Bizarre fantasies of an Extensible Mark-Up Language

I'm awake at 05:00 unable to sleep and instead working on migrating the working copy of my web site to my laptop so that I can eventually retire my old desktop. One of the importunate thoughts bouncing about in my sleep-deprived brain is something someone at work said about some weirdos he'd heard of who actually (ab)use XML as some sort of (standard) generalized mark-up language. This bizarre (to him) concept involves one choosing a set of XML tags to express the structure of the text (as if text could have structure!), which is ludicrous because how would anyone be able to read it? One would have to use XSLT to transform it in to HTML, so why not use HTML in the first place, eh?

In a way this shows just how much the data-heads have beaten the the document-processing crowd when it comes to managers' mind-share. By data-heads I mean those people who cannot see or understand anything unless it is expressible as relational database tables. Nothing else exists: Object-oriented analysis and design is merely a complex variation on entity-relationship diagrams, themselves an affectation when you should just be diving in to a wysiwyg schema-designer; windows and dialogue boxes and GUIs in general only make sense as views of a database table (or occasionally of the results of a query); XML exists purely as a serialization format for database tables.

The last of these is particularly annoying. XML is a lousy serialization format. If XML had been designed as a text-based serialization of data, then it would look more like YAML than SGML. In fact, for database programmers' needs, even YAML is overqualified; for expressing collections of tabular data, you could just use comma-separated values for each table, packaged as a unit using multipart MIME:

Content-Type: multipart/related; boundary=foobar;

Content-type: application/tables+csv; charset=Windows-1252

Content-type: application/csv; charset=Windows-1252; headerrows=1
Content-id: <>

[... 45000 more rows ...]
Content-type: application/csv; charset=Windows-1252; headerrows=1
Content-id: <>

id,name,cars,suit cost

This would have the advantage of recycling a lot of existing code: database programs already understand how to write and read CSV, and if you are functioning on the internet you need a MIME processor. The application/tables+csv section is a bit dull, and might be extended to include schema information for the tables (a subset of ANSI SQL DDL). All in all, if all the world needs is a way to exchange the contents of relational databases, then this is a zillion times better a solution than XML.

On the other hand, there is more to life than RDBMSs, and as a mark-up language, XML is ... well ... adequate. There are bits I would personally have omitted, such as NOTATION declarations and those features of DTDs that prevent XML documents from being stand-alone (namely declaration of &-entitites, and default attributes). But its main asset is its openness to recursive nesting of elements and mixed-content elements (those combining text with nested elements), features that dataheads loathe because they do not fit in to relational database concepts at all.

And no, people should not need to invent their own tag-sets in order to write a note for the milkman; that is what we have committees thrashing out XML formats like DOCBOOK and XHTML for. But when you want to do a critical edition of Shakespeare, the extra effort of hammering out a new XML format makes sense. Of course you then find you need overlapping elements, and end up wishing for LMNL.