Re: [xml-dev] XML spec and XSD

[Rick Jelliffe <rjelliffe@allette.com.au>]

Mukul Gandhi wrote:
> It seems, that most people argue that a conformant XML document has
> ability to refer to user defined entities, and this can be specified
> only in a DTD. I think, the current XML spec, makes DTDs an
> inseparable part of XML spec is, because of the DTDs ability to
> specify user defined entities, which (the entities) are part of say
> the, XML 1.0 (and 1.1) spec.
>   
And because of attribute defaults and non-significant whitespace 
removal.  And perhaps for ID checking and indexing.
> To change the XML spec, to remove the concept of valid/validity and
> being able to remove DTD s from the XML spec, I think will also
> require us to remove the concepts of entities from the XML spec (I
> think, something like a future XML 2.0 spec should only specify the
> built in entities like & amp; etc and not allow users to specify user
> defined entities, unless they use an external thing like DTD or a
> future XSD (a 1.2 version, which may allow users to specify entities,
> via some global declaration perhaps))
Well, entities for special characters and entities for document or 
schema construction  are  perhaps different
animals.   I have long argued that XML should have all the ISO/Math/W3C 
special characters just built in. There was some argument that this 
would make an XML processor much larger which would be a difficulty for 
mobile devices: it seemed silly to me then.  (The fetishism for an 
absolutely minimal XML is, to me, a classic case of a local optimization 
that would cause a global sub-optimality: when you remove a feature from 
a layer that people need, it just goes to a different layer and you 
don't actually get a total simplification--indeed you may get more 
complexity to support the extra layers especially if they are 
uncoordinated. The W3C answer that the way to handle all the 
uncoordinated layers is to make the part of XProc (i.e. a different file 
to be downloaded in order to know the simple functionality of standard 
markup) is insultingly dumb: whatever XProc is good for, it is not going 
to be an adequate packaging mechanism.)

If you built the special characters in, and you built in XInclude, 
XBase,  xml:id, and some method of saying "this is element content so 
ignore the whitespace" then you could indeed move DTDs to a different 
layer.  Note that at ISO there already is a spec (as part of DSDL) for 
namespace-aware DTDs; if that was useful for anything. 

Also of possible interest is that you could have a DOCTYPE declaration 
use some non-DTD system. So, for example, you could restrict the 
internal subset of the prolog to entity declarations (if you wanted it 
all) and have the external subset being XSD or RELAX NG or Schematron etc.

Cheers
Rick Jelliffe

B.t.w. There is absolutely no problem with this and SGML compliance: it 
would just be a minor change to IS8879 to look after the formalities for 
"tag-valid" documents (in fact, IIRC it was Charles Goldfarb who asked a 
decade ago whether it would be useful for giving the W3C/XML/HTML people 
more freedom in this area, but I think the feeling was that we may as 
well wait for a real use case rather than second guess anyone.)  It 
isn't SGML that holds back innovation, though the belief that SGML can 
only be adhered to or dropped completely can hold back innovation.