Documents, Anchors, Nodes and Context was: RE: Context and... Re: the case of two bats

Tim Berners-Lee wrote:
>
> You wrote,
>
> "Now do you expect both "../2000/vocab#" relative URIs to be *absolutized*
> wrt the base URI and compared?
>
> Of course, is the strings are considered to be URI references.
> The URI ref is just a shorthand form for a URI, which is the
> namespace name.
> Relative URI references are (until;/unless one invents something
> different)
> expressed with respect to the URI of the document.
>
> "This is equally ridiculous."

	What I am saying here is that assuming the context == base URI, and hence
absolutization, does not in and of itself solve the semantic problem of the
two bats. One problem, the example DanC gave, demonstrates the problem when
the examples are in two documents (the context is the document). I expressed
the identical problem when the examples are in two distinct subtrees of the
same document. I view intradocument addressing as an extension of
interdocument addressing, and prefer the model not assume arbitrary
behaviors at document boundaries (beyond what is defined as a document node
in the infoset). There are several XML recs which address (:-) this issue
and these ought be considered *along* with RFC 2396.

	Assuming we step back and use a Grove/Infoset view of the addressing issue,
the URI gets us to a document, and the fragment identifier gets us to a
location within a document. By employing more precise addressing, the
semantic ambiguity is reduced (why assume the nodes of a semantic network
are distinct documents?)

	In the XML view as expressed by the current RECs which define this issue,
i.e. XPointer/XSLT/XPath, the context address includes the fragment
identifier, i.e. a path within the document.

	If we don't care about semantics, or resource retrieval, then the literal
approach solves the problem: the name is indeed a name, and just a name. If
we care about addressing and/or resource retrieval then the absolutize
approach only goes part way and doesn't itself solve the problem. If we
really care about naming as referencing an abstract resource (a node in the
semantic network!) then (in the XML infoset context, or grove context) the
XPointer to the node where the namespace is being defined ought be defined
as the 'namespace context'.

Jonathan Borden

Received on Thursday, 15 June 2000 09:32:58 UTC