Re: URI ref comparison - clarification requested

On Fri, 27 Aug 2004 15:30:29 -0700, Roy T. Fielding <fielding@gbiv.com> wrote:
> > There is still an aspect of this that makes me a little uneasy, though
> > I doubt that it's significant even if my reasoning makes sense. If a
> > spec like RDF says it's using URIs but provides its own comparison
> > mechanism (such as the first approximation of string equiv), then
> > applications built to that spec may systematically, as a group, behave
> > differently than apps built directly to the URI spec (possibly
> > including support for better approximations). That systematic aspect
> > seems a step beyond different apps implementing different variations
> > of the original options.
> >
> > Where the primary practical use of the URI is in the process of
> > obtaining a representation of the resource identified,  the
> > comparisons only (potentially) producing false negatives seem to
> > preclude problems. I suspect it might not be such a failsafe in the
> > general case when is used in constructing logical statements (though I
> > might well be mistaken, IANAL).
> 
> I am kind of curious how a system constructing logical statements
> could somehow fail in a non-safe way just because two equivalent URI
> are considered different.  I think, at most, it just adds one to the
> number of aliases, and thus the admonishment against creating
> arbitrary aliases for a resource still applies.  If the RDF graph
> contains conflicting assertions for two equivalent URIs, then those
> assertions are broken regardless of the comparison algorithm; that
> brokenness is simply made harder to discover due to the lax method
> of comparing URIs -- merely declaring the URIs to be different
> does not cause the assertions to be true.

Good point, this would suggest that no new badness is introduced into
the usual model, assuming aliases, open world etc. I suppose it could
be an amplifier of brokenness in a system that had added things like
implication rules, negation.

> However, it is important to note that the reason RDF specifies
> it that way is because the probability of encountering two
> equivalent but not string-equal URIs in the same RDF graph is
> quite small, and easily avoided by use of canonical URI forms.

One of the Atom issues that led down this path was whether it was a
good idea for publishers of URI-identified pieces of information to
canonicalize the URIs. Your comments  give a clear signal on that, so
thanks again.

Cheers,
Danny. 


-- 

http://dannyayers.com

Received on Saturday, 28 August 2004 07:57:37 UTC