xml-dev - Re: Arbitrary Infoset boundaries (was Re: Common XML -FinalReviewDraft)

[Rick JELLIFFE <ricko@geotempo.com>]

Aaron Skonnard wrote:
> 
> Rick Jelliffe wrote:
 
> > XML Schemas may even lead to a resurgence of the use of DTDs: this is
> > because once schema software is in place as a layer, the particular
> > syntax becomes less important. (Of course, where there are features that
> > cannot be supported, that is different. No magic.)
> 
> What? How? You lost me...
 
The more one kind of schema is supported or used, the more all kinds of
schemas will be supported and used.  It is not unthinkable.

Someone will make DTD->XSDL and XSDL->DTD lossy conversion utilities. If
the XSDL schema is too verbose, they may distribute the DTD.  XML
systems that incorporate XSDL software may find it trivial to also
include DTD->XSLD import preprocessors, so DTDs will become an
alternative syntax for simple XSLD.  Some vendors might decide to stick
with their proprietary schema specs, and some might need to use
something other than XSLD due to some particular niche requirement: DTDs
represent the common subset of features that might be the most that the
parts of a multi-schema-language system could interoperate with.

If the XML Schema WG or XML-DEV or others do not come up with some
convenient notation for prose communication of content models, then the
regular expression will continue to be the dominant way for human
communication: using <!ELEMENT x (y,z)> is more satisfactory than BNF  x
::= y, z   because it avoids the handwaving of "lets abstract all the
other XML rules away".  So it may even be that learning DTDs (not
parameter entities) becomes the preferred learning path for XML Schemas!

XSLD will allow more complicated document systems. These will in turn
make use of more complicated documument construction. This will require
entities and linking. These will require DTDs with entity declarations
or attribute value defaulting. 

There may be some size-sensitive devices which need identification of
ID/IDREFs but for which full XSLD is too big.  Lots of possibilities.  

It is not a prediction. But there are lots of reasons why DTDs may
continue as a nice tool in our amoury.  (I don't see markup declarations
as primarily schemas per se: I see them as a mechanism for moving
invariants to headers.  Schema-ish information is one such invariant.)

Rick Jelliffe