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

Re: [rest-discuss] URI design, part 2

Expand Messages
  • Jan Algermissen
    ... Maybe of interest: http://tech.groups.yahoo.com/group/rest-discuss/message/18998 jan
    Message 1 of 28 , Dec 4, 2012
      On Dec 4, 2012, at 11:40 PM, Max Toro <maxtoroq@...> wrote:

      > Thank you very much for your responses Eric.
      > The short answer is (if I understood correctly): POST /cancel is not
      > REST because it lacks visibility, since it's not possible to
      > understand the client's intent by examining the request.

      Maybe of interest: http://tech.groups.yahoo.com/group/rest-discuss/message/18998


      > Never thought about HTTP/URI resource vs. REST resource, you say that
      > a URI that doesn't implement GET is probably (or always) not REST.
      > --
      > Max Toro
      > On Mon, Dec 3, 2012 at 1:17 PM, Eric J. Bowman <eric@...> wrote:
      > > Max Toro wrote:
      > >>
      > >> > Well, what are you expecting to GET from /cancel, or are you just
      > >> > using that URL to invoke a procedure? If so, then there are a few
      > >> > places Roy's thesis admonishes against it in Chapter 6 -- the rest
      > >> > of REST is about positive, rather than negative, reinforcement of
      > >> > the identification of resources constraint. Suggested reading:
      > >> > 6.5.2; 6.2.1, in particular: "REST [defines] a resource to be the
      > >> > semantics of what the author intends to identify."
      > >>
      > >> To clarify, /orders/1/cancel is used to modify a resource, using POST.
      > >> A GET request would result in a Method Not Allowed response.
      > >>
      > >
      > > Yes, such examples are out there, but in them, /cancel is NOT a resource
      > > in the REST sense so they must be some style of RPC... REST, not so
      > > much. Representational State Transfer means just that -- resources are
      > > manipulated by transferring representations of their current, intended,
      > > desired etc. state. Chapter 5.4:
      > >
      > > "Requests and responses have the appearance of a remote invocation
      > > style, but REST messages are targeted at a conceptual resource rather
      > > than an implementation identifier."
      > >
      > > This example is a REST anti-pattern, as I cannot deduce the current
      > > (sub)state of the order (active or canceled) by dereferencing the URL
      > > I'm given for manipulating that (sub)state. Just making a toggle POST
      > > also fails to transfer any representation of anything, let alone
      > > application state, and isn't proper HTTP (which can never be proper
      > > REST).
      > >
      > > REST isn't about optimizing upstream traffic, it's about optimizing GET.
      > > What advantage does a subresource give when it contains no content from
      > > the parent resource? My goal with subresources is to increase the cache
      > > stickiness of their parent resources. Replace the subresource content
      > > in the parent resource, with a link or a hypertext control linked to the
      > > subresource. The link or hypertext control remains static, and cached,
      > > as the content of the subresource varies.
      > >
      > > That's a RESTful pattern, as the subresource now has a representation
      > > (other than that of the 406 error) we can transfer and manipulate to
      > > effect change of the parent resource.
      > >
      > >>
      > >> > Not the semantics of a method invocation. What does /cancel
      > >> > identify? Sounds to me like a method of tunneling DELETE through
      > >> > POST which identifies nothing, iow a procedure endpoint, which is
      > >> > characteristic of various styles but not of the REST style. The
      > >> > hypertext constraint only makes sense if your resources make sense,
      > >> > in that their URLs are identifiers rather than endpoints.
      > >>
      > >> If I understand correctly, you are saying that if I need to affect a
      > >> resource then I should use the uniform interface on that resource URI,
      > >> and not another URI.
      > >>
      > >
      > > Absolutely not! My example changed /order/1 by manipulating /order/1
      > > /status. The /order/1 resource includes its status, but its 200 OK
      > > representations only include links to the /status subresource, derived
      > > from the /order/1 resource such that manipulating a representation
      > > of /order/1/status updates /order/1 (on the server, you can do anything
      > > you want; on the client, the cached link/control in /order/1 now returns
      > > a different value).
      > >
      > > There's no cost to adding a URI like this, nor does it preclude changing
      > > order status via PUT/PATCH to /order/1. The difference is that /status
      > > uses REST's uniform interface, unlike /cancel. The /order/1/status URL
      > > is only presented within a hypertext control which explains how to
      > > manipulate it -- picture a drop-down list with the current status
      > > highlighted, meeting the hypertext constraint. You can always GET the
      > > status of an order even if you don't have a copy of that order, a
      > > useful separation of concerns beyond just optimizing GET, promoting
      > > serendipitous re-use.
      > >
      > >>
      > >> > Which brings us to Chapter 5, and the short answer to your question:
      > >> > "POSTing to /cancel violates the Identification of Resources
      > >> > constraint, and is therefore unRESTful." But I've found that just
      > >> > giving that answer tends to upset folks who've only read Chapter 5,
      > >> > then they get defensive about why can't they call their API
      > >> > RESTful, accusations of pedantry follow, and threads devolve into
      > >> > general ugliness, heheh...
      > >>
      > >> After reading that chapter again I'm not sure my example violates
      > >> anything, but I'd love to get more clarification from you. Is it the
      > >> use of a verb in the URI? or not using the URI of the resource I'm
      > >> trying to modify directly?
      > >>
      > >
      > > The biggest problem, is that an RPC endpoint which has no GET function
      > > while improperly listening for a method itself as a trigger rather than
      > > taking action based on the content of the entity required by that method
      > > and in obeyance of the semantics of that method, is so far away from
      > > REST that I don't know where to start except by urging that Roy's
      > > thesis be read in its entirety. Because it's obvious to me that this
      > > violates the first three uniform interface constraints, making adherence
      > > to the fourth irrelevant:
      > >
      > > "REST is defined by four interface constraints: identification of
      > > resources; manipulation of resources through representations; self-
      > > descriptive messages; and, hypermedia as the engine of application
      > > state."
      > >
      > > Note that "noun/verb" terminology is not present in the dissertation.
      > > But, yeah, if your URI is a "verb" you're probably getting REST wrong.
      > > There is no "cancel" method in the uniform interface. There are two
      > > basic means of solving this -- one, is refactor your cancel operation
      > > to use DELETE; two, tunnel your proprietary cancel method through POST.
      > >
      > > http://tech.groups.yahoo.com/group/rest-discuss/message/19210
      > >
      > > Making this operation its own URI doesn't make it more RESTful, as the
      > > resulting URI is only a resource in the HTTP/URI sense, but not the
      > > REST sense, of the term. Utilizing subresources to break out more
      > > dynamic aspects of content and cache them separately is RESTful; if the
      > > contents of the GET are also allowable content of a PUT then we're
      > > letting hypertext drive application state instead of listening for POST
      > > events to trigger server-object methods we've failed to refactor to the
      > > uniform REST interface.
      > >
      > > -Eric
    • Eric J. Bowman
      ... I d phrase it differently: POST /cancel violates self-descriptiveness as user intent cannot be discerned by examining the request. This would hold true
      Message 2 of 28 , Dec 5, 2012
        Max Toro wrote:
        > The short answer is (if I understood correctly): POST /cancel is not
        > REST because it lacks visibility, since it's not possible to
        > understand the client's intent by examining the request.

        I'd phrase it differently: "POST /cancel violates self-descriptiveness
        as user intent cannot be discerned by examining the request." This
        would hold true even if GET /cancel had a 200 OK response. Even that's
        oversimplifying in that part of the reason POST is wrong here, is
        because DELETE is right. But, no, I do not believe setting up an URL to
        listen for POST and taking action not based on anything in the POST
        request, is ever an example of the REST style, regardless of the nature
        of the resource.

        My real short answer is, "POST /cancel doesn't use REST's uniform

        Neither is it an example of the REST style for POST to have a definition
        which varies by resource, iow sometimes it works without an entity
        (method-tunnel listener), sometimes it doesn't (create resource); even
        if hypertext-driven, such nonstandard method usage inevitably results
        in coupling between client and server, instead of relying on shared
        understanding of standardized method semantics. Fixing the POST to have
        an entity that's ignored might make it valid HTTP and solve this, but it
        still wouldn't be an example of the REST style, where the requested
        state change is transferred in the entity (with a few exceptions) -- not
        defined per resource by the method being tunneled through / triggered by

        Conceptually, it still violates the identification of resources
        constraint, which is why it isn't visible regardless of how we implement
        it in terms of hypertext or method selection. Using the uniform
        interface means refactoring internal methods like 'cancel' to fit the
        resource/representation model of REST such that the request methods and
        response codes used map generically to those of your chosen protocol.
        Assigning them URIs to toggle with semantically-void messages is simply
        some other style not encompassed by REST's uniform interface constraint,
        not something that's superficially fixable by improving visibility or
        adding a 200 OK response.

        > Never thought about HTTP/URI resource vs. REST resource, you say that
        > a URI that doesn't implement GET is probably (or always) not REST.

        More like, "REST resources are concepts, not actions." Just using URIs
        doesn't bestow compliance with the identification of resources
        constraint upon a system, it's more nuanced than that. I can't imagine
        why a bona-fide REST resource would ever deliberately fail to have a
        retrieval mapping, no.

        I think the more important takeaway is method selection. We have DELETE
        in order to avoid such convoluted POST-based cancel mechanisms. It
        isn't RESTful to use POST when another method's semantics happen to
        closely describe user intent. From Roy's blog:

        "POST only becomes an issue when it is used in a situation for which
        some other method is ideally suited: e.g., retrieval of information
        that should be a representation of some resource (GET), complete
        replacement of a representation (PUT), or any of the other standardized
        methods that tell intermediaries something more valuable than 'this may
        change something.' The other methods are more valuable to
        intermediaries because they say something about how failures can be
        automatically handled and how intermediate caches can optimize their
        behavior. POST does not have those characteristics, but that doesn't
        mean we can live without it. POST serves many useful purposes in HTTP,
        including the general purpose of 'this action isn't worth


        Since DELETE is inherently self-descriptive of user intent to cancel an
        order, and is already "listened for" rather than requiring an entity be
        transferred, and has inherent idempotency matching the inherently
        idempotent user intent captured by a cancel request, DELETE must be the
        proper method for implementation using HTTP.

        Now, maybe your media type doesn't support DELETE, in which case
        contriving a /cancel URI may very well be called for until such time
        as it does. I won't fault anyone for doing this out of pragmatism, so
        long as they understand it as a mismatch, inconsistent with the
        architectural style they're following.

        REST is a tool which allows you, over time, to make improvements to the
        architecture you have in accordance with an idealized model. To me,
        accepting mismatches for what they are, is far more useful than coming
        up with convoluted explanations for why they aren't, for the sake of
        buzzword compliance. Change is easier when it's based on having been
        right all along, rather than based on having to admit error. ;-)

      Your message has been successfully submitted and would be delivered to recipients shortly.