Following my earlier post on semantic documents, I’ve given the subject some thought. In fact, I wrote a paper on a related subject and submitted it to XML Prague for next year’s conference. The paper wasn’t accepted (in all fairness, the paper was off-topic for the themes for the event), but I think the concept is both important and useful.

Briefly, the paper is about profiling XML content. The basics are well known and very frequently used: you profile a node by placing a condition on it. That condition, expressed using an attribute, is then compared to a publishing context defined using a similar condition on the root. If met, the node is included; if not, the node is discarded.

The matching is done with a simple string comparison but the mechanism can be made a lot more advance by, say, imposing Boolean logic on the condition. You need to match something like A AND B AND NOT(C), or the node is discarded. Etc.

The problem is that in the real world, the conditions, the string values, usually represent actual product names or variants, or perhaps an intended reader category. They can be used not only for string matching but for including content inline by using the condition attribute contents as variable text: a product variant, expressed as a string in an attribute in an EMPTY element, can easily be expanded in the resulting publication to provide specific content to personalise the document.

Which is fine and well, until the product variant label or the product itself is changed and the documents need to be updated to reflect this. All kinds of annoyances result, from having to convert values in legacy documents to not being able to do so (because the change is not compatible with the existing documents). Think about it:

If you have a condition “A” and a number of legacy documents using that condition, and need to update the name of the product variant to “B”, you need to update those existing documents accordingly, changing “A” to “B” everywhere. Problem is, someone owning the old product variant “A” now needs to accept documentation for a renamed product “B”. It’s done all the time but still causes confusion.

Or worse, if the change to “B” affects functionality and not just the name itself, you’ll have to add “B” to the list of conditions instead of renaming “A”, which in turn means that even if most of the existing documentation could be reused for both “A” and “B”, it can’t because there is no way to know. You’ll have to add “B” whenever you need to include a node, old or new.

This, in my considered opinion, happens because of the following:

• The name, the condition, is used directly, both as a condition and as a value.
• Conditions are not version handled. If “B” is a new version of “A”, then say so.

My solution? Use an abstraction layer. Define a semantic profile, a basic meaning for the condition, and version handle that profile, updating it when there is a change to the condition. The change could be a simple name change for the corresponding product but it could just as well be a change to the product’s functionality. Doesn’t really matter. A significant change will always requires a new version. Then, represent that semantic profile with a value used when publishing.

Since I like URNs, I think URNs are a terrific way to go. It’s easy to define a suitable URN schema that includes versioning and use the URN string as the condition when filtering, but the URN’s corresponding value as expanded content. In the paper, I suggest some simple ways to do this, including an out-of-line profiling mechanism that is pretty much what the XLink spec included years ago.

Using abstraction layers in profiling is hardly a new approach, then, but it’s not being used, not to my knowledge, and I think it should. I fully intend to.