A general question about performance issues with XSLT. If I have two input documents and I am trying to match nodes using regular expressions, would I achieve better performance making a call to Java? I know this is quite vague, I am just curious if there is a known heavy cost for evaluating and using regexs in XSLT (Saxon8).

Saxon uses the underlying regex engine in Java so I wouldn't expect it to make much difference. There is an overhead, because the XPath regex dialect isn't quite the same as the Java dialect and some constructs have to be translated, but that's usually a one-off cost at compile time. There is also an overhead in making calls from XSLT to Java, which can sometimes exceed the cost of actually doing the work.

As always with performance questions, the only way to find out is to measure it. Do remember that to get repeatable results from Java performance measurements you need to run the VM for at least a minute.

Michael Kay
http://www.saxonica.com/
Author, XSLT Programmer's Reference and XPath 2.0 Programmer's Reference

Is there a way to keep hold of previously compiled regular expressions, it seems that you always need to recompile them. Using java, we compile once only and then work with the compiled objects.

You just have to trust the optimizer on this. With Saxon, provided the regular expression is written as a string literal in a place where a regex is expected, it will be compiled once at compile time.

Michael Kay
http://www.saxonica.com/
Author, XSLT Programmer's Reference and XPath 2.0 Programmer's Reference