RE: new issue: drop {direction} property from message reference components

I agree the issue should not be closed for similar reasons; during the 9
Jan 2003 teleconference [1] and 20 Jan F2F [2], we carefully discussed
what information a WSDL processor should be able to infer if it didn't
recognize a pattern URI and decided to keep direction.

Two scenarios that are enabled by WSDL 1.1 and the status quo (even sans
pattern) are direction-based filtering by firewalls and monitoring. I'll
speak to the first; as we know, a firewall may allow initiating a TCP
connection from inside while preventing it from outside because it
distinguishes the direction of the TCP initiation message. Next
generation firewalls will want to do something comparable for Web
services; given the WSDL for a service, the firewall will want to allow
only input messages in (analogous to TCP connection initiation) and only
output messages out (to prevent Trojan horse-related attacks). Such
firewalls will _typically_ understand the pattern URI, but will want to
provide a base level of filtering based on the XML Schema for the
messages and direction otherwise.

However, I do agree that the direction information is redundant between
the operation children and the values specified within the pattern URI
definition; if that redundancy bothers people, they we should consider
removing the other source.


[2] <> 

> -----Original Message-----
> From: []
> Behalf Of Roberto Chinnici
> Sent: Monday, September 29, 2003 4:14 PM
> To: Sanjiva Weerawarana
> Cc:
> Subject: Re: new issue: drop {direction} property from message
> components
> I don't think we should close this issue until we have addressed how
> the properties of the message reference component are filled in
> automatically (when omitted in the syntax) using information from
> the pattern.
> For instance, suppose I run into a pattern I don't know. Then, per
> proposal, I'd build a component model containing some message
> components whose direction is unknown, thus losing some of the
> information present in the WSDL document.
> Roberto
> Sanjiva Weerawarana wrote:
> > The current draft component model has a {direction} property for
> > message reference components [1]. I propose that we drop this
> > property as in reality message references do not have such a
> > property: the direction the message flows in is *completely*
> > defined by the pattern of which this message is a part of.
> >
> > I had edited this out earlier but put it back in at the chair's
> > suggestion that this is a substantial change and hence required
> > WG approval. So, please approve removing this redundancy.
> >
> > Note that this DOES NOT affect the syntax at all. It will change
> > the wording on <input> etc. to say that the direction of the
> > message (from the pattern) must be consistent with the local
> > name of the EII ("input" etc.).
> >
> > Sanjiva.
> >
> > [1]

Received on Wednesday, 1 October 2003 22:15:06 UTC