Providers ordering
Pavel Bucek
Dear experts, there is couple of filed issues related to provider ordering, namely: https://github.com/jax-rs/api/issues/538 We already do have them implemented in the RI and we discovered that we can do similar things for almost all providers (including readers/writers, ..), which don't have already defined @Priority handling (filters and interceptors do have that already). For these providers, lover priority value means higher priority. For example, if you have following ExceptionMappers:
then "MySecondExceptionMapper" will be invoked when "MyException"
is thrown.
@Provider @Priority(300) public static class MyFirstExceptionMapper implements ExceptionMapper<MyException> { // ...} @Provider @Priority(100) public static class MySecondExceptionMapper implements ExceptionMapper<MyException> { // ... } @Provider @Priority(200) public static class MyThirdExceptionMapper implements ExceptionMapper<MyException> { // ... } Do you have any comments or suggestions before we incorporate that change into the spec document? Thanks and regards, Pavel
|
|
Guillermo González de Agüero
Hi,
toggle quoted messageShow quoted text
Have you considered how this will behave when the provider is a CDI @Alternative or @Stereotype? I'm unsure how much those could interfere but I think it's something to take into account nonetheless, specially since improved CDI integration is already planned [1]. Regards, Guillermo González de Agüero
El vie., 19 de mayo de 2017 14:04, Pavel Bucek <pavel.bucek@...> escribió:
|
|
Hi everyone, I'm very happy to see that the issue of provider ordering is addressed in the upcoming JAX-RS release. Thanks a lot for your work on that. I would like to point the EG to a discussion from one of the issues which was about whether the priority ordering should be used only for "custom providers" or for all providers (custom and built-in providers). See [1] for details. During the discussion Pavel mentioned that JAX-RS already mandates that custom providers are preferred over built-in providers. I didn't find this anywhere in the spec document, but maybe I just missed it. The one thing that still makes me think that the priority ordering should be used for all providers is that the definition of "built-in provider" is a bit blurry. What is a built-in provider? What about providers which are deployed as app server modules and therefore are provided out of the box by the container? Are they built-in or custom providers? Perhaps this should simply be clarified in the spec? I would be happy to hear about other opinions. Christian 2017-05-19 14:08 GMT+02:00 Pavel Bucek <pavel.bucek@...>:
--
|
|
Pavel Bucek
Hi Christian, see chapter 4.2.4:
Which i s the base of what I mentioned - there is one issue though - this is only about entity providers, not about ExceptionMappers, which are covered by chapter 4.3. I believe we could include more general statement in chapter 4,
which would say that application provided Providers have priority
over build-ins (and maybe define what build-in is) OR we can say
that implementation providers should have some concrete priority
set - which could be simpler than formally defining built-in
provider. Santiago, others, what do you think? Thanks, On 21/05/2017 10:48, Christian
Kaltepoth wrote:
|
|
Guillermo González de Agüero
Hi, On Sun, May 21, 2017 at 11:03 AM, Pavel Bucek <pavel.bucek@...> wrote:
I was writing to propose exactly that. That's be the more flexible approach, and would resolve my open question about what exactly is a built-in provider (https://github.com/jax-rs/api/issues/537#issuecomment-302853229).
Regards, Guillermo González de Agüero.
|
|
Hey Pavel, thanks a lot for your quick reply.
Thanks a lot for pointing me to section 4.2.4. But as you already wrote the current version of this statement just covers entity providers.
I would be fine with both options as both would fix the problem I mentioned. Although I think the latter one would should be preferred because it is much simpler. Best regards Christian
|
|
Sergey Beryozkin
Hi Pavel Not sure it is relevant but it is not exactly the case with MessageBodyReader/Writers where the custom providers can still lose to built-in providers. By the way, what is the actual point of ordering say 2 ExceptionMapper<MyObject> - would it go a bit too far without bringing any real benefit ? Thanks Sergey
On 21/05/17 10:03, Pavel Bucek wrote:
|
|
Hi Sergey,
Of course nobody will want to create two ExceptionMapper<MyObject> implementations in his application. It is more about the case that some framework provides a "default" exception mapper for a specific exception type and the developer should still be able to provide a custom one which is preferred by the JAX-RS implementation. We had this case in JSR-371 (MVC 1.0): If the validation of a CSRF token fails, the MVC implementation throws a CsrfValidationException. The MVC implementation provides a default mapper ExceptionMapper<CsrfValidationException> which just sends a 403 status code. But the developer should be able to provide a custom exception mapper which does something else, like rendering some HTML page. However, this is currently not possible, because if there are two ExceptionMapper<CsrfValidationException> implementations discovered, there is no way to tell the JAX-RS implementation which one to choose. Christian
|
|
Sergey Beryozkin
Hi Christian Right, do you mean that both of these providers are custom as far as the JAX-RS implementation is concerned ? Sure, I see what you mean now. I guess my only doubt at this stage is, should the providers offered by MVC implementation (or some other *spec-implementaion* API) have a special treatment status, as far as JAX-RS is concerned ? Example, be annotated with some @DefaultProvider annotation (to be introduced in JAX-RS) ? In this case JAX-RS, seeing two MyObject exception mappers, will select the one which has no @DefaultProvider. May be using @Priority is simpler, but it would require application mappers to know how to set their own @Priority to ensure it is high or low enough. The problem with @DefaultProvider is that it is probably too late as ex MVC API would need updated after 2.1 is out. Cheers. Sergey
On 21/05/17 15:58, Christian Kaltepoth wrote:
|
|
Hey Sergey, thanks for your quick reply.
Yes, exactly. In this case JAX-RS currently doesn't specify which one should be used.
Not sure if something like @DefaultProvider is flexible enough. Maybe this "default provider" vs "custom provider" use case is just one special case? I mean, it is always possible to have multiple exception mappers on the classpath if they are provided by different libraries. And in this case it would be nice to have some clear rules on how to handle this situation.
For filters and interceptors there is already javax.ws.rs.Priorities which contains constants which can be used for ordering. So this could also be used for exception mappers and param providers.
That wouldn't be a problem, because MVC will go final after JAX-RS and therefore could depend on JAX-RS 2.1. Best regards Christian
|
|
Sergey Beryozkin
Continuing with the idea of the special
treatment of the providers shipped with the spec implementations. Rather than pursue a @DefaultProvider idea, how about treating the providers in the javax.* namespace as 'default' ones ? Thus if say an MVC API ships its own CSRF exception mapper then it will be selected, unless the application has registered its own. Moreover, perhaps this idea and the idea of using @Priority are not mutually exclusive. IMHO the former puts less 'pressure' on the application providers to make it right... Sergey
On 21/05/17 16:19, Sergey Beryozkin wrote:
|
|
Sergey Beryozkin
Hi Christian
On 21/05/17 18:22, Christian Kaltepoth wrote: Thanks too, I agree using @Priority is a viable option. But as I implied earlier, this approach (apart from solving the issue shown with MVC vs custom CSRF mapper) raises the question - why to support two ExceptionMapper<MyObject> or ParamConverters in general. Perhaps it does not really matter, but I guess we'd like to avoid supporting the solutions where the developers will start thinking about dropping 3 ExceptionMapper<MyObject>, and ordering them 1, 2, 3 and making each of them, say, 1 and 2, not to map, for 3 to do the final mapping :-). That said, may be this example is hypothetical, and using @Priority just works, though I thought I'd suggest a 'default mapper' around javax.* idea too... Thanks, Sergey
|
|
Santiago Pericas-Geertsen
I agree. Naturally, user-defined providers should take precedence over built-in ones. Clearly, this statement is implicit rather than explicit and should be clarified —at least for some providers. That was always the intent. As for what is user-defined vs. built-in, I don’t thing that is hard to define: if it is registered using the JAX-RS API or discovered via class scanning (i.e. annotated with @Provider), then it is user-defined regardless of their physical location (library, container, etc.). Built-in ones should not be made available using those mechanisms. — Santiago
|
|
Santiago Pericas-Geertsen
— Santiago
|
|
We should not mix up concerns here. @Priority should only beak tie. Whether or not to choose built-in vs. application-provided should be separately declarable. For example, in case A it might make sense that an appliation provides a simplistic fallback provider just in case the JAX-RS implementation does not provide support for a particular feature (like PDF support for example). On the other hand, the application might want to provide a superior, really-expensive, full-monty provider for PDF, which shall always override a possible existing built-in one. This example showcases that it MUST be up to the application programmer to declare precedence. But I won't use @Priority for that. It should be a separate annotation.
|
|
Santiago Pericas-Geertsen
Hi Markus,
toggle quoted messageShow quoted text
The precedence of application-defined vs. built-in providers has already been established and cannot be changed at this point (but should be clarified better as mentioned by others). Conceptually, we have two ordered sets and we only consider the built-in set iff the application-defined set is empty after filtering. In my view, the introduction of @Priority should only apply to application-defined providers —which are in fact the ones that developers can add annotations to. — Santiago
|
|
I do not see that we cannot change this. If we decide that a new, optional annotation will modify the precedence, this would be perfectly backwards compatible, as all non-annotated code will still work correctly.
|
|
Sergey Beryozkin
I guess that what I also suggested when we talked with Christian about @DefaultProvider - the only problem from what I understood MVC will've been already released by the time JAX-RS 2.1 gets out.
To be honest, I'm still thinking, checking the origin (package) of the mapper can work. If for ex MVC ships its ExceptionMapper it is still a built-in mapper in context of MVC, and it can be identified to be a built-in one because it will be in a javax. namespace. Such providers should get a lower priority OOB (compared to the otherwise equal competing mappers), without having to depend on @Priority. @Priority is there to sort the providers, making sure they execute in the right order. Using it as a mechanism to exclude does not seem quite right...Though it can work... Cheers, Sergey
|
|
Santiago Pericas-Geertsen
— Santiago
|
|
Sergey Beryozkin
Hi Santiago In what other cases, apart from ensuring a mapper shipped with MVC (or other spec APIs depending on JAX-RS), we would want to support sorting two ExceptionMapper<SomeClass> ? Cheers, Sergey
On 25/05/17 15:01, Santiago Pericas-Geertsen wrote:
|
|