I think the original intention of that statement was to ensure
that the two target elements were siblings, so that you couldn’t conref
to <li>s in two different lists or conref to unbalanced ranges. Michael
brings up a good point though, that validation of the innards of the range must
be ensured. Checking the parent element of both the source and the targets could
provide this validation.
Another option would be to restrict all the element siblings in
the range to be all of the same type e.g. all <li>s or all <step>s
etc. This would prevent users from creating a conref to the following types of
ranges (from x -> y):
<b id=”x”>Some bold text</b><mySpecialisedElement>Some
non-bold </mySpecialisedElement> text<b id=”y”>More
bold</b>
This is probably be too restrictive to be useful though.
Michael.
From: Michael Priestley
[mailto:mpriestl@ca.ibm.com]
Sent: September 27, 2007 11:00
To: Grosso, Paul
Cc: dita@lists.oasis-open.org
Subject: RE: [dita] Groups - DITA Proposed Feature #12013 Referencing a
range of elements (conrefrange.html) uploaded
> 1) "The user must only create
ranges on target elements that
> share the same
> parent element. " -- Does this mean that a source range
> inside <body> can
> only be to a target inside <body>?
One reason for
it to mean that would be to avoid situations like allowing conref of a range of
elements into a context where only one of them is allowed, or where the range
includes elements disallowed in the referencing context. If the parent elements
are the same, then the rules of what's allowed in the range are the same.
With
generalization-on-the-fly, the parent element of the range could be a
specialization of the parent element of the reference, but it does seem to me
that you'd get an important validity guarantee by comparing the parents in the
range case. Otherwise conref-with-range would open up a world of referencing
invalid content that conref currently prevents. In other words, without a check
on the parent element, we would be dramatically loosening the constraints on
what conref allows, in just this one place.
Speaking of
which, I think Yas also mentioned:
>It is already possible to include
invalid content
Is this true?
It's definitely not the intent of conref. I thought it was doing a pretty good
job of avoiding validation errors. What case were you thinking of?
Michael
Priestley
Lead IBM DITA Architect
mpriestl@ca.ibm.com
http://dita.xml.org/blog/25
|
"Grosso,
Paul" <pgrosso@ptc.com>
09/27/2007
01:47 PM
|
|
To
|
<dita@lists.oasis-open.org>
|
|
cc
|
|
|
Subject
|
RE:
[dita] Groups - DITA Proposed Feature #12013 Referencing a range of
elements (conrefrange.html) uploaded
|
|
fwiw, a couple comments from me on Robert's
email below.
>
Original Message-----
> From: Robert D Anderson [mailto:robander@us.ibm.com]
> Sent: Thursday, 2007 September 27 11:34
> To: Yas Etessam
> Cc: dita@lists.oasis-open.org
> Subject: RE: [dita] Groups - DITA Proposed Feature #12013
> Referencing a range of elements (conrefrange.html) uploaded
>
> Hi Yas,
>
> I've just got a few comments about the proposal before we
> have a meeting on
> it. Aside from what's below, I have an initial positive impression of
> Paul's suggestion to work purely with singletons, though I
> probably have
> not thought it through completely yet.
>
> 1) "The user must only create ranges on target elements that
> share the same
> parent element. " -- Does this mean that a source range
> inside <body> can
> only be to a target inside <body>? Or does it just mean that
> the target
> start/end must be siblings and thus share the same parent
> node? I think it
> means the latter, but I know the former was discussed a while back.
I think it means the latter, and this is irrelevant if we
go with the singleton model (which is one big advantage of
the singleton model).
>
> 2) The "Processing behaviors" section seems to state that
the domains
> attribute on the source and target topics must be an exact
> match. That is
> more restrictive than the current restriction on conref, in which
> compatibility is guaranteed when "the list of domains in the
> referenced
> topic instance (declared on the domains attribute) is the same as or a
> subset of the list of domains in the referencing document"
> [1]. That is,
> when the source topic has domains="(topic hi-d) (topic
> sw-d)", target topic
> values of "(topic hi-d)" and "(topic sw-d) (topic
hi-d)"
> should both be
> valid, but requiring an exact match would rule them out. The
> language here
> should match the existing language for single point conrefs.
I agree that we should (and, I assumed, meant to) match what
happens for single point conrefs in this regard.
paul