Re: [xsl] optimization of complex XPath

[Michael Kay <mike@xxxxxxxxxxxx>]

On 19/11/2010 09:24, Wolfgang Laun wrote:
> What's the fastest and memory-cheapest way to extract these links and
> targets to two different (or even a single) plain text files?
> Post-processing with utilities such as grep, sort -u and comm wouldn't
> be "complex".
> -W

The streaming facilities in Saxon-EE 9.3 (or, in fact, in previous 
releases) make it reasonably easy to extract the links and targets to a 
single file in one streaming pass over the input, or to two files in two 
passes over the input.

Producing two output files from a single pass of the input is a problem 
well-known to the working group (we call it the "coloured widgets 
problem", after a use-case put forward by Oliver Becker) and the current 
solution in the spec is the xsl:fork instruction, not yet implemented in 
Saxon. The only way to do it in Saxon would be with side-effectful 
extension functions, or some kind of clever use of xsl:message with a 
user-written MessageListener.

The streamed merging/comparison of two sorted inputs is handled in the 
spec using the xsl:merge instruction - that's not implemented in Saxon 
9.3 but we're working on it right now.

Michael Kay
Saxonica