On Fri, Dec 9, 2011 at 2:24 PM, Simon Laws <simonsl...@googlemail.com> wrote:
> On Fri, Dec 9, 2011 at 2:16 PM, ant elder <antel...@apache.org> wrote:
>> On Fri, Dec 9, 2011 at 2:04 PM, Simon Laws <simonsl...@googlemail.com> wrote:
>>> On Fri, Dec 9, 2011 at 1:46 PM, ant elder <antel...@apache.org> wrote:
>>>> On Thu, Dec 8, 2011 at 5:39 PM, Simon Laws <simonsl...@googlemail.com>
>>>> wrote:
>>>>> On Thu, Dec 8, 2011 at 4:22 PM, ant elder <ant.el...@gmail.com> wrote:
>>>>>> On Thu, Dec 8, 2011 at 10:57 AM, Simon Nash <n...@apache.org> wrote:
>>>>>>> ant elder wrote:
>>>>>>>>
>>>>>>>> This is good and i will have a go in 2.x to see if a similar approach
>>>>>>>> works there. But i can't help thinking its a slightly convoluted
>>>>>>>> approach and as we have more flexibility in 2.x to add more APIs I
>>>>>>>> wonder if we should just add a more direct API method to set the URIs
>>>>>>>> with extra methods on Node, or also perhaps something like being able
>>>>>>>> to do: ((TuscanyServiceReference)serviceReference).setURI(uri)?
>>>>>>>>
>>>>>>> +1 for providing a more convenient solution in 2.x.
>>>>>>>
>>>>>>> A setURI() method would only work for certain binding types. Perhaps
>>>>>>> this could be generalized to use some kind of notion of endpoint.
>>>>>>>
>>>>>>> There's also the mysterious wiredByImpl notion from the SCA spec.
>>>>>>> It might be good to use that as the base concept for this capability.
>>>>>>>
>>>>>>
>>>>>> Ok i had a quick stab at this in r1211944 and 1211945 which enables
>>>>>> setting the binding URI of a ServiceReference by casting it to
>>>>>> org.apache.tuscany.sca.core.context.impl.ServiceReferenceImpl and
>>>>>> calling setBindingURI. Eg
>>>>>>
>>>>>> ServiceReference<Helloworld> sr =
>>>>>> componentContext.getServiceReference(Helloworld.class,
>>>>>> "helloworldService");
>>>>>>
>>>>>> ((ServiceReferenceImpl)sr).setBindingURI("http://localhost:8080/HelloworldService/Helloworld";);
>>>>>> return "client: " + sr.getService().sayHello(name);
>>>>>>
>>>>>> Obviously its not ideal to be casting to a that impl class and it
>>>>>> should rather have an API/SPI interface to cast to. I haven't done a
>>>>>> whole lot of testing but that that seems to work ok, can anyone see
>>>>>> any obvious issues?
>>>>>>
>>>>>> ...ant
>>>>>
>>>>> Ant
>>>>>
>>>>> I like the idea of being able to prod things into the service
>>>>> reference. I have two immediate thoughts that are not well formed....
>>>>>
>>>>> - Could we remove the cast by providing a Tuscany specific
>>>>> ComponentContext for wireByImp impls. We already have the Tuscany
>>>>> RuntimeComponentContext.
>>>>> - Are there binding specific things we will want to do. Nothing comes
>>>>> to mind just yet although policy configuration might be a possibility.
>>>>>
>>>>
>>>> The main issue with the current cast is that its casting to an impl
>>>> class so just using some existing or new interface to have the
>>>> setBindingURI method and casting to that would be better, and that
>>>> allows adding whatever else we think of that might be useful to expose
>>>> to users. I guess an alternative or addition would be to define a
>>>> Tuscany specific annotation to get hold of the Tuscany specific
>>>> context instead of using the OASIS annotation.
>>>>
>>>> ...ant
>>>
>>> IIRC the OASIS annotation code is smart enough (or can be made that
>>> way) to inject the right thing based on the Type of the field to be
>>> injected. We could gate this algorithm based on whether the reference
>>> is marked as wire by impl. I.e. are you allowed to affect the wiring
>>> from within the implementation. The assembly spec says this about
>>> wiredByImpl
>>>
>>> If set to "true" it indicates that the target of the
>>> 353 reference is set at runtime by the implementation code (e.g. by
>>> the code obtaining an endpoint
>>> 354 reference by some means and setting this as the target of the
>>> reference through the use of
>>> 355 programming interfaces defined by the relevant Client and
>>> Implementation specification).If
>>> 356 @wiredByImpl is set to "true", then any reference targets
>>> configured for this reference MUST be
>>> 357 ignored by the runtime. [ASM40006]
>>>
>>> I've never been sure whether "implementation" here refers to the
>>> component implementation or the component implementation type (the
>>> infrastructure code) or either.
>>>
>>> I supposed there's nothing stopping us from making the feature
>>> available regardless of whether wiredByImpl is set but. if nothing
>>> else, we could use it to explain what wiredByImpl is for.
>>>
>>
>> The issues I have with having wiredByImpl control whether or not this
>> is possible are that it defaults to false and if its true then you
>> can't have a default for the reference defined in the SCDL (asm40006).
>> Perhaps I don't understand the reasoning behind those but it seems
>> more useable the other way around to me.
>>
>> ...ant
>
> That's what I was saying at the end (I think) that we make the
> facility available regardless but turning wiredByImpl on means that
> you have to use it. It you don't turn wiredByImpl on you can either
> set the reference through the new API or rely on the target in the
> SCDL.
>
I've committed some changes so that you can use some Tuscany versions
of the OASIS ComponentContext and ServiceReference interfaces. There's
an example of them being used to dynamically alter a Web service
reference URL at:
https://svn.apache.org/repos/asf/tuscany/sca-java-2.x/trunk/testing/itest/dynamicRefURI/src/main/java/org/apache/tuscany/sca/itest/HelloworldClient.java
I've not yet added this to all the OASIS ComponentContext methods (eg
getServiceReferences) and there is now some rationalization of the
Tuscany interfaces and classes I think we should do (eg
ServiceReferenceExt among others), but this gives an idea of what we
could do. I've not done anything with wiredByImpl either as i'm still
not sure its terribly useful.
...ant
