[tei-council] numbered divs, a proposed solution

Mon Jan 8 12:48:10 EST 2007

I may not get to give this issue full attention for a little while,
but some brief thoughts follow.

> > You can hardly have failed to notice the argument on TEI-L
> > which revolved around Ron vd B's innocuous expectation
> > that if he  deleted mumbered divs in his ODD, he would
> > get a valid DTD.
> 
> I agree that this is an innocuous expectation and that he should be
> able to delete numbered divs.

I agree, too, although as I think I've said before, this could be
documented around, rather than really solved. 

> When we discussed and decided to drop support for SGML (around
> 2006-09-14) it was mentioned that the support for DTDs was also
> difficult. I'm assuming this is an example of that. Does this
> problem exist in the generated schemas, or only the DTDs?

Exactly correct. The problem exists in DTDs and probably in W3C
Schemata, although I don't know that anyone has tested the latter.
The problem does *not* exist in Relax NG.

> > I know one should not solve a general problem by dealing with a
> > special case case, but I also think that there is a time for
> > compromise and practicality, so I have a firm proposal which I'd
> > like you to consider. I claim it will solve this problem at a
> > relatively small cost, compared to the public embarassment this
> > will keep on causing.
> 
> I agree that the problem needs to be solved one way or another.

I have not looked at Sebastian's proposal here closely enough to give
a real opinion on it, so I am *not* saying here that I prefer the
brute-force or hack methods better. But I am curious: does the above
statement mean that you have looked at the FAND ODDs I posted on the
wiki and don't think they are viable solutions?[1]

> > The "div-like" would be expressed as a new class model.divLike,
> > and its membership would be div, div0, div1 and divGen.
> 
> I suppose we couldn't re-open the issue of whether we should just
> get rid of the option of div0|div1 and just pick one or the other?

We can, and perhaps should, but it doesn't make any significant
difference. I.e., the non-deterministic content model problem will
exist whenever all numbered divs are deleted in your ODD, whether
div0 is among them or not.

> These are both quite bad in my mind. 1) The whole point of divs vs
> numbered divs is that they are mutually exclusive isn't it? I mean
> I'm not supposed to be able to use both in the same document?

Not within the same <front>, <body>, or <back>.

> So, is the real issue behind this whether we want to rely on
> Schematron for these tests? And to do so for a special case
> solution? Does adopting Schematron for this have any negative
> implications?

These are exactly the important questions to address up front, I
think. 

> > I take it as axiomatic that we cannot simply leave the current
> > setup as it is. We must do _something_ to allow the simple case
> > of "delete numbered divs" to work.
> 
> I agree that something must be done. Does any one have any
> suggested alternatives?

While I may be convinced that something must be done, I am not
convinced yet.

Notes
-----
[1] http://www.tei-c.org/wiki/index.php/FAND1_hack and
    http://www.tei-c.org/wiki/index.php/FAND2_replace