But the way they are added is far from being perfect. The string regular
expressions of XSLT 2.0 is the only place where regular syntax is used,
XML matching is not expressed in terms of regular expresions.

Further on, the regexps in XSLT 2.0 are a mix of those from XML Schema
and perl 5, fit into the XML syntax XSLT 2.0, but without capabilities
of perl to compose and manipulate them.

http://lists.xml.org/archives/xml-dev/200401/msg01040.html

While in perl one can generate, compute, concatenate and split regular
expressions, as well as manipulate their results freely, in XSLT 2.0
the operations are limited to a few built-in commands.

I understand that it is a difficult task to design a regular expression
language to match attribute values and character data; but when there is
no good solution inside XSLT, it should be left out of it -- tool chains
can be long; otherwise the glue must be powerful and flexible. XSLT does
not provide a powerful and flexible glue; it is still XML-oriented.

In RenderX XEP, an XSL FO formatter, the preprocessor is written in XSLT;
it is a complex stylesheet of moderate size, and it does what is best to
do in XML manipulation language.

Regular parsing and type manipulations are done in a different language, because a different language is more suitable.

Instead of providing means for interaction, XSLT 2.0 is going to provide
inferior solutions for problems that can better be handled in other parts
of toolkits.