Re: Fragment Identifiers and Agent Perspectives

On Mon, Oct 10, 2011 at 11:40 AM, Noah Mendelsohn <nrm@arcanedomain.com> wrote:
>
>
> On 10/7/2011 7:11 PM, ashok malhotra wrote:
>>
>> Currently, the specs say "fragment identifier semantics are defined by the
>> media type".
>> We should amend this to say "fragment identifier semantics are defined by
>> the media type and
>> the kind of agent that is making use of the markup".
>
> I understand the proposal, but I'm not convinced it's good architecture.
> Webarch [1] distinguishes direct and indirect identification, making clear
> that a given URI should be a first class identifier for at most one
> resource. If the same URI is to be used to identify something else as well,
> then the identification is viewed as indirect.

Not sure I see the difference. If indirect identification is a kind of
'identification' (as seems to be suggested by the grammar) then you
still have one 'identifier' 'identifying' two things; admittedly with
different adverbs attached, but it's still multiple 'identifications'.

One could rescue unique 'identification' by the somewhat cynical trick
of substituting a different verb, as I think has been suggested a few
times - you could say X 'identifies' Y but 'refers to' or 'names' or
'denotes' or 'designates' Z.

> Crucially, it's not the nature of the agent making the reference, but the
> nature of the dereference operation that's in question. Thus, we might say
> e.g. that http://example.org/people.html#noah is a first class identifier
> for a part of the document people.html, and an indirect identifer for a
> person Noah. Any agent can then decide whether it wishes to do a direct or
> indirect dereference. I think that's far stronger architecturally than
> saying that: "fragment identifier semantics are defined by [...] the kind of
> agent that is making use of the markup".

Absolutely right, thanks for pointing this out. In Common Lisp an
identifier can name many different things at a time (at least four,
maybe seven or more depending on how you count). The choice is made by
syntactic context, not kind of agent. For example an evaluation agent
would see '(f f)' and take the first 'f' to get its binding from the
function namespace and the second 'f' to get its binding from the
value namespace.

However this doesn't substantially change the proposal; it's just a
refinement. Just replace "agent" with "syntactic context". The request
URI in HTTP is one syntactic context, the href attribute in an <a>
element is another, a subject in an RDF (or RDFa) triple is another.

Jonathan

> Noah
>
> [1] http://www.w3.org/TR/webarch/#indirect-identification
>

Received on Tuesday, 11 October 2011 13:15:15 UTC