A few thoughts on what has changed since .992 (our last public reviewspecification)

[Doug Bunting <Doug.Bunting@Sun.COM>]

For grins, I compared the two specification documents.  The version of 
1.086 including all of the identified changes since .992 contains many more 
insertions and deletions than consistent text.  Partially, this results 
from a StarOffice tendency to group small changes into large insertions and 
deletions.  Mostly, it shows how much of the document changed from an 
editorial perspective; 1.086 is a substantial rewrite.

I also compared the protocol schema files.  To start with, two schema files 
(one each for SOAP 1.1 and 1.2) have been merged.  More importantly, 
separate schema files for features and properties, WS-Reliability features 
and properties, and the generic reference container were created.

The specific changes listed since the .992 "core" schema include:
"
0.65: 04/20/04: Replaced 'status' attribute with 'last'
0.66: 04/30/04: Added the GroupAborted Fault and fixed a typo
0.70: 05/03/04: Changed the namespace and made it SOAP version agnostic.
                     Allso, removed the SOAP:mustUnderstand attribute.
0.90: 05/19/04: Use the ServiceRefType
1.1: 06-07-2004: Renamed the file and target namespace
1.1a:07-20-2004: Added BareURI type
1.1b:07-27-2004: Change SequenceNumberRange to SequenceNumRange
1.1c:07-29-2004: Update all mandatory attributes with use=required
1.1d:08-3-2004: Some more updates based on new decisions
1.1c:08-04-2004: Removed the global ReplyTo and localized the references
"

Alone, the 0.65, 0.66, 0.70, 1.1a and 1.1.b changes seem substantive to me.

But wait, there's more!  Just in the 1.1d revision (if I remember 
correctly), a Response@replyPattern attribute was deleted, the Response 
content model was made general enough to validate some of our examples, and 
a number of Fault codes were renamed or deleted.  The 0.90 revision made an 
attribute into an element with an extensible content type.

Back to the specification itself: The .992 document was open to a number of 
interpretations.  I favoured an interpretation in which WS-Reliability was 
viewed as a SOAP Binding -- meaning WSDL describing the application level 
could be very different from WSDL describing the initial Reliable Message 
exchange.  Others viewed the protocol as a set of SOAP Headers and related 
semantics added to an existing application message exchange -- meaning WSDL 
describing the application level and initial Reliable Message delivery 
would be essentially identical.  Importantly, the specification had not 
nailed things down enough to exclude either interpretation and was somewhat 
inconsistent on the implications of both.  The 1.086 document is 100% clear 
that the second model is the only interpretation.  An implementation 
following my model would no longer be conformant.

Further, the .992 document left some important characteristics of the RMP 
"signals" (messages that are not visible to the producer/consumer or 
described in an application level WSDL instance) open.  These 
characteristics have been clarified and restricted in the 1.086 document.

In short, .992 implementations may have gotten lucky and chosen 
interpretations which correspond to the 1.086 specification; however 
substantial technical changes have been made in the interim.

thanx,
	doug