Philip,
On 23 Sep 2009, at 09:07, Philip Taylor wrote:
Jeni Tennison wrote:
Philip,
I just updated rdfQuery to address some of the new test cases that
you've done. Thank you for taking the time to make the edge cases
explicit. The latest trunk rdfQuery output will differ from the
samples you give in some places, so just to detail them here and
explain why:
Great, thanks for looking at this! I've made a few changes to the test
cases, and updated the test results to use the current trunk.
Thanks :) And thanks for the inclusion of the Opera results. It might
be interesting to include results for other browsers too.
2. For 'Underscore xmlns prefix', where the test case is:
<p xmlns:_="http://example.org/"; property="_:test">Test</p>
rdfQuery retains the legal namespace declaration but because the
CURIE begins with _:, which is how blank nodes are indicated, it's
interpreted as a blank node with the id 'test'. Predicates cannot
be blank nodes in rdfQuery, so it's treated as a bogus value and
ignored. No triples are created.
I don't see anything in the RDFa spec that justifies this behaviour.
As
far as I can tell, xmlns:_ updates the list of in-scope mappings, and
then the CURIE is converted to URI with the steps "Split the CURIE at
the colon to obtain the prefix and the resource. Using the prefix and
the current in-scope mappings, obtain the URI that the prefix maps to.
Concatenate the mapped URI with the resource value, to obtain an
absolute URI". The prefix is '_' and so the relevant in-scope mapping
applies, just like any other prefix, so the mapped URI comes from
xmlns:_ and the resulting absolute URI is used as the predicate.
This doesn't really sound like sensible behaviour, but it's how I
interpret the spec. Am I misunderstanding something? Does the spec
need
to be fixed?
Well, in Section 7 of the RDFa syntax specification [1] it says:
"In normal evaluation of CURIEs the following context information
would need to be provided:
...
* a mapping to use with the '_' prefix, which is used to generate
unique identifiers (for example, _:p).
...
In RDFa these values are defined as follows:
...
* the mapping to use with the '_' prefix, is not explicitly stated,
but since it is used to generate [bnode]s, its implementation
needs
to be compatible with the RDF definition."
So it seems to me that RDFa holds the '_' prefix in special regard,
thinks that the CURIE specification holds the '_' prefix in special
regard, believes that it should not be interpreted in the same way as
other prefixes, and that it's used to generate bnodes.
What I wish is that the CURIE specification said something clearly
about the priority between "the mapping to use with the '_' prefix"
and the "set of mappings from prefixes to URIs". All it says is [2]:
"The CURIE prefix '_' is reserved for use by languages that support
RDF.
For this reason, the prefix '_' SHOULD be avoided by authors."
My interpretation is that 'reserved' means that the special use of '_'
takes precedence over any other prefix mappings that may be provided
to it.
I do think the CURIE specification could be improved by defining in
much clearer language the four context items that the RDFa
specification states are used to interpret CURIEs.
3. For 'xmlns prefix 'xml' with incorrect URI', where the test case
is:
<p xmlns:xml="http://example.org/"; property="xml:test">Test</p>
rdfQuery ignores the bogus namespace declaration, but uses the
built-in namespace declaration for the prefix 'xml' and therefore
generates the triple:
<> <http://www.w3.org/XML/1998/namespacetest> "Test" .
What built-in namespace declaration? The spec says there is "An
initially empty list of [URI mapping]s, called the [local list of URI
mappings]", and the list is only updated by xmlns:* attributes on
processed elements, so it sounds like a violation if the list has some
built-in entries and is not initially empty.
Same issue for the tests that use xmlns:xmlns="...".
The only one that should generate triples is
xmlns:xml="http://www.w3.org/XML/1998/namespace"; -- XML Namespaces
says
the xml prefix/URI "MAY, but need not, be declared, and MUST NOT be
bound to any other namespace name", so this particular declaration is
legitimate and should add to the list of URI mappings. (It looks like
rdfQuery is only accidentally passing this test, because of the
built-in
mapping for xml, so the bugs are cancelling each other out :-) )
Yes. Short answer is that I don't know what should happen here, and I
deliberately chose to take a different position from your test in the
hope that someone on the RDFa TF would clarify what the correct
interpretation is. Perhaps that was too subtle. I'll post a separate
mail to explicitly call it out.
8. The language-based tests aren't met, because I'm currently at a
loss as to how to amend rdfQuery to work out whether it's being
used in an HTML or XHTML setting and therefore whether xml:lang is
a lang attribute in the XML namespace or an attribute called
'xml:lang' in no namespace.
You shouldn't need to know the setting.
http://whatwg.org/html5#language defines the language of an element in
terms of the DOM, independent of the syntax, so it's the same for
documents parsed from HTML and XHTML.
Yes, it would be nice if I could take that attitude. Even nicer if I
could rely on the DOM's lang property. I'm just not quite sure how to
square it with the requirement to work in legacy browsers (such as IE).
On a related note, I fixed one of my language test cases (the input
markup <p property=... xml:lang="aa">Test</p> in text/html ought to
produce a literal with no language), and I think the rest are all
testing for the correct output (per HTML5's definition of language).
OK.
Jeni
[1]: http://www.w3.org/TR/rdfa-syntax/#s_curies
[2]: http://www.w3.org/TR/curie/#s_syntax
--
Jeni Tennison
http://www.jenitennison.com
