In article <d566bp$2rv4$1@n...>,
Alain Frisch <frisch@c...> wrote:

>3 If either O1 or O2 is a pair of not and a value (a namespace name or 
>·absent·) and the other is a set of (namespace names or ·absent·), then 
>that set, minus the negated value if it was in the set, minus ·absent· if 
>it was in the set, must be the value.

>I don't understand the rationale behind "minus ·absent· if it was in the 
>set". This means that ·absent· is removed whenever it appears in the set. 
>It would be natural to remove it iff the second component of the "not" 
>pair is ·absent·. 

>Can someone explain the rationale behind the spec ?

The case described is the intersection of ##other and and set of
namespaces.  In this context, "absent" means "no-namespace".

When the target namespace is absent, ##other means any non-absent
namespace.  When the target namespace is not absent, ##other means any
*non-absent* namespace other than the target namespace.  I think the
idea behind this is that ##other is intended to allow extensibility by
having attributes from other namespaces, not by having attributes in
no namespace.

So in the latter case, the intersection should exclude both the target
namespace and absent.

Now for some reason that is not clear to me [*], the namespace
constraint ##other is not represented as a pair of "not" and (a set of
the target namespace and absent), but merely as a pair of "not" and
the target namespace.  So to make the intersection give the right
answer, the spec has to give the peculiar definition quoted.
Similarly, it has to give a peculiar definition for "Wildcard allows
Namespace Name" a few paragraphs earlier.

[*] It's probably one of those "historical" reasons.

-- Richard