On Wed, Mar 27, 2013 at 2:58 PM, Chris Geer <[email protected]> wrote:
> On Wed, Mar 27, 2013 at 11:50 AM, Matt Franklin <[email protected] > >wrote: > > > On Wed, Mar 27, 2013 at 2:32 PM, Chris Geer <[email protected]> > wrote: > > > On Wed, Mar 27, 2013 at 11:27 AM, Matt Franklin < > > [email protected]>wrote: > > > > > >> On Wed, Mar 27, 2013 at 1:59 PM, Chris Geer <[email protected]> > > wrote: > > >> > On Wed, Mar 27, 2013 at 10:42 AM, Erin Noe-Payne > > >> > <[email protected]>wrote: > > >> > > > >> >> On Wed, Mar 27, 2013 at 1:16 PM, Matt Franklin < > > >> [email protected] > > >> >> >wrote: > > >> >> > > >> >> > On Thu, Mar 21, 2013 at 5:32 PM, Chris Geer < > [email protected] > > > > > >> >> wrote: > > >> >> > > I've done a first cut at adding some new CXF based REST web > > services > > >> >> > which > > >> >> > > use a different data model > > >> >> > > > >> >> > As part of RAVE-924, I have created a new page model for web. > As I > > >> >> > was building it, it occurred to me that there are a couple of > > >> >> > different ways we will want/need to use the REST interface for > > Page: > > >> >> > > > >> >> > 1) As an export mechanism > > >> >> > 2) As an OMDL export mechanism > > >> >> > 3) As an entry point for applications who want to render widgets > > >> >> > (including the portal) > > >> >> > > > >> >> > IMO, #1 is straight forward. For number 2, I was thinking that > it > > >> >> > would be better if there was an OMDL mime type so the logical > > mapping > > >> >> > remains the same (/api/pages/{id}) as in the regular export. > What > > >> >> > does everyone think about using application/vnd.omdl+xml? > > >> >> > > > >> >> > > >> >> +1 here, I think mime type is the right approach. I have no opinion > > on > > >> the > > >> >> actual label of the mime type - that looks fine. > > >> >> > > >> > > > >> > +1 > > >> > > > >> >> > > >> >> > > >> >> > > > >> >> > The hard part is how to deal with rendering. For this "mode" we > > will > > >> >> > need to export the Wookie iFrame URL (which is per-user), the > > >> >> > openSocial security token and the OpenSocial metadata. These > > require > > >> >> > a User to be authenticated and should not be exposed across > things > > >> >> > like the OMDL or Page export. What would everyone think about > the > > >> >> > following url for this case?: /api/pages/{id}?render=true > > >> >> > > > >> >> > > >> >> My gut reaction is that I dont like having a query string parameter > > to > > >> >> dictate this. It seems like it would be better for the endpoint to > be > > >> >> context aware somehow. Possibly check the request and deliver the > > >> >> render-ready data set if its coming from an authenticated rave > user? > > I > > >> need > > >> >> to think about that a bit more... > > >> >> > > >> > > > >> > I agree with Erin, not a fan of the query string. I'd rather see > > >> > /api/pages/{id}/rendered or something similar. > I also do not like /api/pages/{id}/rendered. Url parts should be semantically representative of resources, not of state or action. I'm now thinking... /api/pages/0 - this is a representation of the generic page for export /api/users/erin/pages/0 - this is a representation of the page in the context of user erin, with user-specific data needed for rendering attached. > > >> > > > >> > Is the intent to return the wookie and OS stuff in the same > response? > > I'm > > >> > not a fan of that, especially considering some people won't have > > wookie > > >> > installed at all (or OS). > > >> > > >> The approach I was going to take is to inject all the providers in the > > >> current context into a service and when the render condition is hit, > > >> have the provider return an object that extends the 'core' > > >> RegionWidget with its own properties. This way, there is no coupling > > >> to any specific provider. You would be able to have 0...n providers. > > >> > > >> >Maybe it makes more sense to have a different web > > >> > service for rendering by each provider??? > > >> > > >> This will end up causing serious performance bottlenecks in an already > > >> taxed system. > > >> > > > > > > I don't understand this comment. How would this cause serious > performance > > > bottlenecks? Having additional services won't cause any problems, and > > they > > > should only be called when they are used which would be no more/less > than > > > if it was a single service. I'm ok not doing this, just not sure what > > would > > > cause the major performance problems you are referring to. > > > > I should have been more clear. As we move away from server-side > > templating to client-side MVC to deliver the OOTB interface, these > > services will be used by the framework we have running in the browser. > > If it has to make AJAX calls for each widget to get the necessary > > information to render the widget, we are going to end up with a bunch > > of extra AJAX calls in order to initiate rendering of the the widgets > > on a page. Since widgets are already iFrames and require their own > > set of round trips to the widget provider, we now end up in a > > situation we have even more network requests to services. > > > > My thought in returning this information as part of the initial page > > REST call is that we can eliminate the extra round trips to the server > > to get the provider representation of the widget. > > > > That makes sense. The only thing I want to make sure we can do dynamically > add (reload) a gadget to a page without re-rendering the whole page. > Absolutely agree. This approach is in line with that goal. The basic mechanism would be that the page loads with it's initial state (bootstrapped data we are discussing now). If a user adds a new widget then the client side state is updated and appropriate rendering happens, as well as a post to server to update server side state; no page reload happens. > > > > > > > > >> > > >> >Otherwise this "core" service has > > >> > to know about all the different providers which can be a problem > > moving > > >> > forward. > > >> > > >> It shouldn't have to know anything about any specific provider at all. > > >> > > > > > > ok > > > > > >> > > >> > > > >> >> > > >> >> > > >> >> > > > >> >> > Since we want to be able to "bootstrap" the client MVC framework > > with > > >> >> > a pre-fetched & serialized version of the "Page" we will also > need > > to > > >> >> > do the same translation between what is currently returned from > the > > >> >> > service layer in both the server MVC and the REST API; which > raises > > >> >> > the question as to wether or not we just abstract all of that > > >> >> > functionality in the service layer and only expose the "web" > model > > >> >> > from that layer or do we create yet another layer to translate > from > > >> >> > the current service to the web model? > > >> >> > > > >> >> > > >> >> +0 from me. The work should definitely be happening in some service > > >> layer > > >> >> and keep the controllers light. Beyond that I'm not sure what will > > end > > >> up > > >> >> being cleaner. > > >> >> > > >> >> > > >> >> > > > >> >> > Thoughts? > > >> >> > > > >> >> > > >> > > >
