RE: the necessity of describing responses in-band

On 6 Okt 2015 at 14:17, Ruben Verborgh wrote:
> I've written a blog post that describes the necessity
> of describing responses in-band:
>     http://ruben.verborgh.org/blog/2015/10/06/turtles-all-the-way-down/

Nice post! One question I have is why you blanked the URL bar in the second
screenshot? Even if the resource itself isn't self-describing, the client
would still know how it retrieved it, no?

I have to say though, that I'm not a big fan of exchanging named graphs as
they have undefined semantics...

On 7 Okt 2015 at 23:00, Ruben Verborgh wrote:
> On 7 Okt 2015 at 22:13, John Walker wrote:
>> Another point I'd like to raise is the use of quads rather than triples.
>> As you are most likely aware there are no agreed formal semantics for RDF
>> datasets [1].
> 
> Yes and that's quite horrible.
> anybody knows why the graph IRI is only a syntactical construct?

It is the graph's name but it's semantics are not clear. It is undefined
whether the name denotes the graph or not. The reasons for that are because
the RDF WG at the time couldn't find consensus. Graph names have
historically used for different purposes (timestamp graphs, record
provenance, slice data by subject, ...) and it wasn't clear which to elect
to the "winner" and what would break by doing so. 


> Seemed much more logical to make it the name of the graph,
> would give the term "named graph" a more logical meaning.
> And if you want a graph that has nothing to do with the name,
> just pick a different name then anyway.
> 
>> But do you think that using quads would come at the risk of
interoperability
>> issues?
> 
> Hard to predict, but I don't think so.

With the current state of affairs I think using datasets on the public Web
is basically made impossible. The only thing that's defined, is that if a
client expects a RDF graph but gets a dataset instead, it should use the
dataset's default graph and ignore the named graphs.


>> What would happen if someone did a LOAD operation of one of these quads
>> documents into a store?
> 
> Nothing bad, it seems. Do you think of scenarios where things go wrong?

As it is undefined, it depends solely on the implementation.


--
Markus Lanthaler
@markuslanthaler

Received on Thursday, 8 October 2015 15:24:02 UTC