> *[string-length(normalize-space(.))>0]

> No need to work out the length (which can be surprisingly 
> costly for general unicode strings) you can just do
> 
>  *[normalize-space(.)]
> 
> as the predicate is true if teh string is non empty. 
> (Although I have a feeling saxon for example does that 
> rewrite for you anyway, so it may make no difference.)

Indeed. Saxon will neither construct a string with the whitespace removed,
nor count the characters that remain after removing them: with either
expression it simply scans the string (or rather, its UTF16 representation)
and returns false as soon as it hits a non-whitespace codepoint.

Michael Kay
http://www.saxonica.com/