about latest contribution + 1 observation

[Jacques Durand <JDurand@us.fujitsu.com>]

Title: about latest contribution + 1 observation

I did not send more feedback on the list about the thread on latest contribution from Doug, but discussed them by phone with him yesterday, and Doug's latest upload reflects agreements we have reached - except for the point below:

Doug made the case for his edit of 2.5 where only SOAP One-way (not SOAP request-response) is now used for defining the Callback reply pattern (and Poll ) in Section 2.5 ( I initially did not write this restriction), by pointing that he was only aligning the definitions so that they are compatible with the HTTP binding requirements (Section 6), that require these messages to adjust to SOAP One-way, and One-way only. So indeed, section 2.5 and section 6 are now more consistent with each other.

I must say I had not focused much on this section (6) before, but I have to point out something important about its current wording:

After a careful reading of Section 6.2 and 6.3,  my conclusion is the following:
For RMP implementations that are BP 1.0 compliant, the current wording of Section 6.2/3 makes it impossible to use Callback and Poll reply patterns for messages that relate to WSDL request-response operation types (see explanation below).

 I know some of us had proposed this (Tom) before, but I am not sure this is what the TC really intended, as we did not pursue this idea.

Certainly, the now defunct Table 5.2 contradicted this restriction.
 I personally have no strong opinion on this yet would rather not have this restriction.

Explanation:
 Section 6.2 describes precise HTTP requirements related to BP 1.0  for a Callback response in case of Faults ("empty HTTP entity-body", etc) .Two statements here are restricting the use of Callback as applicable only to WSDL One-way operation types.

- L1487 (item 2) clearly states that no SOAP envelope should go in the HTTP response, at least in normal (non fault) cases. This combined with BP 1.0 rules out the use of WSDL request-response operations, which bind to HTTP request-response (via the SOAP binding mandated in BP1.0. Although this can be seen as quite restricting, that also can be seen as a major helper  in SOAP interoperability.).

- L1479-1482 requires NO SOAP fault to be returned in case of error. But  R1029 in BP 1.0 : ("Where the normal outcome of processing a SOAP message would have resulted in the transmission of a SOAP response, but rather a SOAP Fault is generated instead, a RECEIVER MUST transmit a SOAP Fault message in place of the response.") requires a SOAP fault to be returned when a SOAP response was in order. And BP 1.0 always requires a SOAP response over HTTP response,  for output messages that are defined for WSDL requrest-responses operations. So these operations must always generate a SOAP Fault on the HTTP response, which conflicts with our requirement.

We havea  similar situation for Poll pattern (6.3)

Proposal:
If the TC intends to accept this restriction, that should be made more explicit and at higher level (our spec is also for WS users who think in terms of what kind of reliability features they can use with what Web service operations, not just for developers who will look at what HTTP binding they should use.)

If not, I would suggest we reword 6.2 / 6.3 (to accommodate both WSDL one-way and request-response), and relax the One-way restriction for Callback reliable message.

Jacques