Experts,
We need to look at this issue again...
The EJB spec is actually very strict on accessing "other" proxy methods:
3.4.4 Session Bean’s No-Interface View
"Only public methods of the bean class and of any superclasses except
java.lang.Object may be invoked through the no-interface view. Attempted
invocations of methods with any other access modifiers via the
no-interface view reference must result in the javax.ejb.EJBException."
Obviously the container can't verify static or private methods.
With this restriction in mind, I'm changing (b) to:
"Only private methods of the bean class and any superclasses may be
declared final."
I'll also add both restrictions to the message listener methods with a
no-methods interface.
Let me know if you see any problems with this change.
thanks,
-marina
On 3/29/13 1:53 PM, Marina Vatkina wrote:
> On 3/29/13 1:39 PM, Pete Muir wrote:
>> Note this is only a problem for session beans with a no-interface view.
>
> Correct. This is what we are looking at.
>>
>> We could clean this up in the CDI spec probably, by adjusting the
>> proxyability requirements to be identical to EJB for no-interface
>> session views.
>>
>> Worth doing?
>
> Yes. Please.
>
> -marina
>
>>
>> On 29 Mar 2013, at 19:57, Marina Vatkina <marina.vatkina_at_oracle.com>
>> wrote:
>>
>>> On 3/29/13 12:45 PM, Pete Muir wrote:
>>>> On 29 Mar 2013, at 18:42, Jeremy Bauer <jrbauer_at_us.ibm.com> wrote:
>>>>
>>>>> I may have overlooked something, but does CDI allow non-private
>>>>> non-public methods to be final?
>>>> No. Not *if* a proxy is required.
>>> If the EJB container created the proxy, it won't include the
>>> non-public methods. Even if the CDI spec allows more, a compliant
>>> EJB container won't intercept on those...
>>>
>>> -marina
>>>>> With the new rules, couldn't we have this situation?
>>>>>
>>>>> @Stateless public class MyBean {
>>>>> final void m() { };
>>>>> }
>>>>>
>>>>> public class Other {
>>>>> @EJB MyBean field; // PASS
>>>>> @Inject MyBean field; // FAIL - unproxyable
>>>>> }
>>>>>
>>>>>
>>>>> -Jeremy
>>>>>
>>>>>
>>>>>
>>>>> From: Marina Vatkina <marina.vatkina_at_oracle.com>
>>>>> To: jsr345-experts_at_ejb-spec.java.net,
>>>>> Cc: Pete Muir <pmuir_at_bleepbleep.org.uk>, Jeremy
>>>>> Bauer/Rochester/IBM_at_IBMUS
>>>>> Date: 03/28/2013 02:06 PM
>>>>> Subject: [jsr345-experts] Re: Final methods in a bean class?
>>>>>
>>>>>
>>>>>
>>>>> So we are just catching up with the CDI spec ;)
>>>>>
>>>>> Thanks Pete.
>>>>>
>>>>> -marina
>>>>>
>>>>> On 3/28/13 11:21 AM, Pete Muir wrote:
>>>>>> Static methods were never considered as a block to proxability.
>>>>>>
>>>>>> On 28 Mar 2013, at 18:20, Marina Vatkina
>>>>>> <marina.vatkina_at_oracle.com> wrote:
>>>>>>
>>>>>>> How about static methods?
>>>>>>>
>>>>>>> thanks,
>>>>>>> -marina
>>>>>>>
>>>>>>> On 3/28/13 11:12 AM, Pete Muir wrote:
>>>>>>>> We updated the requirements in CDI to specifically exclude
>>>>>>>> private methods from proxability requirements. Note we moved
>>>>>>>> this to it's own section - 3.15
>>>>>>>>
>>>>>>>> Pete
>>>>>>>>
>>>>>>>> On 28 Mar 2013, at 17:16, Marina Vatkina
>>>>>>>> <marina.vatkina_at_oracle.com> wrote:
>>>>>>>>
>>>>>>>>> Well, Pete is on this list (and I'm cc-ing him here
>>>>>>>>> explicitly), but I don't think private final (or static final)
>>>>>>>>> methods should be restricted...
>>>>>>>>>
>>>>>>>>> -marina
>>>>>>>>>
>>>>>>>>> On 3/28/13 9:30 AM, Jeremy Bauer wrote:
>>>>>>>>>> Hi Marina,
>>>>>>>>>>
>>>>>>>>>> Won't allowing final methods will cause an inconsistency with
>>>>>>>>>> the CDI spec? CDI 1.0 section 5.4.1 says:
>>>>>>>>>>
>>>>>>>>>> Certain legal bean types cannot be proxied by the container:
>>>>>>>>>> [...]
>>>>>>>>>> • classes which are declared final or have
>>>>>>>>>> final methods,
>>>>>>>>>>
>>>>>>>>>> And I believe this is true in CDI 1.1, but I don't have that
>>>>>>>>>> spec handy.
>>>>>>>>>>
>>>>>>>>>> -Jeremy
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> From: Marina Vatkina <marina.vatkina_at_oracle.com>
>>>>>>>>>> To: jsr345-experts_at_ejb-spec.java.net,
>>>>>>>>>> Date: 03/27/2013 06:21 PM
>>>>>>>>>> Subject: [jsr345-experts] Re: Final methods in a bean
>>>>>>>>>> class?
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> After a longer discussion with Mark offline and not to allow
>>>>>>>>>> users to
>>>>>>>>>> shoot themselves into a foot, the new proposal is to change both
>>>>>>>>>> statements as follow:
>>>>>>>>>>
>>>>>>>>>> (a) "All non-static public methods of the bean class and of any
>>>>>>>>>> superclasses except the java.lang.Object, are exposed as
>>>>>>>>>> business methods."
>>>>>>>>>>
>>>>>>>>>> (b) "Business methods exposed through the no-interface view
>>>>>>>>>> must not be
>>>>>>>>>> declared final."
>>>>>>>>>>
>>>>>>>>>> Let me know if anybody disagrees.
>>>>>>>>>>
>>>>>>>>>> thanks,
>>>>>>>>>> -marina
>>>>>>>>>>
>>>>>>>>>> On 3/27/13 11:36 AM, Marina Vatkina wrote:
>>>>>>>>>>> Experts,
>>>>>>>>>>>
>>>>>>>>>>> The EJB spec currently requires that if a bean exposes a
>>>>>>>>>>> no-interface
>>>>>>>>>>> view (see 4.9.8 Session Bean’s No-Interface View):
>>>>>>>>>>>
>>>>>>>>>>> (a) "All public methods of the bean class and of any
>>>>>>>>>>> superclasses
>>>>>>>>>>> except the java.lang.Object, are exposed as business methods
>>>>>>>>>>> through
>>>>>>>>>>> the no-interface view."
>>>>>>>>>>> (b) "All methods of the bean class and any superclasses must
>>>>>>>>>>> not be
>>>>>>>>>>> declared final."
>>>>>>>>>>>
>>>>>>>>>>> For all other views the rules for a *business method* is
>>>>>>>>>>> "The method must not be declared as final or static."
>>>>>>>>>>>
>>>>>>>>>>> Does any of you see a problem with removing the rule (b) and
>>>>>>>>>>> changing
>>>>>>>>>>> the rule (a) to:
>>>>>>>>>>>
>>>>>>>>>>> "All non-final non-static public methods of the bean class
>>>>>>>>>>> and of any
>>>>>>>>>>> superclasses except the java.lang.Object, are exposed as
>>>>>>>>>>> business
>>>>>>>>>>> methods through the no-interface view."
>>>>>>>>>>>
>>>>>>>>>>> thanks,
>>>>>>>>>>> -marina
>>>>>>>>>>>
>>>>>>>>>>>
>