Re: namespace viz validation [was RE: DSIGS]

Martin Bryan <mtbryan@sgml.u-net.com> writes:

> The problem is how to ensure uniqueness of ids when you are cutting and
> pasting things from different documents produced with different DTDs. 

Surely this is a similar problem to that addressed by Formal Public Identifiers:
how to get globally unique names.  The FPI solution is of course to prefix the
name with organisation names and so on.   If XML adopted a separator delimiter
by convention (e.g. '::', where ':' is allowed as a NMCHAR) and gave some 
convention for use, like  company::dtd::name  (e.g.  ms::cdf::amount )  you can 
cut and paste between XML documents with different namespaces today.

Which, in turn, suggests that namespaces are, in part, really about identifier
minimisation rather than type. 

XML is just starting.  If terseness is not our goal, and people want to be able to
make their documents cut-and-pastable, then why not add these conventions
now, so that new DTDs have them?

And if a vendor does not follow this convention, and just uses simple names,
then that makes their documents less easy to cut-and-paste, which they
might like.  If cutting and pasting is important, then export your data in a
unique-ized form. 

> AsBernhard Weichel pointed out:
>>   Reusing instance fragments should be possible with out renaming all
> >  elements, so a GI mangling approach isnīt helpful.

All what elements?  In all of which documents?  XML is just starting!

SGML documents will have to be  converted to fit the sole tag and so on.  
SGML DTDs will have to be converted for XML anyway. Now is exactly the 
time when we can make GIs (and other identifiers)  more powerful.  

If we don't define some nice convention that works now, then we'll have to 
retrofit some new big fat namespace system with new declarations etc etc, 
all to achieve the unique naming we could have had from the start.  XML 2.0
can have the algorithm for minimising names, XML 1.0 just needs to set the
data convention for making sure that names are unique in XML 1.0 documents,
and therefore compatible with XML 2.0 fullname minimisation.


Rick Jelliffe

--------------------------------------------------------
Back on casting. etc, I point out that in C casting does 2 things: type coersion
(e.g. make an int a pointer) which is regarded as poor, and type narrowing
(casting to a typedef) which is regarded as good.  There is no hint of type 
coersion in my use of 'cast'.

 I suggest that the ISO 8879:2001 SGML declaration should allow: 

CONCUR	NO			-- e.g. <para> --
CONCUR	NAMESPACE 		-- e.g. <martin::para> --
CONCUR	YES			-- e.g. <(rick)p>--
CONCUR	ARCHITECTURE   	-- e.g. <(martin::para)p --

where the last is a kind of architectural form.

I am sure there is something nice hidden somewhere in CONCUR.  

Received on Thursday, 19 June 1997 11:57:05 UTC