Re: a or b or both - mystery..

[Marcus Carr <mrc@allette.com.au>]

Murali Mani wrote:

> Also, if you look at TREX's goal, it says non-deterministic content
> models are almost necessary and definitely convenient for several use
> cases.

There's only one structure that I know of where deterministic models are unable to
express the desire of the author, but a case can certainly be made that
non-deterministic models might be more convenient in some cases.

> I should say this -- I believe *very* strongly in non-deterministic
> content models, and I think that is the way document processing can be
> supported neatly. So I am *very* biased, and I use RELAX for my purposes -
> and not XML Schema. Actually I think I cannot afford to use XML Schema.

Can you provide some examples where you have run into problems? Granted, the "and"
operator removed most of these issues in SGML, but I've never really felt that
there was much to pick between deterministic and non-deterministic models. Problems
don't occur frequently enough to make me feel strongly either way, though I don't
have much experience with TREX or other XML schemas.

> If you want to know more about deterministic content models, you should
> read Appendix E of XML 1.0, it is very good. Also, I think I am right when
> I say that Anne Bruggemann Klein is the first person to analyze these
> deterministic content models as in DTDs...

Perhaps, though Sam Willmott of OmniMark released a whitepaper somewhere round 1990
or 1991 dealing with content model algebra, not that it's really important who was
first...

> but she also opines that
> deterministic content models are *very* restrictive and is not the way it
> should be.

I didn't get that impression from her conclusions - are you referring to the paper
located at the other end of
ftp://ftp.informatik.uni-freiburg.de/documents/papers/brueggem/habil.ps? On the
contrary, it appears that she contributed her "decision algorithm for unambiguous
expressions" to James Clark for his SGML parser. Ambiguity in SGML was occasionally
inconvenient, but never more than that. My understanding was that the requirement
for deterministic models was a concession made to those who were writing parsers at
the expense of those who were creating DTDs. It was no big deal though - you
learned to spot where problems might occur and avoided them.

> I also heard this: XML Schema WG has discussed deterministic content
> models a lot, and they decided to use deterministic content models largely
> for parser simplicity -- which I think is a non-valid reason at this
> point.

I suppose the working group felt that if this aspect of a processor was going to be
made more complicated in XML than it was in SGML, it would discourage the writing
of processors. Given that there were (arguably) only ever two nearly complete SGML
(James Clark's and OmniMark) parsers, I tend to agree with them.


--
Regards,

Marcus Carr                      email:  mrc@allette.com.au
___________________________________________________________________
Allette Systems (Australia)      www:    http://www.allette.com.au
___________________________________________________________________
"Everything should be made as simple as possible, but not simpler."
       - Einstein