IMO, taking into account compatibility with existing applications, we should not modify the exceptions thrown by these applications. So IMO, this is what we should do:
- JAX-RS runtime MUST throw a specific WAE subtype. This must be indicated in the spec.
- JAX-RS runtime MUST NOT modify WAE instances raised by the application code (to avoid compatibility issues).
I filed a new Jira issue to track this:
http://java.net/jira/browse/JAX_RS_SPEC-306
The above will ensure that JAX-RS as well as new applications are able to leverage the new exception hierarchy. It also ensures that existing applications that throw a WAE today will not se a change in the behavior with the respect to application-thrown exceptions. Also, it avoids the need to do too much magic at runtime.
Marek
On Nov 2, 2012, at 1:35 PM, Sergey Beryozkin <sberyozkin_at_talend.com> wrote:
> Not that it should matter for the group and indeed for the spec :-), but I need to decide whether to optimize the existing (and buggy) CXF code which deals with mapping WebApplicationExceptions to subclass mappers on the server (and causes headaches for the users :-)) or simply revert to the original 1.1 code, which, for example,
>
> given a thrown WebApplicationExceptions(404) and the existence of WebApplicationException and NotFoundException mappers, will ignore NotFoundException mapper and select WebApplicationException mapper, while, a thrown NotFoundException will be mapped to NotFoundException mapper.
>
> As I said before, given that WebApplicationException(404) and NotFoundException represent the same error condition, and given the specific fact the runtime 'knows' about it by the virtue of both exception classes being part of API, it makes sense to make an exception to the algorithm specifically and only for the WebApplicationException hierarchy,
>
> but I won't be to upset with reverting to the original code either
>
> Cheers, Sergey
>
>
>
> On 30/10/12 10:41, Sergey Beryozkin wrote:
>> On 25/10/12 22:23, Sergey Beryozkin wrote:
>>> On 25/10/12 22:16, Marek Potociar wrote:
>>>> FWIW, what I recall we did agree upon, which is however not yet in the
>>>> spec is that we will make sure that all internal server-side WAEs will
>>>> be changed to proper sub-types.
>>>
>>> Can you clarify that a bit please ? What would be a difference between
>>> the internal code throwing WebApplicationException(500) and application
>>> code doing the same, as far as mapping either of these instances to a
>>> registered InternalServerErrorException mapper ?
>>
>> I'd like to get the agreement on this please. What makes me feel that
>> mapping of WebApplicationException to more specific exception mappers on
>> the server side should work is because say
>>
>> WebApplicationException(500) and InternalServerErrorException are meant
>> to represent exactly the same error condition, so, even though having
>> the current WebApplicationException handled by
>> InternalServerErrorException mapper does not work with the current
>> exception mapping algorithm, the exception has to be made specifically
>> for WebApplicationException hierarchy - it would just avoid the
>> ambiguities, example,
>>
>> "I want to have a generic exception mapping code, implemented with
>> WebApplicationException mapper, and something more specific done in case
>> of 500, implemented with InternalServerErrorException mapper".
>>
>> I know it all can be easily done at the WebApplicationException mapper
>> level itself, but the introduction of the new exception hierarchy will
>> inevitably lead to users wishing to write a cleaner base code for
>> handling all the exceptions without "ifs"
>>
>> Sergey
>>
>>>
>>>> Similarly, we agreed that on client side, proper sub-type will be
>>>> thrown based on response error code instead of a generic WAE.
>>>>
>>> That definitely makes sense
>>>
>>>> As for your question, I'm not sure it is wise to interfere with
>>>> application-thrown exceptions. (And I still cannot recall what we
>>>> decided...)
>>>>
>>> The question is whether some specific treatment is applied to WAE
>>> instances (irrespectively of where they originated from) or not, I've no
>>> strong opinion, in fact I've already implemented what I thought we
>>> might've agreed :-), but that is not important, I can revert, we just
>>> need to agree
>>>
>>> Sergey
>>>
>>>> Marek
>>>>
>>>> On Oct 25, 2012, at 11:10 PM, Marek
>>>> Potociar<marek.potociar_at_oracle.com> wrote:
>>>>
>>>>>
>>>>> On Oct 25, 2012, at 10:57 PM, Sergey Beryozkin<sberyozkin_at_talend.com>
>>>>> wrote:
>>>>>
>>>>>> Hi
>>>>>>
>>>>>> I would like to clarify the details of the way
>>>>>> WebApplicationExceptions are mapped on the server.
>>>>>>
>>>>>> Here is what I recall we talked about the other day.
>>>>>>
>>>>>> The code throws "new WebApplicationException(404)", and both
>>>>>> WebApplicationException and NotFoundException mappers are available.
>>>>>>
>>>>>> Given NotFoundException is effectively WebApplicationException(404),
>>>>>> NotFoundException mapper is chosen.
>>>>>
>>>>> Hmm... are you referring to a specific section in the spec? I cannot
>>>>> recall we agreed on that one, but I may be wrong.
>>>>>
>>>>>>
>>>>>> Similarly, if the code throws "new ServerErrorException(500)" and both
>>>>>> ServerErrorException and InternalServerErrorException mappers are
>>>>>> available, InternalServerErrorException gets chosen.
>>>>>>
>>>>>> Is it the way it should work ? This obviously is an exception to the
>>>>>> default mapping algorithm, but it appears it is logical given that
>>>>>> the runtime understands the relationship between various API
>>>>>> exception classes
>>>>>
>>>>> As I said, I don't recall such agreement, but my memory is not
>>>>> flawless... If you see it in the spec, then we did agree on that :)
>>>>> (Which still doesn't mean we didn't if you don't see it there...)
>>>>>
>>>>> Marek
>>>>>
>>>>>>
>>>>>> Cheers. Sergey
>>>>>
>>>>
>>>
>>>