On 13/12/11 12:15, Marek Potociar wrote:
>
>
> On Tue 13 Dec 2011 12:11:49 PM CET, Sergey Beryozkin wrote:
>> Hi Marek
>> On 13/12/11 10:59, Marek Potociar wrote:
>>> Just wanted to add that this functionality is not only focused on some low-percentage use case client extensions. I
>>> believe it is possible to expect that many users will want to access specifics of Jersey, RestEasy or CXF etc. client
>>> implementations and configurations in a type-safe way.
>>>
>>
>> I think many users who will migrate to 2.0 client API will be very sensitive about writing unportable code; the
>> example you shown is technically interesting but I wonder if people will ever write such code,
>
> If you properly encapsulate the code using proprietary features/config,
> there is no reason why to shy away of extensions completely.
>
>>
>> I may be wrong, but a code like this:
>>
>> Client client = ...
>> MyJaxrsImplStack.cast(client).setRequestQueueCapacity(10) is equally 'portable'
>
> I believe you miss the fact that your cast may fail if the client
> instance is not of the expected type. But this is just my rant to this
> particular solution. As I said previously, there are ways how to use
> extension even without the API.
That was not a rant at all, just mild objection :-)
By the way, I'll withdraw my argument about the portability which was
wrong,
I've no strong objections to having these interfaces retained; but seems
like you and Bill can agree on dropping them which is good for me,
As I said that example looks OK to me; but if the goal of these
interfaces is to make it possible only to set the extensions properties
in a typed safe way then may be we should let them go if doing so will
simplify the client API
Sergey
>
> Marek
>
>>
>> Cheers, Sergey
>>
>>> Similarly, clients supporting HTTP protocol extensions (e.g. WebDAV) are another large use case group.
>>>
>> That is more interesting, would be useful to see an example from Markus or yourself; just would like to understand if
>> we should get some setProperties(Map) method instead for various extensions to get configured
>>
>> Cheers, Sergey
>>
>>> Marek
>>>
>>> On 12/13/2011 11:53 AM, Marek Potociar wrote:
>>>> The idea is that with these interfaces you will be able to provide a fluent common API for a type-safe access to any
>>>> client extension. The two-phase process (Factory->Builder) is there to enable the ability to type-safely configure the
>>>> produced custom client before it is built. I took the inspiration in Bean Validation spec. Additionally, the API was
>>>> supposed to provide injection support for the custom client builder artifacts.
>>>>
>>>> In an example, I wanted to make sure users are able to write this (see also in JAX-RS API client examples):
>>>>
>>>> ThrottledClient Client = ClientFactory
>>>> .newClientBy(ThrottledClient.Builder.Factory.class)
>>>> .requestQueueCapacity(10)
>>>> .build();
>>>>
>>>> The above produces a ThrottledClient instance that uses a blocking queue to throttle the amount of client requests
>>>> flowing through the client instance (see Staged Event Driven Architecture (SEDA) for more on this concept).
>>>>
>>>> Let's analyze the details of the example:
>>>>
>>>> ClientFactory.newClientBy(ThrottledClient.Builder.Factory.class)
>>>>
>>>> instantiates and possibly injects the factory that returns a typed ThrottledClient.Builder instance. The returned
>>>> builder exposes a type-safe configuration method...
>>>>
>>>> .requestQueueCapacity(10)
>>>>
>>>> ... that is used to configure the client instance request queue capacity (note that this capacity is not modifiable
>>>> once
>>>> the client instance is created). At last a typed ThrottledClient instance is built and returned
>>>>
>>>> .build()
>>>>
>>>> so that the instance information can further be accessed in a type safe way (e.g. client.getRequestQueueCapacity()).
>>>>
>>>> That's for the explanation of why it is designed the way it is designed.
>>>>
>>>> I do not feel too strongly about this part of the API though. There are obviously other ways of achieving at least part
>>>> of the same functionality even though I have yet to see a solution that would look more elegant in the code (from the
>>>> user's perspective).
>>>>
>>>> To sum up, I like the current API but if you can find more support in EG for removing the API or propose a solid
>>>> replacement for the functionality covered by the API, I would not stand in your way...
>>>>
>>>> Marek
>>>>
>>>> On 12/12/2011 02:47 PM, Bill Burke wrote:
>>>>> Maybe I should give more detail (again):
>>>>>
>>>>> * We don't need a set of interfaces the application developer will never see
>>>>> * The interfaces are redudant. All you really need is to specify a javax.ws.rs.ext.ClientFactory with the class
>>>>> name of
>>>>> your impl's factory within it.
>>>>> * If you want to be able to support extensions (i.e. an extended interface to ClientFactory or Client), then it should
>>>>> be left up to the provider how they want to support it. For example, I don't want Resteasy to be forced to use your
>>>>> interfaces, especially when a typecast or even a constructor invocation would be good enough for this scenario.
>>>>>
>>>>> On 12/12/11 8:35 AM, Bill Burke wrote:
>>>>>> I've already complained about this before and gave in-depth reasons why
>>>>>> we don't need these interfaces, but can we please remove Client.Builder
>>>>>> and ClientFactoryBuilder? They are not needed by either the jaxrs
>>>>>> implementation or the application developer. All logic can be contained
>>>>>> within ClientFactory or Java EE app server.
>>>>>>
>>>>>> I have yet to hear a good reason from you guys why we need this extra
>>>>>> bloat.
>>>>>>
>>>>>> Bill
>>>>>>
>>>>>
>>
>>