> > I think the simplest solution will be to say that <key> and <keyRef>
don't
> > match empty strings even if the included pattern can match an empty
string,
>
> "" (the string of length 0) is not the only one. To be precise, "empty
> string" is a string that consists of #x20,#xD,#xA.
>
> Because patterns like
>
> <element name="foo">
> <choice>
> <empty/>
> <key>
> <data type="string">
> </key>
> </choice>
> </element>
>
> and an instance <foo> &xD; </foo> is still ambiguous.
Right, but we don't have to deal with that in the inference rule for key and
keyRef because of the stripSpace() in 6.2.8.
I believe it is sufficient to add
not(s = "")
as an antecedent to (key) and (keyRef) rules in 6.2.12.
James
