W3C home > Mailing lists > Public > public-rdf-wg@w3.org > June 2011

Re: C14N use case: version control

From: Gavin Carothers <gavin@topquadrant.com>
Date: Tue, 28 Jun 2011 07:48:55 -0700
Message-ID: <BANLkTikfQudiw3YnTGnDhzjWJicVuUXocw@mail.gmail.com>
To: Andy Seaborne <andy.seaborne@epimorphics.com>
Cc: RDF-WG <public-rdf-wg@w3.org>
On Tue, Jun 28, 2011 at 6:55 AM, Andy Seaborne
<andy.seaborne@epimorphics.com> wrote:
>
>
> On 27/06/11 19:55, Jeremy Carroll wrote:
>>
>> This is the use case that TopQuadrant has internally that prompted
>> discussion between me and Gavin leading to this thread on this mailing
>> list.
>>
>> A significant portion of our product source is in RDF.
>> We are migrating our version control system to GIT to reduce cost of
>> merging
>> This will not work for RDF in the form that we currently store it,
>> because simple changes result in completely different documents
>> We are now working on a version of my earlier paper with additional
>> steps to insure reasonable stability of blank node IDs.
>>
>> (In the terms of the paper the bnode ids will be based on a hashcode
>> generated from the first distinctive triple for that bnode).
>
> I'm curious - why not store skolemized data?  The skolemization URI could
> record sufficient information related to when the bNode id was first
> created.  It's then fully reversible in syntax terms (with a little parser
> processing "deskolemization") to make bNodes reconstructable.

Is skolemization fully transitive? Once the parser has created the
blank node will another seralizer produce the same skolemized blank
node? I don't think the current proposals do that. How would that
happen? Would all implementations keep track of the skolem id after
reconstituting the blank node? In which case aren't we just turning
blank nodes into syntactic sugar for a skolem iri? (No objection to
that btw, but it's a bit odd)

>
> For me, this is the point of skolemization - finding a bNode again need not
> be just across the web; it can also be temporally across serialized data.

I agree, this FELT like the point of skolemization, but I'm not sure
that skolemization actually provides this. I'm just not sure how the
skolemization process is supposed to provide a stable blank node label
without either solving graph isomorphism or implementing something
very much like the c14n proposal here. In other words I do think that
skolemization could be used rather then "modifying the graph" ...
mostly since skolemization modifies the graph.

--Gavin
Received on Tuesday, 28 June 2011 14:49:41 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 16:25:44 GMT