Found problem with learningObjectMap, learningGroMap DTDs

Hi,

Yesterday I discovered an unfortunate issue with the learningObjectMap and learningGroupMap DTDs while working on the contains / contained-by tables.

Each of these maps includes a constraint module, learningAggregationsTopicrefConstraint. Unfortunately this constraint is working in the RNG grammar files, not working in the DTD, and unclear in the XSD.

The purpose of the constraint is to say - "in any location that would normally allow <topicref> and its domain extensions, only allow keydef, mapref, and topicgroup". This means that the constraint removes the following elements:
* topicref
* From mapgroup domain: anchorref, topichead, topicset, topicsetref

I believe this works in the RNG. As a child of <learningObjectMap>, the legal topic references include the 3 expected, plus learningObject, learningGroup, learningObjectMapRef, and learningGroupMapRef. It also allows <ditavalref>, which is added as an independent domain. None of the excluded elements above are allowed.

The DTD constraint is not working - it allows everything that the RNG excludes:
* It is included in the wrong spot in the shell - so redefined entities are completely ignored
* The syntax in the constraint isn't quite right; moving it to the right spot in the shell results in a DTD that doesn't parse. It needs several changes in order to match what is done in RNG (drop topicref, add only 3 mapgroup extensions, and keep ditavalref).

I'd appreciate it if somebody could test the XSD - my test failed, not sure if it's a problem with my test or with the XSD itself.

Options moving forward:

Fix the DTD (and maybe the XSD) in the errata. This has the potential to break any documents created with 1.3 using the learning maps. Kris points out to me that the RNG is normative, and that it might be better to break the few docs that exist today versus allowing 5 years of broken docs.

Relax the RNG. But this does feel like more of a normative change.

Almost status quo: DTD allows stuff that RNG doesn't (but fix the constraint module so that it *could* be used properly). This also raises the question "what should the containment models say in the appendix".

Thanks,
Robert D. Anderson 
DITA-OT lead and Co-editor DITA 1.3 specification,
Digital Services Group 

E-mail: 
Digital Services Group 

11501 BURNET RD,, TX, 78758-3400, AUSTIN, USA