Re: +blank node identifiers was- Re: [Graphs] Fwd: Comments on "SPARQL 1.1 Uniform HTTP Protocol for Managing RDF Graphs"

Gavin Carothers wrote:
> On Fri, Mar 18, 2011 at 2:25 PM, Nathan <nathan@webr3.org> wrote:
>> Nathan wrote:
>>> Sandro Hawke wrote:
>>>> On Fri, 2011-03-18 at 19:22 +0000, Andy Seaborne wrote:
>>>>> Is g-snap->g-text is just a function of the content type?
>>>> Well, probably, for our purposes, I think so. There's a trivial case
>>>> where it's not: the  arbitrary non-semantic
>>>> variability in serialization, eg whitespace.  So, some notion of
>>>> equivalence class of g-texts may be important.
>>>>
>>>> There's a related problem I don't know if we can or should address,
>>>> which is how to deal with websites which use cookies or other
>>>> information (IP address, browser sniffing, etc) to customize content.
>>> I was going to raise that, it's where the "resource state" http/rest story
>>> and the rdf "snapshot" story both breaks down.
>>>
>>> Perhaps we should just scrap that story early on and have Named-G-Box = (
>>> <u>, GB ) where GB = { gt1, gt2, gt3, ... }
>>>
>>> Where gt* are g-texts, and GB is a g-box. A g-box being a set of g-texts
>>> over time. And of course a named-g-box is just a GB associated with a URI.
>>>
>>> If we need g-snap, which I'm sure we do, then perhaps each g-text
>>> encodes/serializes a g-snap, and several g-texts may all encode/serialize
>>> equivalent g-snaps, but that requires g-snap equality to be determined.
>>>
>>> O, actually I quite like that, then all g-snap's are anonymous abstract
>>> sets of rdf triples, and it's the set of g-texts (g-box) that is associated
>>> with a name.
>> Ahh, then blank node identifiers would be a property of the g-text, and the
>> g-texts can be associated with a named-g-box, which would surely set blank
>> node identifier scoping to the named-g-box level?
> 
> I think the issue with that is the idea of union graphs, and as Andy
> mentioned on the telecon how that interacts with inference graphs. I'm
> honestly not sure of what to do about blank nodes with graphs at all.
> In all my use of named graphs I've always ended up avoiding them like
> the plague.

I think I agree, and also that the above perhaps says that blank node 
identifiers are in fact little more than serialization/g-text specific 
features that let you reassemble a graph - that is, not all 
serializations would need blank node identifiers, and they have no 
meaning in rdf semantics or concepts.

That said, I feel like there's something/some-detail I'm missing, need 
to think on it harder.

Best,

Nathan

Received on Friday, 18 March 2011 23:16:52 UTC