Greg Wilkins wrote on 05/31/2012 12:51 AM:
> On 30 May 2012 19:23, Bill Shannon <bill.shannon_at_oracle.com> wrote:
>> We don't intend that the behavior of the descriptor changes from release
>> to release.
>
> But we have done exactly that with 3.0
How so?
>> The only issue I'm aware of that's even close to this is what
>> should happen if an application written for an older version of the spec
>> includes annotations that the older version of the spec didn't know about,
>> and is then run on a product that supports a newer version of the spec that
>> does know about such annotations. We would lose much of the ease of use of
>> annotations if developers were required to explicitly enable understanding
>> of annotations before they could use them.
>
> I don't think we lose very much. Most new applications will be written
> with a 3.0 or 3.1 descriptor anyway, so they are already aware that
> scanning for annotations and context initialisers will take place.
> It is not much to ask of a 2.5 webapp, that if it wants to use these
> features then it should simply change the 2.5 to 3.0 or 3.1 in the
> header of the web.xml
>
>> And it would be particularly bad
>> if an application that previously didn't need a descriptor because it was
>> using annotations suddenly was required to have a descriptor to enable use
>> of new annotations.
>
> I don't think anybody is saying this. If a web app does not have a
> descriptor, then it does not declare a version and thus will obviously
> work with the semantics of the container in which it was deployed.
And adding a descriptor shouldn't change that.
>> If you want to run an application in an environment that's limited to exactly
>> and only the features available in a previous version of a spec, run it on a
>> product that supports only that version of the spec. Products may also offer
>> modes that provide different environments limited to different versions of the
>> spec if they believe that this level of compatibility is important.
>
> I think you will find this is unacceptable in many/most production
> environments. You will find that containers are updated because of
> security fixes or corporate infrastructure beyond the control of the
> web application creators. I really don't think it unreasonable to say
> that a 2.5 webapp with meta-data complete set should be able to be
> deployed without new annotations and servlet context initialisers
> being fired (which may be brought in from updated dependencies).
metadata-complete will cause annotations to be ignored.
Servlet context initializers are part of the extensibility mechanism of
the container and shouldn't be disabled.
> I agree that we don't want to provide fine grained 2.5 compatibility,
> but turning off some auto discovery mechanisms for 2.5 webapps is not
> that difficult to do. If you want autodiscovery, it is trivial to
> update a 2.5 descriptor to 3.x
We're not going to change this, sorry.