xml-dev - Re: XML Schemas: ref'ing versus in-lining

[Rick JELLIFFE <ricko@geotempo.com>]

Curt Arnold wrote:
 
> Having to detect when multiple declarations are exactly (or
> sufficiently) identical seems to be unnecessary complexity.

I don't think this is really so complex. Forgetting any impact of
abstract elements: 
 * a complex type declaration has its own symbol space for elements
 * an element declaration (in a complex type declaration) not only
defines the existance 
   of the element, it also binds the name to a type (directly or by
reference) 
 * an subsequent uses of the element must also bind the name to the same
type (directly
   or by reference) 

A good question is whether a subsequent element can have a derived type
of the first
element's type?  Henry?

The reason why this is good is that *how* something is declared (where
there are several 
alternative syntaxes) should not affect what is declared. Of course,
there may be 
other considerations (implementation difficulties would be the most
compelling) that
could alter the tradeoffs, and the WG is very interested in being told
them.

RELAX takes, I think, a sugar-free approach of making you declare types
before/separately 
to binding them to element names. But this results in very shallow trees
and therefore 
more cross-referencing: it may be more difficult to read and
unacceptable (remembering
that RELAX does not have type-derivation so its schemas will tend to be
smaller: 
XML Schemas schemas may allow much more detailed description of
datatypes and
consequently be larger, so reducing page-flipping may be good for humans
at least).  
When the next draft comes out, I hope people will think about this to
see if they 
would prefer RELAX-style or allow inline declarations of type as we
currently have.  
I tend to think it is matter of taste with no clear reasons to favour
one
way or another (which is why implementation feedback is really
important: 
concrete annecdotes are better than taste).

Rick Jelliffe