- From: Steve DeRose <Steven_DeRose@brown.edu>
- Date: Tue, 29 Feb 2000 09:20:04 -0800 (PST)
- To: <www-dom@w3.org>
In XML the order of nodes is defined to be significant. Even if you construct cases where it is not "conceptually", such as an RDB export into XML syntax, the order is still significant by definition; just as the order or records in a relation is by definition *not* significant, even if there is some serial-number or other field floating around that implies an order to apps/people that know. Given that XML is ordered, it does seem pretty strange to not be able to determine what order any given two things are in. Put another way, the lack of such functionality contradicts the data model. I would say that is *why* operations are being asked for that turn out to be very expensive to simulate. Just like when you simulate XML with an RDB: traversing nodes just to do an identity export of a document in the right order, requires extra fields or relations plus a whole lot of sorting -- one big sort if your RDB models XML via generic nodes, a gazillion little sub-selects and sorts if it instead uses a separate relation for each element type. It isn't surprising in either case that you can simulate the other data model -- but it costs. The data model and data architecture mismatch, and that generally leads to high costs down the line. Much better to add the facility and make the interface match the data model better. Calculate O() for the common operations needed to keep all the intermediate results in XPath (and therefore also XPointer) in order, it will add up real fast. Steven_DeRose@Brown.edu; http://www.stg.brown.edu/~sjd Chief Scientist, Scholarly Technology Group, and Adjunct Associate Professor, Brown University North American Editor, the Text Encoding Initiative
Received on Tuesday, 29 February 2000 12:20:12 UTC