W3C home > Mailing lists > Public > w3c-rdfcore-wg@w3.org > July 2001

Re: #rdfms-identity-anon-resources: provenance

From: Brian McBride <bwm@hplb.hpl.hp.com>
Date: Wed, 25 Jul 2001 17:47:41 +0100
Message-ID: <3B5EF82D.8E147A7A@hplb.hpl.hp.com>
To: Graham Klyne <Graham.Klyne@Baltimore.com>
CC: w3c-rdfcore-wg@w3.org

Graham Klyne wrote:
> Brian,
> You said:
> >To represent exactly the information recieved, anon resources must
> >be a part of the model/abstract syntax.  Yes? No?
> I think that to "represent exactly the information received" the entire
> RDF/XML document received must be represented verbatim.  I don't think
> we're going there.

I agree.  I should have made clear that my hypothesis here is that it is 
the 'model' that was signed, not the document.

> In defining RDF, and in particular its semantics, we are in the process of
> identifying the significant information in a received RDF/XML document.  We
> are making choices about what constitutes "information received" (guided by
> the existing XML and RDF M&S specs).
> Signatures present an interesting case, because the most trivial change
> (e.g. adding or removing a space in an XML comment) changes/invalidates the
> signature.  

That was not what I meant.

> So if we mean to convey the exact meaning of a signature (i.e.
> that a certain key signed a certain sequence of bits) we must keep a copy
> of the bits (or at least a copy of the digest of the bits that was signed).
> When one takes a signed XML document and interprets it as RDF, I think one
> needs to make some choices about what it means, which will be some function of:
> (a) the information conveyed by the bare RDF, and
> (b) the information conveyed by the signature
> So suppose I receive a statement:
> (1)   _:exists-X hasProperty Y .
> which is signed Alice.  Also suppose that I apply Skolemization represent
> this information:
> (2)   skolem:12345 hasProperty Y .
> where I am certain that nobody will independently use skolem:12345 to
> represent any resource.  Now I am not entitled to say that the statement
> (2) was signed by A, but I might legitimately say that I received some
> document signed by A whose content was interpreted to mean (2).

Good example.  You can't say that (2) was signed by A because that is
not the statement A sent.

And the question for me is do we prefer an abstract syntax
that allows us to express what A sent to us, or are we happy to loose
information in the process of transforming RDF/XML to the abstract syntax.

> When I sign a cheque, it is a piece of paper that I sign.  When the cheque
> is presented for payment, the bank interpret the signature to mean that I
> have authorized the payment of some amount to some person.  If someone
> writes a telephone number in the margin of the check, the exact content of
> the document is changed, but the authorization conveyed by the signature is
> not.  Thus, any signature as well as the signed document need to be
> interpreted.

I didn't intend that we get hung up on the nature of a signature.  The
intent was to try to draw attention to the issue of whether the received
model is the same information as the sent one.  I think we've agreed
that if the receiver skolemizes, then the information is not the same.

Consider running the scenario twice and the second time the anon resource
gets skolemized to skolem:54321.  Given 

  (3) skolem:54321 hasProperty Y .

do (2) and (3) represent the same information.  If the answer is yes,
then you must be treating skolem URI's differently from say URL's.  If
the answer is no, they can't both represent the same information as in (1).

Received on Wednesday, 25 July 2001 12:50:46 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 14:53:50 UTC