Re: Infoset consistency between versions

Paul Cotton wrote:

> BTW, another interesting question is when should a revised specification
> introduce a new namespace?  From my view if a specification is being
> revised and its namespace does not change, then I would assume that
> there or no "cascading impacts" in the revision and the only changes
> being included are editorial errata.

This is a difficult and important issue, but I question whether it's 
possible to write down a set of rules you can follow mechanically.  For 
example, suppose there's an XHTML 2, and 3, and 4, and they all have the 
  "h1" element.  Some application classes are going to want to know not 
only whether it's XHTML 1 or 2, but whether it's 1.03 or 1.02.   Others 
(renderers, crawlers, indexers, probably the vast majority) have the 
semantics of XHTML "h1" wired-in and just want to know whether this is 
an HTML "h1" or not.  So there are good reasons both to rev and not to 
rev the namespace.

So I think the community of language designers is facing tough design 
choices, and probably all the TAG can do is write a principle saying 
that they have to face up to them and think about them seriously and 
justify their namespace-revving decisions.

Anyone who's worked seriously in content-management systems knows that 
versioning is in general a horrible rats'-nest of issues with semantics 
all over the map and little commonality of practice. -Tim

Received on Tuesday, 9 July 2002 14:44:58 UTC