Cart
[Home] [By Thread] [By Date] [Recent Entries]
On Friday 28 January 2005 16:48, Michael Kay wrote: > > One of the worst is the fact that attributes can use > namespace prefixes that haven't yet been declared. This > means you need some kind of lookahead. > Thanks Michael, I had forgotton all about that nasty. I did get some results based on my initial assumptions about what needed to be checked. In short it appears as though about 50% of the benefit is by not performing a full set of checks and the rest is the quicker parsing in the cases I looked at. I should quantify this by saying I did not put any real amount of effort into optimising the checks so this is probably overstating the costs a bit in a production version. What is a little surprising is that the result you get is quite finely balanced on your choice of parser, encoding format and test documents. With carefull selection you could probably get just about any result you wanted, e.g. on one very text centric document I saw a difference of x25 when the checks were added. Kev.
|
