My guess is (and I wasn't there) is that it added 
complexity and duplicated features in the tree. SGML 
is complicated enough without adding yet another 
feature that isn't useful in a large number of cases. 

However, turning this on the head a bit, if XML 
was to be a simplified SGML, why would we have added 
features that are not very useful in a large number 
of cases?  

Keep in mind, SGML had to be modified 
slightly to make XML a proper subset so it is 
conceivable that something along this line could 
have been added.  Again was it 'very useful in a large 
number of cases' where those cases did not include 
all of the things to which XML has been applied 
in the aftermath that the right people considered 
important then.

Only a few wild and crazy people thought XML would 
become the lingua franca for all bits on the wire. 

len


From: ian.graham@u... [mailto:ian.graham@u...]

Quoting "DuCharme, Bob (LNG-CHO)" <bob.ducharme@l...>:

> I think the general idea is that if the ordering of pieces of
> information associated with an element matter to an application, then
> they should be declared in the content model as child elements, where
> you have various regular-expression-like options for describing ordering
> options. 

I agree with this interpretation, but it does not explain _why_ attribute
order
is explicitly not relevant.