W3C home > Mailing lists > Public > public-webarch-comments@w3.org > July to September 2004

RE: Use of "assign" for URI -> resource

From: Larry Masinter <LMM@acm.org>
Date: Thu, 16 Sep 2004 15:03:25 -0700
To: "'Stuart Williams'" <skw@hp.com>
Cc: public-webarch-comments@w3.org
Message-id: <0I450076TMLPXN@mailsj-v1.corp.adobe.com>

> The direction you offer is largely to dispense with the concept of "URI 
> ownership" and shift the emphasis toward ownership to resources.

Well, I'd like to get away from ownership as much as possible,
and shift the emphasis to URI use. But insofar as WebARCH seems
to have made a big investment in 'ownership', it's more plausable
that resources are owned than URIs (except for URNs, which *are*
owned).

> I made 
> a stab at framing "URI Ownership" in terms of the rights and 
> responsibilities of  "URI owners".  I have some difficulty accepting 
> that there is not a sense in which at least some URI are 
> owned, and its clear that not all resources are own - so we cannot say 
> generally that it is resource owners that make assignment of a URI as an 
> identifier of a resource - eg. the planet Mars.

I was trying to be careful: resource owners don't make the
"assignment" of a URI as an identifier of a resource; in fact,
I was also trying to eliminate the verb "assign". The only URIs
that get "assigned" are the ones for which assignment is
part of the semantics, e.g., URNs.

Instead, the meaning of a URI -- to the reciever of a URI --
comes from the URI scheme and its definition. No out-of-band
information that isn't directly derivable from the characters
of the URI and the definition of the scheme and its delegation
should come into play.  If you want to cause a URI to mean some
resource, so that you can talk to someone else about that
resource, you must arrange those resources under your control,
as best you can, so that some URI that you can speak identifies
the resource that you want to speak about.

To talk to you about Mars, I can give some reference indirected
through RDF and the "#" mechanism or use 'tdb'.
(If we change the question to 'how to speak about
Venus', I'll note that Frege had trouble with the morning star
and the evening star as two different RIs that might or might
not be about the same resource.)

I would be happy to dispense entirely with the notion of
"ownership", since I think it makes the web architecture
messy and adds an unnecessary complexity. I'd be happier
to talk about "capabilities" rather than "ownership", and
then, just for resources. There are some individuals, groups,
or technical (why 'social'?) entities that have the capability
to modify resources, and change the behavior seen when
resources are contacted or asked for representation.

> BTW:  I also get vexed with the number of different words that we get 
> prefixing or post fixing "URI", which are synonymous and 
> which are not.
> 
> URI Allocation: a transfer of ownership rights over one or more URI to 
> some social entity (which can include scheme specifications) - an act of 
> delegation in URI space.

I'm certainly not happy with this formulation, since I believe
what actually happens here doesn't have much to do with "ownership
rights" and more to do with "capability to modify resources identified
by one or more URIs"


> Is there not some entity that obtains a right to make an association 
> between a given URI and a resource?

No, not in general. "data:,12" means what it means. No entity has
any right to associate that with Mars.  "http://a.b.c/d/e/f"
means, to a receiver, the resource you get when you open a HTTP
connection to a.b.c and talk about "/d/e/f" to it. No entity
has a 'right' to 'make an association' between that resource
and the planet Mars. Perhaps, the domain admin for c can allocate
to the domain admin for b.c to the sys admin to a.b.c to the
owner of directory /d to the owner of /d/e to actually modify
the resource that's associated with /d/e/f, but these acts aren't
"make an association between a given URI and a resource", but rather
"arrange resources such that the URIs which identifies that resource
reaches a resource that can be used as a proxy for Mars".

> Such rights may come to them through some delegation chain rooted 
> in the URI spec. and the IANA scheme registry.

Wrong end of the telescope. Neither the URI spec and the IANA scheme
registry give anyone any rights. All they do is tell someone
who gets a URI how they might interpret the URI.  If there
are any 'rights' (or, better, capabilities), it is all on the
side of resource owners.

> It's also not clear that all resources have owners. If resources is 
> scoped so large as to includes say planets or galaxies, I'm not sure I'd 
> be able to attribute any particilar owner to say the planet Mars - 
> though I guess some may stake a claim :-). However, I may have rights in 
> some sense to associate the URI http://example.org/planets/Mars or 
> http://example.org/planets#Mars with the planet Mars in such 
> away that one or both of those URI are said to identify (in a 
> denotational sense) the planet Mars.

No, only in the sense of the indirect references.
http://example.org/planets/Mars
already has a clear definition, it is the resource that you get
when you open a HTTP connection to "example.org" and talk about
"/planet/Mars".
Any association this has with the actual planet Mars is a matter
of association or imagination or interpretation of that resource
itself.

> I might then want to back this up by making available representations of 
> the planet Mars to anyone who happens to dereference one or both of 
> those URI. Of itself that exposes another conuderum... by deploying 
> representations have I :

> 1) magically turned Mars (which I do not own) into an 
> information resources.

This trick doesn't work. Mars is Mars.  You can't manipulate Mars.

> 2) deployed a new information resource (which I do own) that 
> then stands proxy for the planet Mars, and gives me the grief that I may
want to 
> speak independently of the proxy and the planet.

You have this grief. It is one of the fundamental problems of
language: you can never get to Mars, but only to its proxies.
"tdb" might give you a hope of speaking independently of the
proxy and the planet.

> 3) been entirely misguided and the URI (either of then) never 
> identified the planet Mars in the first place.

Yes, indeed.

> Anyway...  would you claim any rights over the URI based on the DNS name 
> "larry.masinter.net" (not the resources they identify, but the URI 
> themselves) ? 

I don't think the personal reference changes my opinion,
even if I wanted to maximize my personal rights.  I can't
suddenly say "by http://larry.masinter.net
I don't mean "the resource you get when you open a HTTP
connection to 'larry.masinter.net' and talk about '/'".
The URI means what it means, I have no control over that
meaning, only over behavior.

> ># Many URI schemes are used to described resources
> >  
> >
>                                                               
>     ^^^^^^^^^
> URI schemes "describe" resources? Maybe - "describe" doesn't 
> feel like the right verb.

You're right. 'identify' is the best verb to use to describe
what Uniform Resource Identifiers do to Resources.

> >This gets rid of the notion that the social entity
> >owns the URIs: it owns the resources, which makes
> >more sense.
> >  
> >
> I can't see that it is necessarily the case that all resources have 
> owners, and I'd be willing to concede that not all URI have owners. 
> However, I think that, socially, we vest ownership of the URI 
> space in the IANA registry and the social process (es) that result in the 
> inclusion of scheme specifcations in that registry.

Wrong end of telescope. There is no ownership needed. The IANA
registry just tells users how to interpret URIs, and the
social process is that writers (those who toss around
URIs) can assume that readers (those who get URIs and
want to know what Rs they I) agree about what the IANA
registry says.

> In a sense, when a scheme is registered, ownership of that chunk of a URI
space 
> passes to the registered scheme specification (and the community that
maintains 
> it)... and the scheme specification itself my further delegate ownership 
> to other registry/specification combinations (eg. URN namespaces) and/or 
> organisations, roles within the organisation and so-forth.

Wrong end of telescope. When a scheme is registered, it lets
recievers of URIs know how they should interpret URIs that
start with that scheme. No ownership is given to anyone.

> >I think it's possible to eliminate "URI ownership"
> >in the rest of the document fairly easily.
> >  
> >
> You used the word "use" above with respect to URI has a very 
> egalitarian 
> feel to it in the sense of there being no primacy of use. 

I don't mean to be egalitarian, I mean that the
power is with the readers, the interpreters, the
utterers, those who are involved in a communication
about resources and want to use URIs to identify the
resources they're talking about.

> That seems to me to work as far as a "USE" of a URI is it's occurence in 
> representations or as a protocol element or written on a 
> piece of paper.
> 
> But "USE" does not establish who/what gets to form an association 
> between a URI and the resource it identifies (either 
> operationally [*] or denotationally [#]).

Well, again, you're stuck in 'assign' land, where you think
that someone "gets to form an association". But I believe
that it is a better model that the association is intrinsic
in the definition (except, again, in the special case of URNs
where meaning *does* get assigned.)

> [*] operationally in the sense of following the documented mechanisms 
> for dereferencing URI - which I believe is the way you use the word 
> identify.
> [#] denotationally in the sense of folks using URI to name 
> things in the world (like planets) - which I believe is also a way in
which 
> the word idenify is used.

Yes, I accept both modes of identification.

> When a URI identifies what it is used to denote then 
> there appears to be a level of harmony (modulo some discussion of 
> whether its denotation is an information resource/hypertext 
> dispenser or the thing that described or depicted by that information
resource).

Just because you can think of URI identification both operationally
and denotationally doesn't mean that the resources identified
are different.

In order to do most kinds of denotation (like to talk about
the planet Mars), you need some kind of implicit or explicit
indirection.

> ># The owner of a resource may arrange the resources
> ># such that a URI can be used to obtain representations
> ># of the resource identified by the URI. For example, when
> ># a resource owner offers the HTTP protocol ...
> ># ... on behalf of the resource owner to provide ...
> >  
> >
> Can you explain how this would work for the planet Mars?

Only through indirection. In fact, it's unnecessary to arrange
any resources: I can use urn:tdb::data,the%20planet%20Mars
to talk to you about the planet Mars. No ownership required,
either of URIs or of resources.

Or I can use
 http://larry.masinter.net/phrases.rdf#the%20planet%20Mars

and arrange some resources: create a file phrases.rdf with
a 'the planet Mars' label in it, and 'arrange the resources'
by sticking the file on the server that servers 
http://larry.masinter.net (Thanks to John Masinter, who offered
free web hosting to anyone named 'Masinter').
 
> >>A URI owner SHOULD NOT associate arbitrarily different URIs 
> with the same resource.
> >
> >I think this makes little sense as stated, and is
> >much more effective changing 'URI owner' to 'resource owner'.
> >  
> >
> I think this makes it a little difficult to associate a URI with a 
> resource like the planet Mars (that may be deliberate).

Well, I don't really like the result all that much. How
about

# Users SHOULD NOT use arbitrarily different URIs for the
# same URI.

Larry
-- 
http://larry.masinter.net
Received on Thursday, 16 September 2004 22:03:48 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 20:26:47 UTC