that we can completely do without such capabilities.

On the other hand, the counter arguments against other forms of variable substitution and conditional metadata remain intact -- at times, it is just too costly to keep them working, especially in an environment that changes frequently. And nesting everything may well be overkill, when so few forms of nesting are actually indispensable. This post summarizes the arguments we considered. Do they demolish the case for structured documents in a highly fluid setting like the software industry? Do they demolish the case for structured documents and reuse? Are they wrong in some important respect? Or do they overlook some vitally important point that makes structured document formats irreplacable?

Read the complete article by Eric Armstrong.