XInclude namespace issue draft response [was: Minutes for XML Core WG telcon of 2015 February 18]

On 2015-02-18 10:59, Paul Grosso wrote:
>

>>
>> Eric van der Vlist (via Norm) asks what the post-XInclude infoset
>> is for this element:
>>
>> <xi:include xmlns:my="MYNAMESPACE"
>> href="mydoc.xml" fragid="element(foo)"
>> my:root="true"/>
>>
>> when mydoc.xml is:
>>
>> <doc xmlns:my="YOURNAMESPACE">
>>   <my:note xml:id="foo">
>> <my:x/>
>> </my:note>
>> </doc>
>>
>> Paul suggested it would be
>>
>>   <my:note xml:id="foo" my2:root="true" xmlns:my="YOURNAMESPACE" 
>> xmlns:my2="MYNAMESPACE">
>> <my:x/>
>> </my:note>
>>
>> where the "spelling" of the prefix "my2" is implementation dependent?
>>
>> What do others think, and can we find an answer in the spec?
>>
>
> Henry agrees with Paul that we are including infosets,
> not text, so Paul's solution is fine.  He points out
> that this isn't an infoset problem, it's just a
> serialization issue--what prefixes to use for the
> various namespaces, but what matters in the infoset
> is the value of the namespace property, not the prefix.
>
> We still need to look at the spec to see if it's clear
> (or at least doesn't say something to the contrary).
>
> Jirka says we should check section 4.3.  Henry says
> there it talks about qualified names which has to do
> with local names and namespace URIs, no prefixes involved.
>
> We will check with Norm to see what he says, and then
> Paul will write a response.
>

Here's my draft response--please comment.  Especially
comment on whether you feel we should add some (minor)
wording/Note to XInclude 1.1 as we go to CR:


Regarding the example in the comment at [0], the resulting
infoset is clearly:

<{YOURNAMESPACE}:note xml:id="foo" {MYNAMESPACE}:root="true">
     <{YOURNAMESPACE}:x/>
</{YOURNAMESPACE}:note>

The only complication comes when trying to serialize this infoset.

(Note also the discussion in the Infoset spec about "Synthetic
Infosets" [1] and also the comments in XInclude's section on
Namespace Fixup [2].  See also the analogous discussion of
adding namespace declarations in the XSLT Recommendation [3].)

When serializing this infoset, the namespace prefixes emitted by
the XInclude implementation are implementation dependent.  See the
fourth paragraph under section 4.7 "Creating the Result Infoset"
at [4] where it says:

  Some processors may not be able to represent an element's
  in-scope namespaces property if it does not include bindings
  for all the prefixes bound in its parent's in-scope namespaces.
  Such processors may therefore include additional namespace
  bindings inherited from the include parent in the in-scope
  namespaces of the included items.

We could perhaps either augment the above paragraph or the
Namespace Fixup section to make an explicit reference to the
case where prefixes may need to be modified to avoid emitting
namespace declarations with the same prefix and different
namespace names, but in my opinion, that behavior is already
implied by the existing wording of the spec.

For example, one reasonable serialization of the infoset
resulting from the example in question might be:

<my:note xml:id="foo" my2:root="true"
          xmlns:my="YOURNAMESPACE" xmlns:my2="MYNAMESPACE">
   <my:x/>
</my:note>

paul

[0] {eventual URL of the comment in the www-xml-xinclude-comments list}
[1] http://www.w3.org/TR/2004/REC-xml-infoset-20040204/#intro.synthetic
[2] http://www.w3.org/TR/2014/WD-xinclude-11-20141216/#namespaces
[3] http://www.w3.org/TR/xslt#section-XML-Output-Method
[4] http://www.w3.org/TR/2014/WD-xinclude-11-20141216/#creating-result