[jax-rs-spec users] Re: Hypermedia API

From: Markus KARG <>
Date: Wed, 17 Dec 2014 15:28:45 +0100


thank you for your comments.

To answer your questions it is necessary to distinguish the proposal posted
here yesterday and the actual solution used in-field. The latter is written
against JAX-RS 1.0 specification hence the links are hard-coded and actually
auto-expanded. Due to the nature of that actual application this is not a
real problem. The proposal however, as I told today, is _solely_ about the
definition of the @StructuralLink annotation and the lookupLink /
lookupObject methods. As it is up to any MBW / MBR author to _use_ that
methods or not, it hence is also _his_ decision whether he implements
auto-expand or not. Or in short: This is out of scope of my proposal. But if
you ask me, how I would do it if I had to write another MBW / MBR and JAX-RS
2.1 would be finished and would include my proposal, then my answer would be
that the MBW auto-expands links, but not to the actual object, but to a
lazy-expanding wrapper. That way, the object model would be "a.b", but only
when accessing "a.getB()" for the first time, the wrapper around B would
invoke "lookupObject" _then_. As a result, the unmarshalling is fast,
consuming low memory, and auto-expands as lazy as possible. Again, this is
not part of the proposal, but it is the MBR author's decision.

In fact for writing, yes, link building happens eagerly. The reason is that
it is a rather cheap operation and the link definitively is needed rather
soon as the MBW is about to push the document to the stream anyways.

Yes, this solution can be done in an interceptor, as I already told Santiago
today. The proposal is _not_ to include any particular MBW / MBW but solely
the definition of the @StructuralLink annotation and the lookup-Methods. It
is up to any author of MBW / MBR authors, and / or up to any author of
Interceptors, to use them or not.

About the missing metadata can you please elaborate what you think is
missing hence imposes a need to "guess"?

About the necessarity of this API: Better support for hypermedia is a static
item on the JSR 370 charter. Hypermedia is about structural links and
transitional links. My proposal was announced as an example how structural
links could be supported in future, IF THIS IS WANTED. If you ask me: Build
it and they will come. My experience is that people did not use it in the
past, because it was not there. It was there in your product, but it was not
there in the standard. And many people are reluctant with non-standard

About building links by hand: This is only possible if the application
vendor and the MBW author are the same. But -at least- in my case, these are
different people. The solution we're using in-field is provided by me, the
application using it is not. See, for _some_ companies standards and
development roles do play a role, even if only for a few.

So the question is: Do we want to support structural links, yes or no? If
the answer is yes and my proposal doesn't fulfil all needs, the question is
who proposes an alternative? And if the answer is no then the question is
why asking me for a proposal in the end? ;-)


-----Original Message-----
From: Bill Burke []
Sent: Mittwoch, 17. Dezember 2014 14:53
Subject: Re: Hypermedia API

Are you automatically expanding URI links into data when reading (MBR)?
  And automatically linking when writing (MBW)? BTW, this
resolving/building of links could be done in a
Writer/ReaderInterceptor.What I don't understand is how is the container
supposed to resolve/build links. There's not enough metadata in your
example and the container would have to guess which methods to use to
expand/build the link.

I think some facility to automatically create links when marshalling
might be useful. We had a contributor build something for us, but I was
never completely satisfied with it. [1]. Honestly though I found that
I'd rather just build the links by hand. We've had this feature for
years and I've rarely gotten a question on it. Also, as I've said
before, Hypermedia/linking is pretty much ignored by app developers.
Some of the REST talks I've been to over the years suggest that it is
not a very popular "feature" of REST.

As far as automatically expanding links? I'm really not convinced it
would be useful and is a real edge case.

This does make me think it might be interesting to include the opposite
of a UriBuilder...a UriParser in the specification.

On 12/16/2014 12:41 PM, Markus KARG wrote:
> Santiago,
> I hacked a short pseudo code to illustrate what our current solution works
> like:
> ---snip---
> @Path("demo")
> public class StructuralLinksDemo {
> public class A {
> int a_id = 0;
> @StructuralLink B b_ref;
> A(int id, B ref) {
> this.a_id = id;
> this.b_ref = ref;
> }
> }
> public class B {
> int b_id;
> B(int id) {
> this.b_id = id;
> }
> }
> public class MyMBW implements MessageBodyWriter<A> {
> @Override
> public void writeTo(A a, Class<?> type, Type genericType,
> Annotation[] annotations,
> MediaType mediaType, MultivaluedMap<String,
> Object> httpHeaders,
> OutputStream entityStream) throws
> IOException, WebApplicationException {
> /*
> * When rendering instance of 'A' then reference
> 'b_ref' shall get replaced
> * by '.../demo/b/1' in this demo application!
> *
> * As MyMBW is provided by a third party vendor, it
> can neither know
> * (1) what URI pattern is used in this application
> / resource for 'GET B',
> * (2) nor whether 'b_ref' actually is intended by
> the application vendor to be a structural link or not,
> * which can be tricky in case multiple
> alternative MBW (e. g. JAXB vs. JSON-B) are to be used with
> * the same data model.
> * (3) nor what @ApplicationPath or resource's @Path
> are.
> *
> * Hence a JAX-RS API is needed which is neutral
> w.r.t to vendor of MyMBW, data format, and
> * application. This API will...
> * (1, 3) allow the MBW to 'lookupLink(Object,
> params);', which returns link of ".../demo/b/1"
> * taken from that resource method in this
> application which has '_at_GET' and returns 'B'.
> * (2) allow the MBW to identify those references in
> the model which are to be rewritten without
> * enforcing MBW-specific annotations in the
> data model.
> */
> Link b_link = JAXRS.lookupLink(a, a.b_ref.b_id); //
> link effectively is ".../demo/b/1" now!
> // 'b_link' is written to stream instead of 'b_ref',
> e. g. JAXB can be used with custom adapter!
> }
> }
> public class MyMBR implements MessageBodyReader<A> {
> @Override
> public A readFrom(Class<A> type, Type genericType,
> Annotation[] annotations,
> MediaType mediaType, MultivaluedMap<String,
> String> httpHeaders,
> InputStream entityStream) throws
> IOException, WebApplicationException {
> /*
> * When parsing document 'A' then URI '.../demo/b/1"
> has to be replaced by instance of B.
> *
> * The problems for MBR are the same as for MBW.
> *
> * Hence a JAX-RS API is needed which allows...
> * (1, 3) to forward URI to JAX-RS for resolution
> without a complete network stack loopback roundtrip
> * (2) to identify those references in the data
> model which are URIs and shall be replaced by Java object
> */
> return new A(JAXRS.lookupObject(b_link)); // returns
> instance of 'B' produced by implicit invocation of 'get_b(1);'
> }
> }
> @GET @Path("a/{a_id}")
> public A get_a(@PathParam("a_id") int a_id) {
> // MyMBW replaces instance of B by URI '.../demo/b/1'
> return new A(a_id, new B(1));
> }
> @PUT @Path("a/{a_id}")
> public void put_a(@PathParam("a_id") int a_id) {
> // MyMBR replaces URI '.../demo/b/1' by instance of B.
> ...
> }
> @GET @Path("b/{b_id}")
> public B get_b(@PathParam("b_id") int b_id) {
> // Client's MBR will invoke 'conditional GET' when replacing
> URI by instance of B.
> return new B(b_id);
> }
> @PUT @Path("b/{b_id}")
> public void put_b(@PathParam("b_id") int b_id) {
> // Client's MBW will invoke 'conditional PUT' when replacing
> instance of B by URI.
> ...
> }
> }
> ---snip---
> I think the comments expain pretty well how it works. Our particular MyMBW
> MyMBR utilizes JAXB's XmlAdapter to exchange objects with URIs and vice
> versa, and it works on both, client and server side. But as I said before:
> It is our particular data format, so it is _not_ a generic XML solution,
> we must know how the syntax for links is like. Anyways, it pretty well
> depicts the demand on behalf of a structural links API.
> What the API cannot do is -unfortunately- simplify client programming:
> Unless JAX-RS makes WADL discovery mandatory, the client cannot know the
> patterns needed. So the proposal is a first step for server side only.
> I assume that Casey's solution works similar and would benefit from the
> proposed annotation and lookupLink / lookupObject methods in the same way.
> If there are questions, just ask. :-)
> Regards
> -Markus
> -----Original Message-----
> From: Santiago Pericas-Geertsen
> Sent: Montag, 15. Dezember 2014 16:14
> To:
> Subject: Re: Hypermedia API
> Markus,
> I have a lot of questions about the proposal, but I feel most can be
> answered if you translate your ideas into a sample application. Assuming
> what you're proposing is available in JAX-RS, how would your application
> look like? Could you provide a sample app?
> -- Santiago
> On Dec 15, 2014, at 7:38 AM, Markus KARG <> wrote:
>> Sergey,
>> yes and exactly at that point we'd like to give the data model
>> designer an annotation at hand which marks that field as an injection
>> point for the actual URI at runtime. That's the sole idea of the
>> proposed API. :-)
>> Regards
>> -Markus
>> -----Original Message-----
>> From: Sergey Beryozkin []
>> Sent: Montag, 15. Dezember 2014 11:25
>> To:
>> Subject: Re: Hypermedia API
>> Hi Markus
>> What I meant is that data model designers may not necessarily need the
>> help of JAX-RS in order to design the data representations that can
>> accommodate links.
>> Example, a data designer wishing for a given piece of data have a link
>> would add a field such as href, etc... May it is oversimplifying it...
>> Cheers, Sergey
>> On 15/12/14 09:46, Markus KARG wrote:
>>> It is not as simple as you say. You're right that for XML and other
>>> generic syntax there cannot be a generic entity provider. Hence, a
>>> generic XML entity provider will not be able to fulfil the
>>> application author's wish, obviously. But in that case, the request
>>> is invalid, as it relies on the false assumption that there can be a
>>> generic solution with pure XML. XML cannot do magic tricks, hence JAX-RS
> cannot.
>>> Certainly the proposed API makes only sense for "valid wishes", i. e.
>>> the application assembler co-bundles entity providers (one or many)
>>> which are non-generic, i. e. are written according for particular XML
>>> schemas (hence not generically
>>> @Produces("application/xml") but partcularly
>>> @Produces("application/xml+foobar") for example, where "foobar" is a
>>> link-aware schema. I understand that for JAX-RS _implementation
>>> vendors_ this looks like a rather seldom case, but from the view of a
>>> JAX-RS _application vendor_ and _extension vendor_ this is my daily
>>> work,
>> actually.
>>> You're pretty right that this API is only good for the data model
>>> designers, absolutely. But I am representing these people in this
>>> expert group. JAX-RS is not only about SPI topics (which are
>>> essential for JAX-RS implementation
>>> vendors) but to a great extend is an API for application vendors. :-)
>>> -----Original Message-----
>>> From: Sergey Beryozkin []
>>> Sent: Sonntag, 14. Dezember 2014 21:30
>>> To:
>>> Subject: Re: Hypermedia API
>>> Hi Markus
>>> On 12/12/14 18:34, Markus KARG wrote:
>>>> This is correct, hence it is impossible to have a generic solution,
>>>> but need support by Entity Providers. As Entitiy Providers are aware
>>>> of the abilitiy how to merge structural links, they -and only they-
>>>> can decide about the sole correct way to add them at time of coding
>>>> and how to resolve them at time of decoding.
>>> The above is a conflicting statement. A solution involving the entity
>>> providers can not qualify as a generic solution because the providers
>>> have no idea about the schema constraints possibly applied to a
>>> representation to be produced from a given bean.
>>> This solution can work in some cases where no schemas are involved or
>>> a consumer does not care about the validation or when schemas are
>>> designed to allow the extra content. But it is not a generic solution.
>>> To be honest I do not expect such solutions to become mainstream.
>>> It is really about people designing the data model, the one meant for
>>> the external consumption, with the links in mind, auto-augmenting the
>>> existing data can be interesting but does not appear to be something
>>> 2.1 should be spending much time on
>>> Sergey
>>>> -----Original Message-----
>>>> From: Sergey Beryozkin []
>>>> Sent: Donnerstag, 11. Dezember 2014 22:24
>>>> To:
>>>> Subject: Re: Hypermedia API
>>>> The problem with auto-augmenting XML with links is that it can
>>>> produce non valid XML, schema - invalid. A corresponding schema
>>>> instance (and for most serious applications there will be a schema)
>>>> may not be open enough for a given XML instance to include extra
>>>> attributes or elements representing the links.
>>>> Sorry if I misunderstood
>>>> Sergey
>>>> On 11/12/14 21:02, Markus KARG wrote:
>>>>> I do not see how JSON-LD is any better than XML based links, as it
>>>>> doesn't solve the root problem: In the end it plays no role whether
>>>>> the document syntax is JSON or XML. The problem is the missing API.
>>>>> To make structural links work, there must be standard annotations
>>>>> in the entity POJOs, and the entity providers must process them,
>>>>> and the spec must unambiguously tell how. Yes this is tough, but I
>>>>> think that shouldn't be an excuse for not standardizing it. Rather
>>>>> we should concentrate on the question whether we _want_ standardize
>>>>> structural links or not, and in case we do, who provides the RI for
>>>>> that, as possibly Casey has something which can be built upon, and
>>>>> implementing it won't be cheap, and it has to be done by _all_
>>>>> JAX-RS
>>> vendors.
>>>>> So my question to the spec leads and vendors is: Shall we really go
>>>>> on with discussing structural links, or is it simply out of scope
>>>>> of JSR 370? Frankly, I would love to have structural links API, but
>>>>> I certainly respect it if none of the vendors wants to pay that. In
>>>>> the end, a standardization should standardize existing products,
>>>>> not enforce _all_ vendors to build something made up synthetically
>>>>> from
>>> scratch.
>>>>> *From:*Santiago Pericas-Geertsen
>>>>> []
>>>>> *Sent:* Donnerstag, 11. Dezember 2014 20:47
>>>>> *To:*
>>>>> *Subject:* Re: Hypermedia API
>>>>> Casey,
>>>>> Yes, structural links in entities is not something that JAX-RS
>>>>> provides any "special" support for (in some cases, people have
>>>>> included them in headers, but it is arguably less clean).
>>>>> The reason why JAX-RS hasn't done much (other than the JAXB Link
>>>>> serialization bit) is that JAX-RS has not been, and likely never
>>>>> will be, in the business of (structured) entity serialization; it
>>>>> delegates to specific JSON and XML libraries for that. Clearly this
>>>>> is an issue for link processing, but the architectural decision of
>>>>> not duplicating existing APIs is certainly sound.
>>>>> As you point out, JAX-RS would need some additional meta-data to
>>>>> "learn" about these links in representations. However, this needs
>>>>> to be done without introducing unnecessary coupling and in a
>>>>> standard manner --that is, not in a way that would require using a
>>>>> JAX-RS implementation for it to work. This is a difficult problem to
> solve.
>>>>> JSON-LD [1] is a step in the right direction, especially since
>>>>> becoming a W3C recommendation. Not having full control of the
>>>>> serialization is still an issue for us, but perhaps there's
>>>>> something we can do working with the new JSON-B EG.
>>>>> -- Santiago
>>>>> [1]
>>>>> On Dec 11, 2014, at 1:43 PM, Casey Lee <
>>>>> <>> wrote:
>>>>> I agree, Markus that the issue here is the technical infrastructure
>>>>> (or lack of) around HATEOAS has limited its adoption and
>>>>> understanding. Specifically, the limitation is with the fact that
>>>>> the links we currently have with JAX-RS 2.0 are only transitional
>>>>> links in the HTTP header, but no support for structural links in
>>>>> the
>>> Entity.
>>>>> I feel the issue is that the current API is all about RESOURCES,
>>>>> which causes server side developers to focus more on the URIs than
>>>>> on the REPRESENTATIONS. Additionally, this has leaked into the
>>>>> client API, causing the client side developer to also have an
>>>>> awareness of the resources, which limits the need to think about
>>>>> links or at best makes the links optional.
>>>>> Is there an opportunity to evolve the API to enable adding (server
>>>>> side) and retrieving (client side) Links from the Entity? One
>>>>> approach would be by annotating your Entity POJOs where Links would
>>>>> be added. This would allow some declaration of the structural (and
>>>>> possibly transitional) links for the representation.
>>>>> This would cause developers (client and server side) to begin to
>>>>> think about links and the structure/relationships of the
>> representations.
>>>>> At our organization, we've developed our own sets of annotations
>>>>> for declaring the structure of your representations and associating
>>>>> them to a media type. All of our documentation and the API that
>>>>> the client uses is based on the following:
>>>>> * Follow a link
>>>>> * Get back a representation
>>>>> * Find a link in the entity
>>>>> * Repeat
>>>>> This causes us to spend most of our effort describing the media
>>>>> types, and very little if any effort describing the URIs.
>>>>> -Casey
>>>>> On Wed, Dec 10, 2014 at 1:36 PM, Markus KARG
>>>>> < <>> wrote:
>>>>> Santiago,
>>>>> you asked for statements on the field of hypermedia and reactive. I
>>>>> will take the chance to start discussion hereby on the field of
>>>>> hypermedia and provide a starter for reactive in a separate thread.
>>>>> Some weeks back I gave a lecture on JAX-RS 2.0 big picture at JUG
>>>> Stuttgart,
>>>>> just as I did infrequently at other events before. I'd like to
>>>>> describe
>>>> the
>>>>> reactions as those were stereotypical, independent of audience,
>>>>> location
>>>> and
>>>>> date. People where convinced by the very clean separation of
>>>>> concerns (application made up from pure domain objects, technical
>>>>> aspects separated from domain model), and the mostly declarative
>>>>> programming style (simply adding annotations to declare needs,
>>>>> engine solves the needs "under the hood" using a sophisticated and
>>>>> extensible technical infrastructure). But when the presentation
>>>>> came to hypermedia support, they were some kind of shocked by the
>>>>> mostly algorithmic coding style needed to make it work, voiding the
>>>>> aforementioned separation of concerns and declarative code style.
>>>>> While the existing API clearly is a foundation to achieve at least
>>>>> "something", real HATEOAS becomes a hack with the existing low
>>>>> level
>>>> support
>>>>> only. Code gets cluttered with old-style techno-punk, which is hard
>>>>> to
>>>> read
>>>>> and understand. This is due to the lack of a declarative way to
>>>>> tell the infrastructure how to make up the links from application
>>>>> domain state, and how to provide the links to an entitiy provider
>>>>> so he can merge them into the wire-level representation. Certainly
>>>>> everbody would vote for a declarative kind of solution fitting into
>>>>> the existing infrastructure. On the other hand, nobody (yes, really
>>>>> zero) people wanted to agree that they have a REAL NEED for HATEOAS
>>>>> (hence, neither for an explicit HATEOAS API)
>>>> as
>>>>> 100% of all attendees admitted that their recent and current
>>>>> RESTful projects are on level 1 or 2 of the REST Maturity Model
>>>>> only, and that the largest obstacle to level 3 is not a techical
>>>>> issue (hence not a missing explicit HATEOAS API) but the fact that
>>>>> HATEOS as a paradigm simply is not well understood by most of them
>>>>> and / or they do not see the actual
>>>> benefit
>>>>> of HATEOAS in the real world: It wouldn't pay off, but it would be
>>>>> cool,
>>>> to
>>>>> sum it up.
>>>>> So the question is: Is HATEOAS commonly understood well enough that
>>>>> it
>>>> makes
>>>>> actual sense to provide an explicit API for it, or does it make
>>>>> sense to make an API even when it is not? And if we define an API,
>>>>> do we all agree that it should support the separation of concerns
>>>>> and declarative style
>>>> that
>>>>> is typical for JAX-RS?
>>>>> I think without an agreement on that general topics, it wouldn't be
>>>>> a good idea to discuss any kind of details of API proposals in the
>>>>> area of
>>>>> Bill and Sergey, what's your opinion on that?
>>>>> Regards
>>>>> -Markus
>>>>> -----Original Message-----
>>>>> From: Santiago Pericas-Geertsen
>>>> []
>>>>> Sent: Mittwoch, 10. Dezember 2014 20:03
>>>>> To:
>>>>> <>
>>>>> Cc: Marek Potociar
>>>>> Subject: Welcome to the JAX-RS 2.1 EG
>>>>> Hello Experts,
>>>>> Welcome to the JAX-RS 2.1 (JSR 370) expert group!
>>>>> This is the official mailing list for the JSR. Note that the old
>>>>> mailing list for JAX-RS 2.0 (JSR 339) is still available for 2.0
>> matters.
>>>>> Before we start any discussions, I would like everyone to take a
>>>>> couple of minutes and read the JSR description one more time to
>>>>> make sure we are all on the same page ;)
>>>>> ===
>>>>> 2.1 Please describe the proposed Specification:
>>>>> Server-Sent Events (SSE) is a new technology defined as part of the
>>>>> HTML5 set of recommendations for a client (e.g., a browser) to
>>>>> automatically get updates from a server via HTTP. It is commonly
>>>>> employed for one-way streaming data transmissions in which a server
>>>>> updates a client
>>>> periodically
>>>>> or every time an event takes place.
>>>>> JAX-RS 2.0 introduced the notion of asynchronous processing for
>>>>> both the client and the server APIs. However, asynchronous
>>>>> processing alone cannot deliver on all the promises of a modern
>>>>> architecture without the help of non-blocking I/O. If only blocking
>>>>> I/O is available, asynchronous
>>>> processing
>>>>> simply pushes the problem from one thread to the next --this is
>>>>> akin to borrowing from a person to pay another, the problem is not
>>>>> really solved, only deferred. Thus, support for non-blocking I/O is
>>>>> necessary to achieve high throughput and efficiently manage
>>>>> resources
>>> like threads.
>>>>> In summary, the following is a list of the tasks in scope for
>>>>> JAX-RS
>> 2.1:
>>>>> * Adding support for SSE.
>>>>> * Improving integration with CDI.
>>>>> * Exploring support for non-blocking I/O in providers (filters,
>>>>> interceptors, etc.).
>>>>> * Evaluating ways in which declarative security can be supported
>>>>> either directly in this JSR or by leveraging other EE-platform JSRs.
>>>>> * Making JAXB conditional on runtimes where it is available.
>>>>> * Providing integration with JSON-B.
>>>>> * Building upon the hypermedia API added in version 2.0.
>>>>> * Investigating the reactive programming paradigm as a way to
>>>>> improve the JAX-RS asynchronous client API.
>>>>> * Evaluating any requirements necessary to support the use of
>>>>> JAX-RS resource classes as controllers in the MVC 1.0 JSR.
>>>>> ===
>>>>> Some useful links:
>>>>> [JSR]
>>>>> [JAX-RS Spec] [JIRA for 2.1]
>>>>> [RI]
>>>>> [E-mail Archives]
>>>>> As before, all of our discussions will be conducted using the
>>>>> expert's alias and (automatically) CCed to the user's alias.
>>>>> Some of the 2.1 tasks above require coordination with other
>>>>> specifications (JSON-B, Security), so these tasks will tackled
>>>>> later on
>>> in the process.
>>>>> We have tentatively selected 2 topics to start our discussions,
>>>>> both of which require some investigation, these are: hypermedia
>>>>> improvements and reactive programming. If you have any
>>>>> suggestions/comments/concerns about these two topics, feel free to
>>>>> start a discussion about them. We will be sending some more info as
>>>>> well
>>> in the upcoming weeks.
>>>>> Looking forward to working with all of you!
>>>>> --
>>>>> Santiago Pericas-Geertsen
>>>>> Marek Potociar
>>>>> JSR 370 Spec Leads

Bill Burke
JBoss, a division of Red Hat