On 24 Sep 2009, at 22:27, Philip Taylor wrote:
Shane McCarron wrote:
[...] I was just trying to reflect in errata the discussion that
went on today on the RDFa Task Force call.
Would you prefer that we turn this around, explicitly indicating
that 'xmlns' is not permitted at all, and that 'xml' is only
permitted to be used in CURIEs if it is declared?
Mostly I want there to be a single unambiguous mapping from any
given source document to a set of triples, so I really don't want
the spec to say "a conforming RDFa Processor MAY provide these
mappings" and introduce uncertainty - it should be a strict
requirement one way or the other.
I agree with Philip. I'd prefer there to be strict rules wherever
possible rather than leaving things implementation-dependent.
In this case I believe the current spec text is requiring that the
mappings must not be provided, and so an errata that says they must
be provided would be changing the normative behaviour defined by the
spec, which seems undesirable. If it's unclear to implementors, it
would be good to have test cases for this; I'd be happy to leave the
spec as it is, though I'd also be happy for the spec to have
explicit text to guide implementors safely around the confusion
caused by XML Namespaces.
I agree that this is what the spec says.
The problem from my perspective is that it is impossible for an XSLT
implementation (such as a GRDDL stylesheet) to know whether or not the
'xml' prefix has been declared through a namespace declaration. In the
XDM on which XSLT and XQuery are based, all elements have an 'xml'
namespace node whether or not one is declared, so documents that
contain xmlns:xml="..." namespace declarations and those that don't
produce equivalent XDMs.
There seem to be the following options:
1. Leave the text as it is now, with the consequence that it's
impossible for an XSLT implementation to be fully conformant.
2. State that the 'xml' prefix is implicitly bound to the XML
namespace such that xml:* will always work.
3. State that the 'xml' prefix is reserved and must not be used for a
CURIE (such CURIEs are then invalid, and ignored).
I would prefer 2 or 3 than 1, because I think it's important for XSLT
and XQuery implementations to be feasible. I don't have a particular
feeling either way about which of 2 or 3 is best.
The situation is different for the 'xmlns' prefix. It's illegal for
people to declare the 'xmlns' prefix in a document anyway, so a
clarification that stated that the use of xmlns:* CURIEs is always
invalid would work just fine.
Cheers,
Jeni
--
Jeni Tennison
http://www.jenitennison.com
