Linktext investigation

[Robert D Anderson <robander@us.ibm.com>]

As Michael requested, I did more in-depth investigation into the linktext
content model query.

It turns out there's a deeper issue apart from the differing domains. The
linktext element is one of two elements (along with shortdesc) that is
defined twice - once in the map module, once in the topic module. We've
discussed this before, and can't bring them in sync until 2.0, because they
use different class values (map/linktext vs topic/linktext) - if we change
that, we'll break either map or topic processors. This has been known since
before DITA 1.1.

Now, with 1.2, the content models have gotten out of sync. Proposal 12020
(easy reuse of small pieces of text) called for adding <ph> to <linktext>.
In the implementation, it was added to the topic version, but not to the
map version. That's the only difference at a base level - the remaining
content matches. This means simply adding the highlight domain will not
bring the two in line, because specializations of <ph> (such as <b>) still
will not be valid in linktext.

Adding <ph> in one spot but not the other seems like a clear oversight -
the question for the TC is whether to fix it at this point. I think the
options are:
1. Do nothing - 1.2 introduces a confusing quirk, but the package remains
stable
2. Add <ph> to <linktext> in maps - custom shells can then add domains so
that models match
3. Add <ph>, and add some domains
4. Add <ph>, and add all topic based domains
5. Add some domains, but don't update <linktext> - this would solve related
issues elsewhere but would do nothing to address the actual spec comment

Note that for domains, we actually ship 2 maps today - basemap and the
original map (now part of the Techncial Content package). I'm assuming that
any base domains we add will go into both, while any Tech Content domains
like programming would only go into the Technical Content map.

Robert D Anderson
IBM Authoring Tools Development
Chief Architect, DITA Open Toolkit