Re: Content negotiation for Turtle files

I feel we should be crisp about these things.
Its not a question of thinking of what things kind of tend
to enhance interoperability, it is defining a protocol
which 100% guarantees interoperability.

Here are three distinct protocols which work,
ie guarantee each client can understand each server.

A) Client accepts various formats including RDF/XML.
  Server provides various formats including RDF/XML.

B) Client accepts various formats including RDF/XML AND turtle.
  Server provides  various formats including either RDF/XML OR turtle.

C) Client accepts various formats including turtle.
  Server provides  various formats including turtle.

These may not ever have been named.
The RDF world used A in fact for a while, but the
Linked Data Platform at last count was using C.
Obviously B has its own advantages but I think that 
we need lightweight clients more than we need lightweight
servers and so being able to build a client without an
XML parser is valuable.

Obviously there is a conservative middle ground D in 
which all clients and servers support both formats,
which could be defined as a practical best practice, 
but we should have a name for, say, C.

We should see whether the LDP group will define
a word for compliance with C.  I hope so, and then
we can all provide that and test for it.

Tim

On 2013-02 -06, at 11:38, Leigh Dodds wrote:

>> From an interoperability point of view, having a default format that
> clients can rely on is reasonable. Until now, RDF/XML has been the
> standardised format that we can all rely on, although shortly we may
> all collectively decide to prefer Turtle. So ensuring that RDF/XML is
> available seems like a reasonable thing for a validator to try and
> test for.
> 
> But there's several ways that test could have been carried out. E.g.
> Vapour could have checked that there was a RDF/XML version and
> provided you with some reasons why that would be useful. Perhaps as a
> warning, rather than a fail.
> 
> The explicit check for RDF/XML being available AND being the default
> preference of the server is raising the bar slightly, but its still
> trying to aim for interop.
> 
> Personally I think I'd implement this kind of check as "ensure there
> is at least one valid RDF serialisation available, either RDF/XML or
> Turtle". I wouldn't force a default on a server, particularly as we
> know that many clients can consume multiple formats.
> 
> This is where automated validation tools have to tread carefully:
> while they play an excellent role in encouraging consistently, the
> tests they perform and the feedback they give need to have some
> nuance.