On Fri, Apr 12, 2013 at 1:24 PM, Chris Geer <[email protected]> wrote: > On Fri, Apr 12, 2013 at 10:19 AM, Erin Noe-Payne > <[email protected]>wrote: > >> On Fri, Apr 12, 2013 at 1:02 PM, Chris Geer <[email protected]> wrote: >> > On Fri, Apr 12, 2013 at 9:55 AM, Erin Noe-Payne < >> [email protected]>wrote: >> > >> >> Hey all, I've pushed the first couple commits to the angular branch >> >> with some extremely basic features in place. I want to start a >> >> discussion to refine our vision for the portal application and keep >> >> everyone on the same page. >> >> >> >> To preview the work so far: >> >> - Check out from >> https://svn.apache.org/repos/asf/rave/branches/angular/ >> >> - Spin up rave >> >> - Hit the url http://localhost:8080/portal/app/angular/portal >> >> >> >> You should see some tabs that you can navigate between, some rendered >> >> widgets. Very little else is working at this point. >> >> >> >> The proposal: >> >> - An implementer should be able to define any custom context that they >> >> want to present through the rave portal application. This corresponds >> >> to the context as we discussed in the pages api [1]. Currently rave >> >> ships with "portal" and "profile" contexts, and that's what I will be >> >> building out. >> >> >> >> - Each context gets its own angular 'single-page' web application. >> >> Moving within a context (IE /profile/erin -> /profile/matt) is all >> >> client side routing & ajax calls. Moving between contexts (/profile -> >> >> /portal) is a full page reload and entirely new angular webapp is >> >> served. The reason for this structure is that each context will want >> >> its own display (markup & css), its own routing rules, etc. >> >> >> >> - The contexts are served from one generic endpoint. Right now this is >> >> /portal/app/angular/{context}/** to avoid collision with other >> >> endpoints. Eventually I see this as moving to root and replacing most >> >> of our current application endpoints. See >> >> org.apache.rave.portal.web.controller.AngularController for the basic >> >> implementation. The idea is that a call to the context endpoint will >> >> always render the same basic view that imports the corresponding >> >> context's markup and angular js app, and that app then handles any >> >> further navigation / client side routing / importing of appropriate >> >> resources. >> >> >> >> - In this way, the implementation of a context is entirely in static >> >> files - html, css, js. If an implementer wants to add a new context >> >> (say portfolio), they only need to create the new static files to >> >> support that context. This means that a new context can be custom >> >> built from the ground up without having to overlay and with complete >> >> flexibility. However... >> >> >> >> - We can still write and provide reusable components. View partials >> >> can be imported using angular's ng-include blocks, common services can >> >> be written as angular modules. >> >> >> >> [1] >> http://mail-archives.apache.org/mod_mbox/rave-dev/201303.mbox/browser >> > >> > >> > I look forward to trying it out. Out of curiosity, have you put any >> thought >> > into how security will work? For example, can I restrict people to >> > particular contexts? How will that work client side? >> > >> >> Definitely. So from the server side perspective we can continue to use >> spring or whatever other security provider we want. We could force >> someone to login before they can hit the {context} endpoint at all - >> you'll see that is the case now but I don't really have an opinion on >> that. I think where you put your security restrictions is on the api >> endpoints that deliver data. >> >> Then from the application perspective, any angular webapp that loads >> in a particular context will need to make api calls to get data. You >> can then write an http interceptor so that for any call that is >> intercepted with a 401, some action is taken. There is a simple >> example of this in the code right now in >> script/angular-portal/routing.js lines 22- 41 (note you can't actually >> see it in action because our endpoints don't return 401, and you have >> to be logged in to see the context endpoint at all). This is a simple >> implementation that assumes that if you receive a 401 you are simply >> not logged in, and get redirected to a login page. But you can easily >> take a more granular approach, and we can provide this as a pluggable >> authentication service that each context webapp can configure and use >> as they see fit. >> > > Ya, I'm curious to know how to handle it client side, I'm good with server > side. Now that the server isn't rendering the pages, the client has to be > aware of permissions so it can show/hide the correct information. For > example, if a user doesn't have permission to view a certain page, the link > should even be an option. That means there needs to be a way to have the UI > get a list of all the permissions a user has and take those into > consideration. I know there are ways to do it, just curious if you've put > anything in place. Honestly right now the backend isn't really setup to > handle that though. We need a more flexible permissions framework probably. > > We have the same problem in our system so on page load we cache all the > permissions a user has client side and then the JS can use that list to > make determinations. >
This is a good question and honestly I only have a vague idea of what it will look like. I think in this case your auth will need to be a rest service. It will probably end up being part of the users api and look something like the pages api with /api/users/@self. >> >> > Chris >>
