Re: 2004-02-12 Action Item: Clarification to the OperationName feature

I must say, it's terrifically satisfying to me to see the WG finally
having this discussion, even if it's a few years late.  8-) It's going
very well too.

As this relates to my issue though, there's at least two more things
to keep in mind.

One is whether or not the operation name is in the message.  If the
WS-Arch WG was around, I'd raise this with them, and would hope that
they'd agree with me when I say that it should be best practice to
always have the operation name in the message (and by "message" I mean
all bits that are sent, including all underlying protocols (e.g.
HTTP, TCP/IP, Ethernet).  The only exception should be when the
operation name is self-descriptively associated with the message via
some public specification and/or registration (i.e. data arriving on TCP
port 9999 will always be bound to such-and-such an operation); this
isn't as extensible, but it's still self-descriptive, which is what I
believe to be most important.  I strongly believe that the operation
name must be specified in either of those two ways for any Internet
scale architecture (of course, you're all probably familiar with my
additional beliefs in this respect, but first things first 8-).

The other issue is *how* to identify the operation name, and this is
being discussed, but I don't think the breadth of the problem has
been recognized.  From what I can see, we're only currently
discussing the following cases;

- SOAPAction feature
- RPC style (GED)
- implicit (matching WSDL description to received messages)

(note that "implicit" goes against my belief described in the
paragraph above)

But I think we also need to tackle the cases where the operation
name is;

- inherited from an application protocol, or
- specified in a SOAP header, e.g. wsa:Action

Thanks.

Mark.

Received on Thursday, 26 February 2004 10:41:40 UTC