RE: [ws-caf] Fault message formats

["Murray, Bryan P." <bryan.murray@hp.com>]

Since the TC members are clearly opposed to using the SOAP Fault element
to convey fault information related to WS-Context messages, I will
refrain from further comments on the subject. However, I thought the
group might be interested in the results of a very unscientific and
incomplete research to see how some other specifications handle faults.

I divided specs into 3 categories. First there were specs that either
did not directly define messages or did not specify a WSDL document.
BPEL and WSFL fall into this category, although the text and examples in
WSFL imply that SOAP Faults are intended to be used with the lifetime
operations that are defined.

The next category is for specs that explicitly do not use the SOAP Fault
element. In my quick review I only ran across WS-Reliability for this
category. This spec does require the use of the SOAP Fault element in
request-response exchanges, but does not put any information in the
Fault element, using a SOAP Header element instead.

The last category is for specs which use the wsdl:fault element and have
text implying the use of the SOAP Fault element. In this category are
WS-MetaDataExchange, WS-ReliableMessaging, and WS-Coordination.
WS-Coordination is underspecified in the area of faults, but the
following statement from the introduction to the section titled
"Coordination Faults" is a pretty strong indication of the intent: "For
example, when used with SOAP 1.2 the identifier code is the fault
sub-code and any additional information is passed in the detail." I
think an earlier version of WS-Coordination did define their own fault
messages.

I also looked at ebXML Messaging. Although that spec talks about using
SOAP Faults when the severity is error but not when it is warning, I
think it is more like WS-Reliability uses them. It is not very clear in
the spec.

I did not include specs which build on those listed above such as
WS-AtomicTransaction, or other ebXML specs. I also did not include the
specs this TC is writing or some more contentious specs such as WSRF.

This review did not produce any clear direction although it seems that
more recently written specs have a stronger tendency to use the SOAP
Fault element to convey error information then older specs did.

Bryan