On 3/21/12 6:18 PM, Sergey Beryozkin wrote:
> Hi Bill,
>
> On 21/03/12 19:08, Bill Burke wrote:
>>
>>
>> On 3/21/12 3:01 PM, Marek Potociar wrote:
>>>
>>> On Mar 19, 2012, at 3:01 PM, Bill Burke wrote:
>>>
>>>>
>>>>
>>>> On 3/15/12 4:15 PM, Santiago Pericas-Geertsen wrote:
>>>>>
>>>>> On Mar 15, 2012, at 4:00 PM, Bill Burke wrote:
>>>>>
>>>>>>
>>>>>>
>>>>>> On 3/15/12 3:54 PM, Santiago Pericas-Geertsen wrote:
>>>>>>>
>>>>>>> On Mar 15, 2012, at 3:31 PM, Bill Burke wrote:
>>>>>>>
>>>>>>>> I would say do nothing because it would break existing 1.1
>>>>>>>> applications when deployed.
>>>>>>>
>>>>>>> How so? Before you had to always specify the mapping, and that
>>>>>>> will continue to take precedence. Can you elaborate?
>>>>>>>
>>>>>> We assumed everything could be scanned, you could have an empty
>>>>>> web.xml file, and no Application class. This was an incorrect
>>>>>> assumption? I know a lot of people use us in this manner. Maybe I
>>>>>> just interpreted the 1.1 spec wrong?
>>>>>
>>>>> Class scanning and servlet mapping are orthogonal. The only way you
>>>>> can get the servlet mapping in 1.1 is from (i) a web.xml or (ii) the
>>>>> @ApplicationPath annotation on an Application subclass AFAICT, with
>>>>> (i) overriding (ii) if both are present. For the other cases, 1.1
>>>>> states that "the application MUST be packaged with a web.xml that
>>>>> specifies a servlet mapping for the added servlet".
>>>>>
>>>>> Perhaps you're already using a default in Resteasy?
>>>>>
>>>>
>>>> Well, then the spec should allow the case for *no* Application class
>>>> and no web.xml listing. The default mapping should be "/*" and should
>>>> work with static content too. This way its very simple for users and
>>>> no thought has to be put into anything.
>>>>
>>>
>>> I think so far we can all agree, that a default mapping is good. I
>>> want to however point out that suggested "/*" as a default mapping is
>>> too aggressive in connection with any other technology that uses some
>>> default mapping as well (e.g. JSF). For that reason, I suggest to
>>> choose a named default mapping directly under the root path e.g. the
>>> "/webapi/*" as proposed earlier.
>>>
>>
>> Please tell me why "/*" is too aggressive? its easily handled if your
>> implementation is Filter based.
>>
>> But, this is orthogonal to the issue of requiring a web.xml or
>> Application class. I do not think either of these artifacts should be
>> required to deploy a JAX-RS service.
>
> Without Application, what would be the most portable way to deploy a
> simple JAX-RS service across multiple JAX-RS stacks ? Sorry may be this
> question is also orthogonal :-),
>
Not sure I understand you. just put a resource class in /WEB-INF/classes
and have the scanner discover it. Seems pretty straightforward to me.
YOu want a full Java EE container to have value-add. One of the value
adds is scanning.
--
Bill Burke
JBoss, a division of Red Hat
http://bill.burkecentral.com