Linkin' Park

A recent discussion on XML-DEV turned to the question of what was left undone in the great XML project, and thus conversation turned again to linking. In this column, I'll attempt to summarise the debate, which contained plenty of resonance for those who've been following XML awhile.

Can We Make Linking Work?

Jonathan Robie picked up on my comment in last week's column that XQuery was the "last great project of XML standardization," and asked XML-DEV if they thought there was anything left to do. Bob DuCharme quickly noted that some thought XML linking was still an unattained goal.

Liam Quin provided his assessment of the state of linking, and in particular its deployment on the Web. It's not purely, if at all, a technical problem, says Quin.

I think part of the problem with XLink's deployment on the Web is that it didn't have clear relationship to HTML and to web-based multimedia. There are lots of use cases for "linkbases," not least in connection with distributed annotations, but the ability to take an arbitrary attribute in an XML or XHTML document, or the content of an arbitrary element, and say, "use this URI as the destination for *this* link, with this text in French and this in Italian, and if the user clicks here, offer a choice of these seven links with their associated titles" was a step beyond the horizon of most web developers, and old hat to the SGML hypertext crowd.

I don't know how to bridge that gap, and I'm not sure who does, if anyone, but until it's bridged, I don't think we'll see great leaps forward in the XLink area. It's cultural and political more than technical.

Quin goes on to provide a list of what he would have liked to have seen on the Web by now. It's not strictly related to the linking discussion, but worth noting just to reinforce the lowly state of browser technology.

The corporate desktop isn't somewhere you'd expect to see innovation, and indeed, now that the corporate desktop seems to drive the Web, we're not seeing innovation.

I'd love to see a more sophisticated XSL-FO being used in browsers. I'd love to see web browsers making use of embedded RDF to display information about web pages. I'd love to see multi-channel audio being used on the Web, both for accessibility and for the user experience.

So, that's linking from the Web's point of view. What about another big user of linking, SVG? Robin Berjon provided feedback on XLink's deployment in that specification.

On the good side: contrary to a lot of what has been heard, people don't complain about the namespace thing. Having discoverable links is generally liked ... with the increased integration of SVG with arbitrary namespaces, having clearly identified links is turning out to be quite useful.

On the bad side: you can't have an XLink without specifying at least two attributes, xlink:href and xlink:type, even for simple links. That's something I would call a glitch and would suggest fixing with an erratum ... Another downside is that the simple web authoring community didn't get the href/src/ref distinction that it's used to dealing with anywhere near as easily as without XLink, while the link junkies get nowhere near where they would like to go.

So much for an assessment of XLink's success this far. Not everyone's on board with the idea that XLink's a necessity anyway. No, thank you, said Michael Kay.

Hyperlinks belong in the user interface space; XML should represent information independently of the user interface. It was always architecturally wrong to do hyperlinking at the XML level and the attempt should not be repeated.

"Modelling relationships in XML" -- that would be different.

What consummate skill in setting the stage for the debate! In just two paragraphs Kay not only baited the acolytes of Nelson, but provided a foothold for the RDF permathread to get started too. Surely a classic discussion is in prospect.

Let's take the user interface point first. Len Bullard pointed out that practically all implementations of linking in the past have at least partly used queries and procedural link building anyway. So is there any point in pursuing a declarative language for putting links in documents, when we can do what we want with mechanisms such as JavaScript. In response, Quin outlined some persuasive points for including links in documents.

Google can't follow procedural links, and isn't likely to. Procedural (ECMAscript) links are hard to manage and maintain. They're hard to reason about. They're hard for archiving bots to follow. They're often not made accessible, because the web designer made them rather than the browser/UA developer. They can be hard to internationalize.

Now onto modeling relationships. The notion of relationships between data items tends to subsume links, which become relationships that are made navigable in some way by an application. Enter stage-left XML linking's prodigal son, Ben Trafford.

I think that modeling relationships sets a foundation for linking. Linking behavior belongs specifically in the user interface space. And we've seen that CSS and XSL-FO can be used hand-in-hand with XML quite nicely to determine document behavior.

You might at this point be reminded, as Len Bullard was, of the Opera browser's implementation of links in CSS.

Trafford continues to note that the separation of links from the content authoring is all very well from the modeling point of view, but it provides real difficulties for content authors themselves.

I may have a specific reason when authoring content to want a link to appear in a certain way, but I may have no skill or ability to write the stylesheet that makes sure that happens. I need to know how to write my tag to specify the linking behavior I want.

In other words, I need for that preference to be somehow displayed at the XML level.

Kay sniffed out Trafford's document-centricity, and accepted those use cases, but returned to the problems of XLink as a solution to the problem.

Part of the problem, I think, is the focus on URIs as identifiers (and links). I've heard a number of talks recently advocating that we should use URIs whenever we want to identify anything, and I simply don't think that's the right direction. To my mind <postcode>RG4 7BS</postcode> is a perfectly good identifier (for a small piece of geography in which my house is found), and any technology that requires me to write it differently if I'm going to use it for linking purposes is too constraining.

So what would Kay do? Not so fast, he was only defining the problem, not a solution. But the unease certainly runs deep:

I'm not even comfortable that the hierarchic relationships should be special. Why can't we have multiple hierarchic views of the same network? Why do all my queries have to change depending on whether my footnotes are inline, out-of-line referenced by IDREFs, or in external documents referenced by URI? What happened to the old doctrine of data independence?

How to break out of the hierarchies? Meanwhile, Quin suggests that queries are one way to do this, or maybe RDF-based annotations.

Speaking of RDF, doesn't it solve all the relationship-modeling problems XLink could anyway? DuCharme suggests that once user interface is out of the picture, RDF should do.

I think that modeling relationships is a bit ambitious, though; a "model" makes me think of an ordered structure of components. Perhaps "representing relationships" would be the lower-hanging fruit, but we don't need a new standard to say that resource X has relationship Y to resource Z; we've got RDF for that.

Ben Trafford couldn't disagree more.

You're proposing we say, "Look, you want links? Well, first write yourself a DTD. Then a stylesheet. Oh, and by the way, you'll need to write an RDF vocabulary to represent the data relationships, too. Hope you like colons!"

Somehow, I just don't think people will be jumping on that bandwagon.

Taken on the level of using the RDF/XML syntax, Trafford has a point. However, integration of any linking technology with RDF would make lots of sense on the Web, given that RDF definitely has been a success within its sphere. Earlier in the discussion Quin again affirms this in the context of web metadata.

Embedding metadata makes a lot of sense -- e.g. the author of this document. Doing so in a model that's compatible with RDF also makes some sense. I don't care about the syntax.

Most RDF aficionados would agree with this viewpoint these days: it's the model that counts.

So where will a solution be found that works effectively across documents? Mike Champion echoes Quin's suggestions that query could be an answer, picking up on Kay's link-by-value use case for his postcode.

Maybe the lesson here is that the relational model approach of defining links dynamically based on relationships on the values of information items rather than predefined links really is the way to do what XLink tried to do.

The thing I always liked best about XQuery is simply the addition of a Join operation into the XML corpus. Until this thread I hadn't thought of this as a replacement for XLink, but that idea is starting to take root in my head ...

Eric van der Vlist opined that RDF was in just the right place to do this, clarifying that he meant of course the data model and not the syntax! One of his reasons was that he didn't think Champion's addition of the Join operation was enough.

... the Join operation isn't enough if all you can join are tree fragments that are by nature not "merge friendly" and one of the biggest (and usually underestimated) benefit of RDF is its ability to "auto-merge" information from multiple sources.

And there we will leave the linking debate for now. In short: more questions and few answers. Personally I think there's distinct promise in a link-by-query approach. One of the things that makes me think this is the increasing use of search on the Web and now on the desktop for locating data. Search will never be as good as explicit linking, but it may well be the "good enough" solution for many uses that the Web was to hypertext.

Before I finish, I should follow up with one more item of historical resonance: Len Bullard's invocation of architectural forms. It just goes to show that all good XML developers should know their SGML history.

The Last Word on XQuery

Last week's XML-Deviant covered most of the debate about the length of time XQuery was taking to develop, but I wanted to bring you some of the closing words in the thread. Hindsight may be 20:20, but people still seem to see different things.

Tim Bray:

Bah. If the WG had followed the advice they got three years ago to just abandon all the type-based elaborations and the tight coupling to XSD, then they would have shipped XQuery two years ago, and they would have discovered that the market apparently cares more about update facilities than it does about schema-driven querying, and they could be building that right now.

Michael Kay:

If the WG had followed all the advice they got three years ago then they would have produced nothing whatsoever.

Many of us could have produced a better query language much quicker, but only by ignoring the advice of others.

Births, Deaths, and Marriages

The latest announcements from the XML-DEV mailing list.

SAX for .NET 1.0 Released

Teaching a new dog old tricks. Includes a wrapper for Expat, with a port of Ælfred planned, too.

YesLogic Prince 4.0

While it sounds like a CD that a well-meaning nephew might buy you for Christmas, YesLogic Prince is a batch formatter for converting XML and CSS into PDF or PostScript. Evaluation versions available.

XMLmind XML Editor V2.8

XML editors need radical features to compete these days, it seems. XMLmind's new product incorporates a spreadsheet application, using the XML document as the spreadsheet.

XMLmind FO Converter 2.0 - WML Support

XMLmind's formatting engine adds support for Rich Text Format (RTF) or WordprocessingML (WML) output.

xframe - xsddoc-0.6-beta Released

An open source documentation tool for W3C XML Schema based on XSLT. With xsddoc you can generate documentation of your XML Schema in a JavaDoc-like visualisation.

Scrapings

If an element type falls in a forest and nobody has a name for it ... DuCharme's metadata fling helpfully augmenting mail headers ... 164 messages to XML-DEV last week, XLink quotient 20% ... XML Spy continues to attract ire in its interpretation of W3C XML Schema ... Quotes and reportage from a conference focusing on the Microsoft and web services view of XML ... Apparently Don Box kept his clothes on.