For something that seems so simple, this is turning into a quagmire. The web service binding processor is not the best place to test intents because the builder obviously has not yet run and propagated intents down to the binding. It would only be able to test the intent on the binding itself.
The other option is to do the selection in the reference binding provider, and actually it does happen that way now. By the point where the provider gets control, the binding's port is null. Axis2ReferenceBindingProvider has code to select the port. Unlike the web service binding provider, it doesn't just pick the first. It gives preference to the first port with a SOAP 1.1 address element, and it can't find one it takes the first SOAP 1.2 port. How is the binding's port null in the provider if the processor previously selected the first port? Well, WSDLServiceGenerator tests if the user WSDL provided a port by calling binding.getPortName(). Since the binding model is still marked unresolved, it returns null (this is wsdl.service so there is no port name). This causes WSDLServiceGenerator to import all the bindings and set the binding's port to null. Why is the binding model still unresolved? Well, the processor's resolve operation never marks it resolved. So, if the provider already has to select the port, why not have it use the SOAP intent to drive a selection? Well, when the binding processor selected the first port, it set the binding uri to that port's address. Then when WSDLServiceGenerator copies the ports over to the wrapper WSDL, it stores the binding uri into the port address. So the address to use is clobbered. Ok, let's change the binding processor to not select a port for wsdl.service since the provider's going to choose it anyway. Well, when I tried this, I got a NoSuchElementException in WebServiceBindingImpl.setIsDocumentStyle(). The binding is null, so it looks for the first WSDL Message in the Definition to determine the document style. In my case the main WSDL document has no Messages of its own but imports them from another file. I suppose this is a problem that could be hit in other ways and I just got unlucky. I guess I can continue to poke away at this, but I'm beginning to wonder if this functionality is worth the effort. On Wed, Feb 15, 2012 at 12:53 PM, Simon Laws <[email protected]>wrote: > On Wed, Feb 15, 2012 at 4:58 PM, Greg Dritschler > <[email protected]> wrote: > > When a web service binding uses wsdl.service, WebServiceBindingProcessor > > picks the first port. > > > > if (model.getPortName() != null) { > > port = > > service.getElement().getPort(model.getPortName()); > > } else { > > // BWS20006 - no port specified so pick the first > > one > > port = > > (Port)service.getElement().getPorts().values().iterator().next(); > > } > > > > What if the reference requires SOAP.v1_1 or SOAP.v1_2? Shouldn't it > pick a > > port that uses a matching SOAP binding? The web services binding > > specification says: > > > > 139 If the binding is for an SCA reference, the set of available ports > for > > the reference consists of the > > 140 ports in the WSDL service that have portTypes which are compatible > > supersets of the SCA > > 141 reference as defined in the SCA Assembly Model specification > > [SCA-Assembly] and satisfy all > > 142 the policy constraints of the binding. > > Greg > > Sounds right to me. > > Simon > > -- > Apache Tuscany committer: tuscany.apache.org > Co-author of a book about Tuscany and SCA: tuscanyinaction.com >
