Loading ...
Sorry, an error occurred while loading the content.

1017Re: [json] Re: MIME types

Expand Messages
  • Kris Zyp
    May 6, 2008
    • 0 Attachment
      Thank you, that RFC is very helpful, and they definitely make a strong case for option #3 instead of #4. One of the most important influences in the decision is in how you want MIME dispatchers built on existing infrastructure to respond to different JSON types. If you think it would be best for all JSON content to be handled by one application which would then have different internal logic for different subtypes, you would may want some variant on option #4 (although there are additional complications with getting the additional info through in some cases), but if different applications should handle the different types in MIME dispatching infrastructures, we should probably use #3. This is kind of a difficult analysis to use since I believe most JSON applications probably don't even use MIME dispatching architectures (like XHR).

      Anyway, based on that RFC I would be inclined to think option #3 (application/subtype+json) is probably the best bet for things like JSON-RPC, JsonML, JSON Schema, etc. There may be some situations where MIME parameters may be helpful in denoting extra features, for example using JSON referencing would probably be best defined as a MIME parameter (especially since it is applicable to multiple subtypes), and explicitly stating how dates are represented may be something to define in a MIME parameter as well.

      Thanks,
      Kris
      ps
      With great trepidation that I might spark another superset rabbit trail, it is also interesting to note the obvious parallel alternate syntax MIME types that correspond to these types. For example a browser might well say
      Accept: x-application/jsonml+javascript-expr, x-application/jsonml+json
      to indicate that it can accept more than just JSON (comments, dates, etc). However, I want even bother to ask if this seems reasonable lest others think I am suggesting that extra features be added to JSON :).


      ----- Original Message -----
      From: Stephen M. McKamey
      To: json@yahoogroups.com
      Sent: Tuesday, May 06, 2008 9:03 AM
      Subject: [json] Re: MIME types


      I would also support a #4 as you've described as long as whatever it
      is works within the IANA restraints. It has been a while now since I
      sifted through the RFCs but it might be a good time to revisit. The
      commentary which influenced my initial stance was in the XML MIME
      Type RFC:

      http://tools.ietf.org/html/rfc3023#page-32

      Definitely worth a read as they've justified the reasoning for XML's
      usage of the "+xml" style. They also discuss some alternatives.

      Thanks,
      smm

      --- In json@yahoogroups.com, "Kris Zyp" <kriszyp@...> wrote:
      >
      > Thank you Stephen, this exactly what I was asking! Thanks for
      clearly articulating this. I was also wondering if there is an option
      #4, which may be an enhancement of #1, where we also use
      application/json, but we include extra info after a semicolon to
      denote structure/application information. JsonML might be:
      > application/json; type=jsonml
      >
      > Or something like that. I believe there other types that use
      something like this to include extra info; multipart includes
      boundary definition after a semicolon, and I have seen charset info
      included after a semicolon. However, I am not sure what the semantic
      intents and constraints are for this approach, it may be
      inappropriate to define subtypes in this manner. I really am fine
      with any of these approaches and just wanted to get some consensus. I
      am perfectly content to follow your (Stephen) precedent with {x-}
      application/subtype+json.
      >
      > Thanks,
      > Kris
      >
      >
      > ----- Original Message -----
      > From: Stephen M. McKamey
      > To: json@yahoogroups.com
      > Sent: Tuesday, May 06, 2008 8:17 AM
      > Subject: [json] Re: MIME types
      >
      >
      > No offense, but I think the replies thus far have missed Kris'
      > point. Kris is trying to get opinions on additional MIME types
      for
      > specific applications of JSON. These aren't necessarily messaging
      > envelopes and are still platform-agnostic. These are situations
      > where JSON is the format but the structure is important. Examples
      > from the XML world:
      >
      > - ATOM: "application/atom+xml"
      > - XHTML: "application/xhtml+xml"
      > - XSLT: "application/xslt+xml"
      > - SVG: "image/svg+xml"
      >
      > As you can see, these are all valid XML formats but it makes
      sense
      > due to their profileration to give them a MIME type that allows
      > better assumption of a subset of XML. I wouldn't want to
      interpret
      > SVG as an XHTML document and such it is nice to be able to
      > communicate between client-server the "types" that are accepted.
      >
      > As I see it, opinions on this will more or less fall into three
      main
      > camps:
      >
      > 1) avoid MIME variants at all costs
      > 2) standardize and use for every standardized variant
      > 3) standardize and use sparingly where justified
      >
      > There is the strict purist/minimalist perspective which would
      suggest
      > #1. This has the advantage of remaining simple, pure and keeps
      JSON
      > as just JSON. I can respect that. And judging by Kris' question
      he
      > is probably okay with that as well if that is the group concensus.
      >
      > There are circumstances that #2 or #3 would be nice however. For
      > instance when I make a request for JsonML, it is nice to be able
      to
      > inform the server that as a client I can accept a JsonML response
      > (and in response that the format is actually JsonML and not just
      JSON
      > data which must be processed differently). This is a more
      specific
      > requirement than just being able to parse JSON. Essentially JSON
      > data needs to be in a particular structure or it doesn't make
      sense
      > to a JsonML builder.
      >
      > Now that being said, the risk of opening this Pandora's Box is
      that
      > every developer and their mom might begin creating MIME types for
      > every piece of ad hoc JSON data. Granted if we went with #3, we
      > could just rely on IANA to police the JSON MIME type horizon. As
      it
      > is, in order to officially have a "application/XXXX+json" MIME,
      you
      > have to have a formalized standard and they have to accept your
      > application. See [ http://www.iana.org/assignments/media-types/ ]
      > for more info.
      >
      > My only real fear is that if we don't pick a standard of some
      kind
      > then we run into the issues where incompatibility makes the
      feature
      > useless. Initially JavaScript itself suffered from this issue
      until
      > it was later better standardized (see RFC-4329) but this caused
      > confusion and incompatibility for years.
      >
      > My suggestion would be option #3 with IANA policing the
      > official "application/XXXX+json" MIME space, but also some
      standard
      > set in place for "X-" MIME types. For instance a JsonML MIME type
      > isn't official, so technically it should be
      > "x-application/jsonml+json" rather
      than "application/jsonml+json".
      > This is the official way of "experimentally" creating a MIME type.
      >
      > I hope this makes sense and that we can come to some sort of
      > consensus.
      >
      > smm
      >
      > --- In json@yahoogroups.com, "kriszyp" wrote:
      > >
      > > I wanted to see if there was any prevailing consensus on if and
      how
      > > JSON subsets and alternatives should be denoted with MIME
      types.
      > There
      > > are a growing number of JSON subsets like JSON-RPC, JSONT,
      JsonML,
      > > JSON Schema, etc. There may be benefits in being able to define
      > which
      > > subset is begin used as a MIME type. XML has used a subtype+xml
      > > convention, and I noticed that Stephen McKamey is using this
      > precedent
      > > to suggest denoting JsonML as "application/jsonml+json". I am
      > > perfectly fine with using this convention, I just wanted to see
      if
      > > others felt this was the way to go, or if it is preferable to
      do
      > more
      > > "minor" subtyping that would not suggest a need for IANA
      > registration
      > > like "application/json; rpc", or something like that.
      > >
      > > I am also curious if there any thoughts about JSON supersets.
      There
      > > have been little efforts to create any formal proposals in this
      > area,
      > > but I don't think it is totally uncommon to use additional
      > JavaScript
      > > constructs for JSON-esque data, and postel's law application to
      JSON
      > > almost invariably means allowing for additional JavaScript
      defined
      > > flexibility (non-quoted keys, comments, etc). One could think of
      > > JavaScript expressions as a type unto themselves, the true
      superset
      > of
      > > JSON (JavaScript itself is not actually a correct superset of
      JSON
      > > because JSON is not valid JavaScript by itself, it must be
      enclosed
      > in
      > > paranthesis to make it a JavaScript expression). Generally
      > JavaScript
      > > libraries that don't do validation before an eval (Dojo and many
      > > others) are actually full JavaScript expression acceptors, and
      > > non-JavaScript libraries and JavaScript libraries that do
      validation
      > > (like json.org's library) would be only JSON acceptors. I am
      > wondering
      > > if MIME type definition might be valuable here as well
      (providers
      > > would know if they can include comments, for example).
      > >
      > > Anyway, I am not really making any proposals, just curious it
      there
      > > are prevailing feelings towards MIME types for defining
      different
      > JSON
      > > formats.
      > >
      > > Thanks,
      > > Kris





      [Non-text portions of this message have been removed]
    • Show all 9 messages in this topic