- From: Sanjiva Weerawarana <sanjiva@watson.ibm.com>
- Date: Fri, 26 Apr 2002 05:38:17 -0400
- To: <www-ws-desc@w3.org>
Hi Glen, IMO there's a semantic difference between "mustUnderstand" and "required." We may need both, but what WSDL 1.1 has is "required", of course. Sanjiva. ----- Original Message ----- From: "Glen Daniels" <gdaniels@macromedia.com> To: <www-ws-desc@w3.org> Sent: Thursday, April 25, 2002 12:24 PM Subject: FW: Open Content Model > > Oops! This should have gone to the list. --G > > -----Original Message----- > From: Jeffrey Schlimmer [mailto:jeffsch@windows.microsoft.com] > Sent: Thursday, April 25, 2002 11:54 AM > To: Glen Daniels > Subject: RE: Open Content Model > > > Glen, did you want this message to go to the list or just me? --Jeff > > -----Original Message----- > From: Glen Daniels [mailto:gdaniels@macromedia.com] > Sent: Thursday, April 25, 2002 8:29 AM > To: Jeffrey Schlimmer > Subject: RE: Open Content Model > > > Two comments about wsdl:required. > > 1) If in fact the purpose of this is to indicate that the processor who > is reading this document must comprehend the extension in question for a > valid interpretation of the WSDL, why don't we change it to > wsdl:MustUnderstand to mirror the SOAP attribute which indicates > precisely the same semantics? > > 2) Assuming these are the semantics, you can use this, as in SOAP, to > require understanding of anything, not just EIIs. Example: > > <operation name="foo" myNS:fancyAttribute="idempotent"> > ... > </operation> > <service> > <myNS:fancyExtension wsdl:MustUnderstand="true"/> > ... > </service> > > So a reader of this document would see the MU="true" on the > myNS:fancyExtension element, which means that they are required to > "understand" all semantics and rules indicated by the specification > which defines this element. In this case, I'm positing that said spec > indicates that you must process the "myNS:fancyAttribute" attribute when > dealing with operations to indicate various characteristics such as > idempotency. > > Elements like myNS:fancyExtension can be defined for fine-grained > extensions, and then it's easy to aggregate these into groups of > commonly used stuff which can be indicated with a single MU="true" > element, something like "<myNS:allMyExtensions>", whose specification > indicates that understanding it implies understanding "fancyExtesion", > "superExtension", and "megaExtension". > > Does this get us what we want here? A potential performance problem > with this is that you need to go and find all the required stuff before > doing any real work comprehending the document, since the semantics of > how you parse the rest might be affected. This also precisely matches > the SOAP 1.2 processing model, and allows for safe extensibility. > > --Glen > > > -----Original Message----- > > From: Jeffrey Schlimmer [mailto:jeffsch@windows.microsoft.com] > > Sent: Wednesday, April 24, 2002 9:41 PM > > To: > > Subject: RE: Open Content Model > > > > > > As appealing as it is, wsdl:required can only be used to annotate > > element information items (EII) but not attribute information items > > (AII). > > > > So, we're going to have to commit to either: (a) AII extensions must > > _always_ be understood, or (b) AII extensions must always be > > ignorable. > > Right? > > > > If we choose (a), we have a split rule for interpreting whether an > > extension is required "by default". That is, an EII without > > wsdl:required is ignorable but an AII without wsdl:required > > (which they > > all will be) must be understood. > > > > For the sake of consistency, seems like (b) is a better choice. Can we > > live with extension AII always being ignorable? > > > > --Jeff > > > > -----Original Message----- > > From: Jonathan Marsh > > Sent: Wednesday, April 24, 2002 4:20 PM > > To: www-ws-desc@w3.org > > Subject: RE: Open Content Model > > > > > -----Original Message----- > > > From: Sanjiva Weerawarana [mailto:sanjiva@watson.ibm.com] > > > Sent: Wednesday, April 24, 2002 1:26 PM > > > To: www-ws-desc@w3.org > > > Subject: Re: Open Content Model > > > > > > I would personally preferred to have open-content on > > selected elements > > > rather than saying "arbitrary XML can be inserted > > anywhere." That is, > > > I prefer architected open-content rather than flat-out openness. > > > > I think open content models are only a small factor in the total > > extensibility of the language. The main parts are the > > processing model > > and the conformance clauses. > > > > > Example: If we decide that the language should not have portType > > > (interface) inheritance, then having an open-content model would > > > allow one to easily supplant that restriction by adding an attribute > > > or element to <portType>. IMO that's bad. > > > > You can imagine lots of ignorable annotations on <portType> like > > my:lastUpdatedOn="20020424" my:lastUpdatedBy="jmarsh@microsoft.com". > > Ignorable annotations are the 90% case. How do you distinguish > > ignorable annotations from behavioral extensions? Perhaps an open > > content model together with a processing model like this > > would meet your > > concern: > > > > 1) extension attributes are allowed everywhere, but cannot change the > > behavior of elements or attributes described in the spec. > > 2) extension elements are allowed everywhere, but cannot change the > > behavior of elements or attributes described in the spec unless marked > > with wsdl:required="true". > > 3) it is an error to process a WSDL document with extension elements > > marked with wsdl:required="true" if the extension is not supported by > > the processor. > > > > > Jonathan: If I recall correctly XSLT is not fully open-content: > > > weren't there some restrictions on top-level elements and > > their impact > > > on the processing model? That's akin to what I would prefer; > > > architected extensionsibility with good rationale. > > > > XSLT has a mechanism for identifying extension elements, and > > distinguishing them from literal result elements inside of templates. > > Such extension elements are given an implicit > > required="true", that is, > > a processor that doesn't understand the extension must halt and catch > > fire. Outside of templates, where there is no ambiguity, arbitrary > > elements are allowed. Attribute extensibility is allowed everywhere. > > The rules given are pretty much like those I give above, although > > wsdl:required provide a granularity of control not available in XSLT. > > > > > Sanjiva. > > > > > > ----- Original Message ----- > > > From: "Keith Ballinger" <keithba@microsoft.com> > > > To: <www-ws-desc@w3.org> > > > Sent: Monday, April 22, 2002 10:06 PM > > > Subject: Open Content Model > > > > > > > > > One item I've had on my plate is to describe why the open content > > model > > > is something we should try to take advantage of as much as possible > > with > > > the WDSL revision. > > > > > > To begin with, when I say open content model, I mostly mean allowing > > > extra element and attributes from other namespaces within a schema. > > This > > > is typically done with the <any/> and <anyAttribute/> schema tags. > > These > > > elements also allow you to specify which namespace > > (including ##other, > > > which means any namespace but the one in the schema), as well as the > > > default processing of these elements. An example of this > > can be found > > > with the <binding> element in WSDL today. This element allows child > > > elements of the type: <any namespace="##other" minOccurs="0" > > > maxOccurs="unbounded" /> > > > > > > Working with SOAP and WSDL over the past few years has shown me the > > > value in allowing this open content model. It is very useful > > especially > > > for extending a specification in new ways. As authors of > > specs, we may > > > cover many use cases and requirements in a first class way, but we > > need > > > to provide for those other requirements that come to us > > that we don't > > > anticipate. The open content model allows us to handle many > > of these. > > > > > > I would also recommend that we keep WSDLs mustUnderstand feature for > > > extensions, the wsdl:required attribute. > > > > > > As a matter of technique, I feel that we should be overly open as > > > opposed to overly closed. This would lead to putting this > > open content > > > model on all elements, and then finding reasons why it > > shouldn't be on > > > one. > > > > > > Cheers, > > > Keith > > > > >
Received on Friday, 26 April 2002 05:41:16 UTC