The rules in section 10.2.1 for binding attribute set names appearing in an [xsl:]use-attribute-sets attribute need to be revised to mention what happens when there are multiple packages, overriding attribute sets, etc.

It would be useful to rephrase the rules to distinguish what happens statically from what happens at execution time.

Specifically:

(1) At package compile time, the use-attribute-sets attribute is tokenized and each QName must match the name of an attribute set declared in the package.

(2) I think the specification would be easier if we spoke in terms of statically merging several same-name attribute set declarations appearing within the same package into a single attribute set component. The merging isn't trivial because if we have

a-set use="X" attributes="Y", "Z"
a-set use="P" attributes="R", "S"

then the result of the merge is

a-set use="X" attributes="Y", "Z" use="P" attributes="R", "S"

where the position of the "use" constructs relative to the "attributes" constructs must be preserved.

(3) Dynamically, there are then two operations: (a) binding the attribute-set names appearing in use-attribute-sets, which follows the general component binding rules in 3.6.3.4, and (b) evaluating the attribute set component, which involves evaluating its constituent xsl:attribute and use-attribute-sets "instructions" in order.