Rick Jelliffe's article on XSLT 1.0 performance

[Michael Kay <mike@saxonica.com>]

Rick Jelliffe has written an interesting article on XSLT 1.0 comparative performance at

https://www.xml.com/articles/2017/01/26/revving-xslt-10-engines-are-they-all-same/

By and large I trust the results because they are in line with previous such exercises including the measurements we made for our XML London paper in 2014.

However, I think there is one important caveat missing from the article, namely that what is being measured is the sum of (stylesheet compilation, tree building, transformation, and serialization) costs, and in many real workloads, it's unrealistic to simply add these numbers up.

Our own measurements showed Saxon being relatively strong on transformation time and relatively weak on compilation time, so we don't show ourselves to best advantage in a benchmark where the stylesheet is being compiled every time it is run. Of course in some workloads that is exactly what happens so we are starting to put more effort into compilation time.

Michael Kay
Saxonica