I'd also add this particular answer from the RFC3023 FAQ, which is 
relevant to what you guys are talking about, as far as assuming anything 
about the relationship between content type "something" and content type 
"something+xml".

A.13 What is the semantic difference between application/foo and
      application/foo+xml?

    MIME processors that are unaware of XML will treat the '+xml' suffix
    as completely opaque, so it is essential that no extra semantics be
    assigned to its presence.  Therefore, application/foo and
    application/foo+xml SHOULD be treated as completely independent media
    types.  Although, for example, text/calendar+xml could be an XML
    version of text/calendar[RFC2445], it is possible that this
    (hypothetical) new media type would include new semantics as well as
    new syntax, and in any case, there would be many applications that
    support text/calendar but had not yet been upgraded to support
    text/calendar+xml.


So.  "application/sru+xml" needs to be registered to be used legally. 
Even if you already had "application/sru" registered, you'd need to have 
registered "application/sru+xml" too, to use it legally.

By standard, anything ending in "+xml" should be valid XML. So you would 
probably be stopped from registering something ending in "+xml" that 
wasn't intended to be.

Beyond that, the nature of what "application/sru+xml" is, is up to your 
own documentation for the content type. So when you registered 
"application/sru+xml", the standard that defines it may have said "it 
will return something that is valid under the SRU Response XML schema" 
-- then that's what makes this so. Not just any implication or 
assumption of the relationship between "application/sru" and 
"application/sru+xml".

I don't think there's any standard about "+json". But you could still 
register a type for "application/sru+json", and the standard that 
establishes the meaning of that registration could say it will be JSON, 
and would be expected to define the particular nature of that JSON. 
Which you could do using JSON schema, or plain English language, or 
whatever other means you wanted; there is nothing that would require it 
to be anything in particular, that'd be up to you standardizing it.

Jonathan

On 2/19/14 11:54 AM, Jonathan Rochkind wrote:
> No.
>
> So, according to the relevant standards, all content types need to be
> registered, you don't just get to make them up -- unless you use some of
> the prefixes for experimental/vendor/unregistered content types.
>
> Not even if they end in "+xml" do you get to make them up.
>
> There is, however, an established convention in the standard that
> content types ending in "+xml" are XML
>
> There is no equivalent established standard for content types ending in
> "+json", even if you think it is "sort of implied."
>
> The "+xml" convention is standardized in this RFC:
> http://tools.ietf.org/search/rfc3023
>
> I don't believe a "+json" convention is standardized anywhere, although
> I don't think there's anything stopping you from ending any content type
> you want in "+json" ("+" is a legal char for content types in general, I
> think?), and I can see how it would be a helpful hint for developers.
>
> You would still need to register your content type, or use one of the
> vendor/experimental/private-use prefix standards, as you would for any
> new content type, ending in "+xml", "+json", or anything else. Or at
> least, you'd need to do that unless you are okay with violating the
> standards.
>
> I do encourage looking at RFC3023 to get a sense of how all this stuff
> works, especially it's FAQ section at the end.
>
> Jonathan
>
> On 2/19/14 10:57 AM, Ray Denenberg wrote:
>> *From:*LeVan,Ralph
>>
>>  > Application/sru+xml means SRU returned as XML.
>>
>> It means more than that. it means that the response must be as defined
>> by the SRU Response schema:
>>
>> http://docs.oasis-open.org/search-ws/searchRetrieve/v1.0/os/schemas/sruResponse.xsd
>>
>>
>>
>> That sort of implies
>>
>>  > that application/sru+json is meaningful.
>>
>> Not unless (or until) there is an analogous json schema.
>>
>>  > The HTTP Accept header or the httpAccept parameter take care of the
>>
>>  > response definition, possibly supplemented with the responseType
>>
>>  > parameter if a sufficiently complicated media subtype can’t be agreed
>>
>>  > upon. (In the example given below of an Atom response with SRU
>>
>>  > semantics, it was suggested that the media type would be
>>
>>  > application/atom+xml combined with a response type (essentially) of
>> SRU.
>>
>> No, it's saying more than just "atom expressing SRU", it is saying there
>> is a (fictitious) definition (which may be a schema, or some other form
>> of definition) whose identifier is *info:srw/1/response-type/atom, * to
>> be used in conjunction with atom, to form a complete SRU response.
>>
>>  > To my mind what that means is that we are refining the Atom semantics
>>
>>  > with SRU semantics. I’d probably fudge the whole thing as
>>
>>  > application/atom.sru+xml, realizing that no one has agreed to the
>>
>>  > semantics of that media type. :)
>>
>> By that approach we either (a) fudge non-standard media types, or (b)
>> try to standardize the media types we want.   I don't like either
>> possibility.
>>
>>  > A similar pattern then applies for the record.  I need to specify a
>>
>>  > media type with semantics and syntax combined.
>> Application/marc21+xml,
>>
>>  > application/vnd.viaf+xml, application/vnd.viaf+json.  The current
>>
>>  > solution is to provide an opaque URI in the recordSchema parameter (or
>>
>>  > a short name for that URI as defined in the Explain record) that
>>
>>  > magically defines semantics and syntax.
>>
>> And what's wrong with the latter?
>>
>>  > I think I prefer a more explicit use of an Internet media type.  It is
>>
>>  > subject to standard rules about negotiation with the client and server
>>
>>  > both able to say something about the ranking of preferences.  The
>>
>>  > recordSchema element in the record response allows the server to tell
>>
>>  > the client what media type was selected for the response.
>>
>> I think you meant to say "selected for the record"?  But no, the server
>> doesn't select a record syntax, that value is echoed in the response,
>> redundantly. The server cannot select an alternative syntax.  At the
>> (broader) response level, it's different. The client has the option of
>> content negotiation, as an alternative to using httpAccept, and, this
>> way to do it (allowing both mechanisms)  was chosen (and as I recall, it
>> was your idea) for two reasons (1) we wanted to allow content
>> negotiation, (2) some clients are not capable of content negotiation.
>>
>> But my point, I think, is that content negotiation is meaningful at the
>> response level, but doesn't really make a whole lot of sense at the
>> record level.
>>
>>  >     The critical standards issue here is that the client needs
>>
>>  > to be able to request something *not* on our approved list of options
>>
>> I don't follow, could you elaborate?
>>
>>  >
>>
>>  > recordXMLEscaping (formerly recordPacking) is a specific example of a
>>
>>  > generic problem.  I might like to escape my json records in my json
>>
>>  > response so that they don’t get parsed as part of the entire payload.
>>
>>  > More importantly, I *probably* need to escape my JSON records in my
>> XML
>>
>>  > response.  I think we need to reconsider recordPacking.
>>
>> Yes, I do see the point here. I'm sure it will be on the table for 2.1.
>>
>> Thanks Ralph.
>>
>> Ray
>>
>>
>> *Sent:* Tuesday, February 18, 2014 4:21 PM
>> *To:* [log in to unmask]
>> *Subject:* Re: SRU recordPacking and JSON
>>