The new fn:analyze-string() function behaves like most string functions in that it accepts an empty sequence as input, and treats it in the same way as a zero-length string (in both cases the output will be an empty fn:analyze-string-result element).

The xsl:analyze-string instruction in XSLT 2.0 does not work this way: it reports an error if the input is an empty sequence. So, for example, if there is no address element then <xsl:analyze-string select="address"> will cause a type error, but if the address element is present and empty, it will produce empty output.

I propose we change this (a) for usability, (b) for consistency, and (c) to make it a little bit easier for implementations to reuse code between xsl:analyze-string and fn:analyze-string.