5

We are looking for some advice on handling URLs (and the state related to each URL) in a web application backed by a HATEOAS REST API, more specifically on

  • how to avoid having the web application URLs coupled with the REST API URLs
  • how to handle multiple resources in a single view

But let me first provide some more context:

We are building an Angular web application on top of a REST layer with Hypermedia constraint. (Note: I prefer simply using the term 'Hypermedia (constraint)' over HATEOAS).

As dictated by the Hypermedia constraint, the available actions and links in the application at any point in time are provided by the REST API. So the web application should not contain any hardcoded urls of the REST API, except for the 'root' (assuming that concept really exists in a REST API).

On the other hand, each page in the web application needs to be bookmarkable. So we cannot create a black-box application (with a single url and all state changes handled in the SPA without changing the URL). This means the web application also has its URL space, which needs somehow to be mapped to the REST API URL space. Which is already a conflict with the Hypermedia idea.

In the Angular application we use UI Router for handling application state. Here is how we got it working:

  • We only define states, no URLS
  • We defined a $urlRouterProvider.otherwise handler that will map the current web application URL to the corrsponding REST API URL, retrieve the representation of the resource that corresponds with that REST URL and pass it to the controller (in $stateParams).
  • The controller can then use the data (and links and actions) in the representation, just like it would if it would have made the REST call itself (or through a service)

So far so good (or not really) because there are some downsides on this approach:

  • The Web application URLs are mapped to the REST API URLs, so both URL spaces are coupled, which conflicts with one of the basic assumptions of using Hypermedia constraint: we cannot change the REST API URLs without having to change the web application.
  • In the $urlRouterProvider.otherwise handler we retrieve the representation of the current web app URL. But in some cases we have two resources in a single view (using UI Router nested states): for example a list of items and a detail of a single item. But there is only a single URL, so only the representation of the item detail is retrieved and the list of items remains empty.

So we would love to hear some suggestions on how we could improve on our approach in handling the two URL spaces. Is there a better way to make the REST API dictate the (available) behaviour of the web application and still have bookmarkable URLs in the webapplication? Because now we have some kind of hybrid approach that does not feel completely right.

Thanks in advance.

Regards,

Luc

Luc Feys
  • 127
  • 1
  • 9

1 Answers1

2

that's a tough setup. Roughly you want bookmarks into your API, and RESTful systems somewhat discourage bookmarks.

One possible solution is a "bookmark service" that returns bookmark (bit.ly like) urls for the current resource being presents that are guaranteed to be fowards compatible because as you may change the canonical url structure, the bookmark service can always translate the bit.ly like url into the canonical url. Sounds complicated, but we see this all the time and we call them SEO urls eg: /product-name/ maps to products/ today, but may be /catalog/old-products/ tomorrow.

How you match that up to a UI that shows 2 resources the first being a list of summary like resources, and the second being a specific resource get's really tricky. I would expect such a page to know contain the state of what's it's displaying in it's url (probably in the fragment). As such since it's [likely] the controller that processing such commands it probably needs both (the list resource and the expanded resource) as input. I bet the url would look something like:

list=http://path/to/list/results&expand=http://self/link/of/path

So the last part you have is to make sure that works going forwards. Again this is the bookmark problem. What i may suggest if you don't want to build a bookmark service is that given you want to have such bookmarks you need to transition people to the new URLs. When a request is made to http://path/to/list/results and you want to switch that over you should be 301 redirecting them to the new canonical url and the app should be updating the bookmark. such a redirect can include the &flag=deprecate_message param to trigger the presentation in the UI that the client's bookmark is old and should be replaced. Alternatively the response can be internally forwarded and the deprecation flag & canonical (or latest) link included in the response to the old URL. This causes a phased transition.

In summary: I have yet to see HATEOAS be a cure all for backwards & forwards compatibility, but it's much better than the existing techniques. that said you must still make decisions in v1 of your API about how you want your users to move to v2.

Chris DaMour
  • 3,650
  • 28
  • 37
  • Thans for taking the time to come up with some good proposals for this pretty complex setup. I like the idea of the bookmark service. It seems like a good solution for decoupling the UI url space from the REST url space. With regard to the 2 resources in a single view, we were able to solve it by adding a link from the detail resource to its parent list resource. A 'pure' hypermedia solution providing the results we needed. – Luc Feys Jun 29 '15 at 10:55
  • All good points by Chris. I know this is a dated post but for future SO searchers I would add that (in addition to the above approaches) you should look into returning a 'bookmark' relation type when returning your list of resource links. See http://www.w3.org/TR/html5/links.html#link-type-bookmark for more info. – Wade Bee Jun 20 '17 at 18:18