> This is already true for DTD-declared IDs: they have type ID even if
> they are not valid.

Okay, this makes sense...

> Also, in plain XML, for a given element name, all attributes with the same
> name have the same type.  Some APIs don't have a way to return different
> types for a given attribute depending on its value.

I am not sure I understand the implication?

> I expect they will continue to do what they do now with invalid
> DTD-declared IDs.

Right, sounds good.

>>  <?xml version="1.0"?>
>>  <!DOCTYPE root [
>>    <!ELEMENT root EMPTY>
>>    <!ATTLIST root
>>      id ID #IMPLIED>
>>  ]>
>>  <root id="ID1" xml:id="ID2"/>
> 
> Not an xml:id error.  It will of course give you a validity error if
> you validate it.

Just so I am clear, the validity error that should be raised is 
undeclared attribute right? Or should you get two validity errors-- 
undeclared attribute and One Id per Element Type? I think the former 
(only undeclared attribute) is correct and not the latter-- as the 
xml:id isn't declared and that is where the VC is applied.

This brings up one additional question which I didn't think of before, 
though I am pretty sure I know the answer:

Does VC: ID Attribute default have any relevance here? I think not-- 
again xml:id has no relevance unless the xml:id is declared in the DTD-- 
in which case a document where an xml:id was declared with a default 
value would produce a validity error, nothing more. Right?

> You're right.  My guess is that the only reason it's not mentioned is
> that the editor was only thinking about conformance to the basic
> infoset when he wrote section 7.

I will send this as a comment to the public comments list so that the WG 
can address it (either as you have, or by adding the additional infoset 
item).

Thanks,
Jeff Rafter