[occi-wg] JSON Rendering

[Ralf Nyren]

On Thu, 5 Apr 2012 09:21:52 +0000, "Feldhaus, Florian"
<florian.feldhaus at gwdg.de> wrote:
> Hi,
> 
> how do we proceed?

The best thing IMO would be to create a version 1.2 of the HTTP Rendering
doc and update it so that it is a clear separation between Protocol and
Data Format. The existing text/occi, text/occi and the JSON data formats
would then be pluggable modules to this spec.

The quick way is to continue writing the JSON rendering as a standalone
HTTP-based OCCI rendering which happens to be quite similar to the HTTP
Rendering. Saves time but causes lots of duplication.

> Following a some responses to your comments:
> 
> Am 04.04.2012 um 10:59 schrieb Ralf Nyren:
> 
>> On Tue, 03 Apr 2012 14:31:24 +0200, Feldhaus, Florian
>> <florian.feldhaus at gwdg.de> wrote:
>> 
>>> Hi,
>>> 
>>> once again I would like to reiterate the JSON rendering. First a short
>>> overview what Alexander and I think are the main points we should
>>> address:
>>> - remove all HTTP Rendering specific parts from the JSON rendering
>> 
>> Remember that an OCCI rendering (as currently specified) includes
_both_
>> protocol and data format at the moment.
> 
> That's only partly true. A "pure" JSON rendering can already exist
> independently from the HTTP Rendering without any trouble in rendering. 

No. Yes.

Probably a misunderstanding here. An OCCI Rendering is defined as a way to
manipulate the Core Model. So in theory you could have 2 different
HTTP-based OCCI Renderings with different semantics where one happen to be
using XML as the data format and the othe JSON for example. This is not
nice but within the definition.

So to be complete an OCCI Rendering must both define the protocol and
whatever data format is used by that protocol.

This does not prevent us from having a single OCCI HTTP Protocol Rendering
with pluggable data formats.
 
>> I like the idea of having a common HTTP Protocol rendering spec which
>> the JSON rendering could be built upon though.
> 
> I second this and would like to move forward. Any comments on the best
> strategy? Do we need to create a version 1.2 for the HTTP rendering?

I believe so yes. It would be mostly backwards compatible though.

>> However, the current HTTP rendering doc lacks things like e.g. request
>> parameters in URL which I would say is necessary to have a sane JSON
>> rendering.
> 
> I don't think so. The rendering should be independently of the transport
> protocol. If I ask your server to send me a file by mail containing the
> JSON rendering of all resources, that should work as well. 

We are probably using different terminology here. I am referring to an
OCCI "rendering". Your statement is 100% true for an OCCI data format.
However, a data format is not enough to create an OCCI Rendering.
  
>>> - consider using RFC 5988 "Web Linking" for collection information
>>> (e.g. index, next, previous,…)
>> 
>> Gary and I had an email conversation which resulted in a solution where
>> all info necessary for pagination would end up in the request URL. I.e.
>> basically eliminating the need for using special headers (such as RFC
>> 5988).
>> 
>> The request parameters to a collection simply allow you to specify the
>> amount of resource instances you want returned either _before_ or
_after_
>> a specific occi.core.id.
> 
> Do you have some examples? IMHO this should go to the revised HTTP
> rendering document.

The mail thread was on occi-wg so should be in the archives.

Agree that it would be best to put this into a 1.2 version of the HTTP
Rendering doc.
 
>>> Examples:
>>> http://pastebin.com/ZK9Uf0K1 (Entities)
>> 
>> Nice, I like that you keep the "attributes" hash now ;)
>> 
>> How do you render Link attributes for the links tied to an OCCI
Resource?
> 
> In the example you can see Links are rendered as a hash containing href
> and kind. The only really necessary part is the href location.
Everything
> else is optional and could be retrieved by the client using separate
HTTP
> requests. It would also be possible to omit the hash and just render all
> link hrefs in an array. To allow for a slim rendering and also allow for
> additional information to be send to the client, I would suggest that we
> specify a hash with at least the href and optional all other parameters
> valid for the link. We could even go so far as to use the link rendering
> for rendering link attributes within resources.

After many long discussions it was decided to have inline rendering of
Link attributes in the OCCI HTTP Rendering. I think the same should apply
to JSON.

>>> To keep the mail short, a detailed discussion can be found in the
>>> attached text document.
>> 
>> Just picking out one thread:
>> 
>>> resources and links should be represented differently. The entry
>>> "links" is unique for resources and the entries "target" and "source"
>>> are unique for links.
>> 
>> Sounds goods. So the top-level hash of the collection format would have
>> one hash "resources" and another hash "links" then?
>> I mean, we still have to cover the case where the client asks for
>> "everything" at the top-level URL and thus gets both Resources and
Links
>> in the response.
> 
> I would suggest to have a content-type for entities. It should contain a
> hash with "links" and "resources". Both then are arrays of hashes.

I rather have a single array, it plays much better with the collection
concept of REST.

>>> In OCCI Core the attribute names should be changed from
>>> occi.core.source and occi.core.target to just source and target, as
both
>>> are representing connections to other resources from within the OCCI
>>> model (similar to links in resources, or kind in entity).
>> 
>> occi.core.source and occi.core.target was named simply source/target up
>> until just before the OCCI HTTP Rendering doc was published.
>> 
>> The fundamental problem here is that we have two different sorts of
>> "attributes".
>> 1. Attributes as part of the OCCI Core model. These include both
>> Entity.id, Entity.title, Resource.summary, Resource.links, Link.target,
>> Link.source, etc
>> 2. Attributes as exposed by an OCCI rendering. The HTTP Rendering
>> exposes id, title, summary as attributes as well as target and source.
>> However the Resource.links attribute is not exposed as an attribute...
>> 
>> There is no clear distinction here which IMO leads to confusion.

No comments on the above?
 
>> Also remember that a subclass of OCCI Link may have Link.target
pointing
>> at some arbitrary external object.
> 
> In my opinion, source and target always point to resources, even if they
> lie outside the OCCI model. They contain complex data types like kind or
> mixin and not primitive data types like id, title or summary.

So, to link to a VNC console you would have the vnc:// URL where? In a VNC
console Resource object?

/Ralf