Re: Proposal for Async Extensions

Marc Hadley wrote:
> On Jun 15, 2005, at 3:01 PM, Anish Karmarkar wrote:
> 
>>
>> In the case where wsdl:required="true" on a wsaw:ResponseBinding,  it 
>> would not make sense to have more than one wsaw:ResponseBinding.  In 
>> fact it would be an error to have two wsaw:ResponseBinding with  
>> wsdl:require='true' on both.
>>
>>
>>> Does that make sense?
>>> Or (re-reading) did you mean that wsdl:required would mean that
>>> anonymous responses aren't allowed?
>>>
>>
>> Yes, that's what I meant.
>>
> That sounds a little odd since anon replies can still come back  before 
> [reply endpoint] takes effect (mU faults e.g.) ...
> 

The mU fault issue is a general issue when using async over HTTP (or any 
protocol with a back-channel).
As Hugo has mentioned on the ws-addr list, this is a problem whenever a 
[fault endpoint] is specified.

But the usecase I want to support is not about faults.
I want the ability to static specify in WSDL that the service will send 
the response (not faults) message asynchronously and therefore the 
invoker of the service must specify the [reply endpoint].

Make sense?

> Even if we get round that problem it seems odd to only be able to  
> demand use of one particular transport rather than just being able to  
> say "must be async".
> 

Why?
The service may only support one transport (HTTP) and wants to indicate 
that the [reply to] endpoint (sent in the request message) must be a 
HTTP url/binding.

But it is certainly a valid usecase to say "must be async" without 
demanding the use of one particular transport.

-Anish

...
<snip/>

Received on Wednesday, 15 June 2005 23:00:52 UTC