The easier a specification is to implement, the more implementations will be made and the more compatible they will be. Easy implementation also means that implementers have time to be creative and come up with novel ways to use the technology. Implementations will be cheaper; in fact they can be virtually free if they can be made by individuals in their spare time. Fixing bugs is among the most expensive operations known to developers, especially if there are people that, unwittingly, have come to rely on them. Avoiding bugs is thus of the foremost importance. Implementability can come from simplicity or from building on well-known, existing components.

Building on existing components means using technologies that programmers are familiar with, such as context-free grammars (expressed as EBNF or some variant) or XML 1.0.

It does not mean adding dependencies on new or ill-understood technologies. XSLT, e.g., should probably not have used XML Namespaces. It could have distinguished between commands and data in more conventional ways. XSLT is now in danger of undergoing the same evolution towards incompatible subsets as HTML, which used SGML, another technology not understood by developers.

As Jakob Nielsen says, it is unsafe to use a new specification until at least a year after it has become official ([Nielsen00], page 34).

W3C's specifications thus far aren't very complex, at least not when compared to those of some other standards organizations, but they are already getting more intricate. W3C probably has to look at the IETF again for how to keep specs simple, because already one of the most often heard complaints is that the number of interdependencies, especially among the XML family of specs, is becoming hard to deal with, especially for developers who create their first XML-based programs (and if we want the Web to develop, we'll need many first-time developers).

Indeed, XML 1.0 itself is already too complicated for some of the applications for which it was designed, as shown by the project that has been started on the Internet to define a simplified version (see [Park00]).

The existence of two implementations of a specification is no guarantee that there will be more, but it does give a hint. It is therefore a good idea that many W3C working groups now ask for two or more implementations before submitting their work as a "Proposed Recommendation."

See also

Nielsen, Jakob. Designing Web usability. Indianapolis, IN. Dec 1999. New Riders.
Park, Don (ed). Minimal XML 1.0. 11 Apr 2000. Draft developed on the SML-DEV [SMLDEV] mailing list. URL:
sml-dev mailing list. 29 Nov 1999. Simple Markup Languages Developer group mailing-list. URL: