Re: PROPOSAL: Drop interface/operation/(input|output)/@headers

Wow, what a shock!

The immediate SOAP proponent in me says no, we can't really just drop 
headers! I haven't ever considered SOAP header blocks as just 
"middleware"; I know some do.

Of course, the feature argument sounds compelling... although it opens 
the door to features being dropped in a (hypothetical) second step, 
which I would oppose.

Obviously, I need to think more about this. I am generally in favour of 
simplification, but I also like flexibility.

Jonathan, since I will be away on Thursday, and since Youenn will have 
just come back from paternity leave, can we please differ discussion of 
Sanjiva's proposal until next week?

Thank you,

Jean-Jacques.

Sanjiva Weerawarana wrote:

> During the <message> elimination process, I proposed that it be
> eliminated by introducing a single element that may go as the 
> body of the message and zero or more "header" elements. The motivation
> was that it was possible in WSDL 1.1 to have such header and body
> parts and we didn't want to lose that functionality.
> 
> The main use of the "header" parts are to enable one to use WSDL
> to describe middleware protocol type "applications" of WSDL. That
> is, one can imagine protocols requiring certain headers to be 
> present etc..
> 
> However, to fully describe such protocols the header stuff has to
> be much much richer. If you look at the original context proposal
> I made back in January this year, you'll see some of that richness,
> but even that is not enough.
> 
> At the same time, complicating WSDL to the extent needed to make
> it possible to fully describe a handful of middleware protocols
> when compared to the millions of "regular" applications doesn't
> seem like the right tradeoff. 
> 
> Thus, we now propose we drop the "headers" concept from input and
> output messages. Messages will simply be a single XML element. 
> 
> I also propose a simple syntactic change to exploit this new 
> simplified structure of messages. Rather than the current form:
> 
>     <input messageReference="xs:NCName" 
>            body="xs:QName"? headers="list of xs:QName"?/>
> 
> I suggest we use:
> 
>     <input messageReference="xs:NCName" 
>            message="xs:QName"/>
> 
> And similarly for <output>. The same syntactic approach can be
> used to make <fault>s consistent with this approach. Rather the
> current form:
> 
>     <infault messageRefernce="xs:NCName" details="xs:QName"/>
> 
> I suggest we use:
> 
>     <infault messageReference="xs:NCName" message="xs:QName"/>
> 
> And similarly for <outfault>.
> 
> Thus, this email proposes the following changes to status quo:
>     - drop interface/operation/(input|output)/@headers
>     - rename interface/operation/(input|output)/@body to
>       interface/operation/(input|output)/@message
>     - rename interface/operation/(infault|outfault)/@details to
>       interface/operation/(infault|outfault)/@message
> 
> Sanjiva.
> 
>

Received on Tuesday, 21 October 2003 12:25:20 UTC