Re: [xml-dev] Rick Jelliffe's article on XSLT 1.0 performance

[Michael Kay <mike@saxonica.com>]

> On 15 Feb 2017, at 02:40, Rick Jelliffe <rjelliffe@allette.com.au> wrote:
> 
> Does anyone have any benchmarks on how much @as can improve performance, or any tips on which particular cases it helps.  I had kinda written it off as one of these theoretical improvements that never really pays off enough to be much use.
> 

I don't have any numbers I can share, and the effect is going to be very variable across stylesheets.

The main benefit of using @as attributes isn't performance, it's maintainability - your code becomes more readable and errors are reported earlier, often at compile time.

The performance effects will usually be fairly marginal - the system doesn't have to do so much run-time checking, and it can make earlier decisions on things like what arithemetic or comparison operator to use. Occasionally, though, the availability of type information opens the door to a much bigger optimization. To take an example, if you have an xsl:choose with conditions like test="$x = 3", then if you know that $x is a single atomic value, you also know that the conditions $x=3 and $x=4 are mutually exclusive, so you can implement the xsl:choose with a hash lookup. If $x can be multi-valued then that isn't the case. Another example: given $x/price, if you know that $x is a single node then you know that you don't have to sort the result into document order and eliminate duplicates.

Michael Kay
Saxonica