Re: Annotations draft updated

[James Clark <jjc@jclark.com>]

> In some cases, users simply give
> up validation and  merely use default values.

It's very true that users may want information about default values without 
*RELAX NG* validation, and the spec should definitely mention annotation 
processing without RELAX NG validation as a possible implementation 
scenario.  This is another justification for keeping infoset modification 
cleanly separated from validation.

> Thus, although level "2"
> appears to built on top of level "1", this is not always the case.

Level 1 is really mostly about checking schema correctness: for instance, 
checking that the schema unambiguously defines a default value for a 
particular attribute/element name pair.  Such checking is a precondition of 
Level 2 and is needed irrespective of whether you do RELAX NG validation. 
The description of Level 1 in the spec needs some work.

For ID/IDREF, there is some validation involved in Level 2, but this is 
just validation of the ID/IDREF.  I think if my application is using 
information about IDs, then I would want, for instance, validation that 
there are no duplicate IDs.

> Do we
> really have to define conformance?

It seems to me as important to define conformance for this spec as it is to 
define it for the RELAX NG spec itself.  Applications may rely critically 
on the infoset modifications, just as much as they rely on validation.

James