On 16 July 2016 at 06:21, Ruben Verborgh <ruben.verbo...@ugent.be> wrote:
> Hi Sarven, Phil, Rob, > > > Are there mediatypes that map to application/ld+json with a profile? > > This begs the question why one would want to do that. > This was investigated by the WG. Over a period of time the pros and cons were collected, and the cons firmly outweighed the pros: https://www.w3.org/wiki/Socialwg/Media_type_for_AS2 However, there is a small but vocal contingent in the WG hostile to Linked Data. For example, one comment from the chair on this topic that was unqualified was "JSON-LD is a non starter". Ultimately this bespoke mime type got voted through IMHO against general consensus, and common sense, and without looking at the evidence collected, which is something that I mentioned in the minutes. So we're stuck with a CR at the moment this is "sort of" linked data, but not quite. It's a real pain. > > > The problem with specific MIME types is > that they are only one-dimensional. > What if multiple dimensions have to be considered? > What if I want an activity represented in RDF, > but not in JSON-LD? Do we need things like > application/activity+turtle or +xml? That doesn't scale… > > Why would we want to have such specific media types > when we already have media types for RDF syntaxes? > After all, the whole idea of RDF data is that it is self-descriptive; > that a client can understand that a document contains > an activity / person / building by interpreting it. > application/activity+json should be totally unnecessary. > > Note that there is currently no official interpretation of the + token, > so a client cannot even assume application/xxx+json is a JSON(-LD) > document. > Clients have to he hard-coded for application/activity+json. > This defeats much of the purpose of RDF. > > >> Format/serialisation/Mime Type, whatever, is not enough. You need to > >> know the profile. Erik Wilde has an internet draft on this [1] too. > > I definitely think profiles are the way to go. > > >> Should we also consider it as another dimension of content negotiation? > > > Definitely. That's the only use case I see for > application/activity+json-ish things, > telling a server that you expect a certain vocabulary / profile. > So I'd expect a client to say > Accept: text/turtle;profile="http://example.org/activity"; > instead. > > >>> The Web Annotation WG decided /not/ to do this, and to only use a > profile. So the media type for the JSON-LD annotation serialization is: > application/ld+json;profile="http://www.w3.org/ns/anno.jsonld"; > > I love it; that's the most scalable way to do this, > and it does justice to what RDF stands for: > the interpretation is inside of the message. > > How is it actually implemented? > It should be sufficient that the client does > Accept: application/ld+json;profile="http://www.w3.org/ns/anno.jsonld"; > and the server only says > Content-Type: application/ld+json > since the rest would be in the message. > > Actually, the client should also be able to say > Accept: text/turtle;profile="http://www.w3.org/ns/anno.jsonld"; > to which the server would reply with > Content-Type: text/turtle > but I guess the notion of "profile" (RFC 6906) > here is to be interpreted more strictly as "JSON-LD profile". > I would be interesting to consider > Accept: application/ld+json;profile="http://www.w3.org/ns/anno"; > without any extension, allowing content negotiation on the profile. > > Best, > > Ruben > > PS Wrote more about that here: > – http://ruben.verborgh.org/phd/ruben-verborgh-phd.pdf#page=103 > – http://ruben.verborgh.org/blog/2015/10/06/turtles-all-the-way-down/ >
