Namespaces and infosets.

In this response to Paul I agree that this topic should drop, but 
just wanted to clarify a few things.

At 2:00 PM -0500 6/23/00, Paul Grosso wrote:
>At 13:45 2000 06 23 -0400, David G. Durand wrote:
>>At 11:32 PM -0500 6/22/00, Dan Connolly wrote:
>>>(as Paul G put it:
>>>2.  say that a document containing an nsattrib whose value is a
>>>      relative URI has no defined infoset.)
>>
>>I'm not 100% sure that this is a good idea. My objection is based on
>>a separate set of issues:
>>
>>   1. Xlink should work for any well-formed XML document.
>
>XLink doesn't work on XML documents at all.  XLink allows
>you to link between all kind of resources.  XLink uses
>XPointer to address into XML documents (and allows the
>use of other things to address into other resources).
>I suspect you meant to say "XPointer" here.

Indeed. Apologies for the imprecision. At least the point got through.

>  >  2. XLink is defined on the information set of an XML document.
>
>I'm not sure this is really related to the primary topic of
>discussion on this list, but (1) XLink isn't defined on the
>information set of an XML document.  You are probably thinking
>of XPointer, not XLink.  (2) XPointer probably should be defined
>on an infoset, and maybe it is, but I'm not sure because XPointer
>is defined on top of XPath and (3) XPath is not defined on top
>of infoset, rather it defines its own data/object model.

I'm quite sure about Xlink. I may be mistaken on Xpointer, but 
certainly people in the Xlink working group have been talking as 
though it is using the infoset. If you're right, then it's good 
fortune that Xlink has sidestepped a serous bug in the infoset.



>This is already not the case.  The infoset only covers well-formed
>XML documents that are namespace-valid.  Long ago, we made the
>decision that the Infoset spec would not cover well-formed documents
>that are not namespace-valid.

I was already getting this idea. As I said in the original posting, 
that's an infoset issue, and not directly a namespace issue. On the 
other hand, I wanted to call attention to the serious problems 
introduced by legal syntax that has not data model. I'm not on the 
infoset discussion groups, and I've not been paying close enough 
attention, clearly, but there are so many hours in my days, and 
they're never enough.

The idea that a relative URI reference (even though it's a bad idea 
and _should be forbidden, on  my own views), would render the other 
information content of a document undefined is baffling to me. 
Clearly this is an issue for the Xlink WG to raise with the infoset 
group, if it disturbs the rest of us as much as it does me.

>  >Otherwise there would be no way for an Xpointer to select an XML
>>element in a "namespace-defective document" because that document
>>would not have an infoset.
>
>Yes, this is an open issue for XPointer already.  For example,
>it is not clear if XPointer can address into an external parsed
>entity because it's not clear that such has an infoset.  This
>isn't just a namespace problem.

Right.


>  >Alternatively, if it's allowed to ask for the infoset of an XML
>>document "explicitly ignoring" any apparent namespace declarations,
>>then a fallback policy would be possible.
>>
>  >This is really an information set issue, that touches on the
>>namespace issues, but it does seem closely enough related that it's
>  >worth thinking about.
>
>Yes and no.  It has long ago been decided that the Infoset spec
>would only define infosets for well-formed XML documents that
>are namespace-valid.  Therefore, all we need to decide now is
>what documents are namespace-valid (and what the actual
>namespaces names are in all cases); that is all that is in scope
>for this discussion.

Right. But it's not quite that clear-cut, as it's reasonable for the 
Wlink WG to raise a formal issue about this with the infoset group. I 
will move any further comments out of this forum.

>After this in-scope issue is addressed, then the XML Core WG can
>decide what will be the infoset of those namespace-valid documents.
>And then perhaps the XLink WG needs to figure out how XPointer is
>involved, and maybe even XPath needs to be rewritten to take the
>Infoset into account (though I'm not really suggesting that just
>now), but the point is that all this is out of scope of the current
>discussion.

But perhaps it's worth noting the issue, as it certainly affects my 
view of whether deprecated relative URIs should prevent the creation 
of an infoset. Because of linking, I believe that an infoset should 
be defined for this ugly case.

  -- David
-- 
_________________________________________
David Durand              dgd@cs.bu.edu  \  david@dynamicDiagrams.com
http://cs-people.bu.edu//dgd/             \  Chief Technical Officer
     Graduate Student no more!              \  Dynamic Diagrams
--------------------------------------------\  http://www.dynamicDiagrams.com/
                                              \__________________________

Received on Friday, 23 June 2000 19:50:17 UTC