axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Thomas Sandholm <sandh...@mcs.anl.gov>
Subject Re: WSDL4J import problem
Date Wed, 06 Mar 2002 16:12:42 GMT
While you are poking at this I'd like to mention a related problem we hit 
with WSDL imports.

If we use relative urls in the import location like ../../bindings.wsdl, we 
run into problems when importing from multiple locations. For example 
WSDL4J (or WSDL2Java) does not recognize that a ../bindings.wsdl and 
../../bindings.wsdl import may actually refer to the same file (and thereby 
target namespace).

This results in the import being performed twice with multiply defined 
symbol errors as a result. If we organize our schemas so that all imports 
of a particular file are done with the exact same location string (e.g. 
always ../../bindings.wsdl), we don't hit this problem.

/Thomas


At 08:24 AM 3/6/2002 -0600, Russell Butek wrote:
>Is the fact that everything is in the same namespace clouding the issue?
>OK, give each WSDL file its own namespace:  binding.wsdl:  "uri:diy.b";
>porttype.wsdl:  "uri.diy.pt".  I still see the same problem.
>
>Definition def; // the definition for tutorial.wsdl
>PortType portType = def.getPortType(new QName("uri:diy.pt",
>"tutorialSoap"));
>Binding binding = def.getBinding(new QName("uri:diy.b", "tutorialSoap"));
>PortType portTypeInBinding = binding.getPortType();
>
>portType and portTypeInBinding are DIFFERENT.  The Definition for
>tutorial.wsdl contains 2 different PortType objects with the same QName!
>HOW CAN THIS BE CONSIDERED GOOD?!?
>
>Russell Butek
>butek@us.ibm.com
>
>
>
>To:    Russell Butek/Austin/IBM@IBMUS
>cc:    axis-dev@xml.apache.org, Francisco Curbera/Watson/IBM@IBMUS, Matthew
>        Duftler/Watson/IBM@IBMUS, Sam Ruby/Raleigh/IBM@IBMUS
>From:  Sanjiva Weerawarana/Watson/IBM@IBMUS
>Subject:    Re: WSDL4J import problem  (Document link: Russell Butek)
>
>Hi Russell,
>
>As Paco said in his reply, the issue here is a WSDL interpretation question
>and you heard what the intent of the authors was. If you want to call that
>intent wrong then you can consider the WSDL4J behavior a bug. Otherwise it
>works the way it should.
>
>Sanjiva.
>
>Russell Butek
>03/06/2002 07:12 PM
>
>To:    axis-dev@xml.apache.org
>cc:    Francisco Curbera/Watson/IBM@IBMUS, Matthew
>        Duftler/Watson/IBM@IBMUS, Sam Ruby/Raleigh/IBM@IBMUS, Sanjiva
>        Weerawarana/Watson/IBM@IBMUS
>From:  Russell Butek/Austin/IBM@IBMUS
>Subject:    Re: WSDL4J import problem  (Document link: Sanjiva Weerawarana)
>
>axis-dev folks, we have a problem with WSDL4J that affects WSDL2Java
>imports.  I presented the problem to the WSDL4J folks, but I lost my
>argument.  I'm posting this chain of mail here to see whether anyone thinks
>I should reopen my argument (I still think it's a bug), or whether I should
>just drop it.  The first note in the chain was posted to axis-dev a few
>weeks ago.  You can work bottom up through the discussion.
>
>Russell Butek
>butek@us.ibm.com
>
>
>Russell Butek
>02/22/2002 07:24 AM
>
>To:    Francisco Curbera/Watson/IBM@IBMUS
>cc:    Matthew Duftler/Watson/IBM@IBMUS, Sam Ruby/Raleigh/IBM@IBMUS,
>        Sanjiva Weerawarana/Watson/IBM@IBMUS
>From:  Russell Butek/Austin/IBM@IBMUS
>Subject:    Re: WSDL4J import problem  (Document link: Russell Butek)
>
>Sam, are you OK with this?  It was your WSDL that found the problem.
>
>I still think it's wrong.  The Definition for tutorial.wsdl should have all
>the pertinent info and all references should be properly resolved.  It
>really really bothers me that asking for a Binding's PortType, which
>happens to be named "uri:diy:tutorialSoap", I get a different PortType then
>if I asked the Definition that contains that Binding for PortType named
>"uri:diy:tutorialSoap".
>
>Russell Butek
>butek@us.ibm.com
>
>
>Francisco Curbera
>02/21/2002 08:37 PM
>
>To:    Russell Butek/Austin/IBM@IBMUS
>cc:    Matthew Duftler/Watson/IBM@IBMUS, Sam Ruby/Raleigh/IBM@IBMUS,
>        Sanjiva Weerawarana/Watson/IBM@IBMUS
>From:  Francisco Curbera/Watson/IBM@IBMUS
>Subject:    Re: WSDL4J import problem  (Document link: Russell Butek)
>
>Russel,
>
>The interpretation that Matt is describing follows the semantics of
><import> in XML Schema, which was the model for WSDL <import>. There you
>only import definitions from namespaces different from the one the document
>belongs to. There are a few differences, in fact, because in XML Schema the
>"location" is optinal... So WSDL is at least little more helpful.
>
>It is true that the spec is not very explicit, but I take that to be the
>correct interpretation of the WSDL spec (all 4 original authors agree on
>this).
>
>The problem your example brings up is that of splitting a single
>"targetNamespace" into multiple documents. The correct mechanism to deal
>with that one would be the equivalent of XML Schema <include>. This may be
>a good addition to WSDL, but it is not supported at this time.
>
>Matt's interpretation is the only one consistent with WSDL 1.1 in my view.
>
>Regards,
>
>Paco
>
>Russell Butek
>02/21/2002 06:47 PM
>
>To:    Matthew Duftler/Watson/IBM@IBMUS
>cc:    Sam Ruby/Raleigh/IBM@IBMUS, Sanjiva Weerawarana/Watson/IBM@IBMUS,
>        Francisco Curbera/Watson/IBM@IBMUS
>From:  Russell Butek/Austin/IBM@IBMUS
>Subject:    Re: WSDL4J import problem  (Document link: Francisco Curbera)
>
>If this is not a bug (I still think it is), then it is inconsistent.  See
><rjb>...</rjb>
>
>Russell Butek
>butek@us.ibm.com
>
>
>
>MATTHEW DUFTLER
>02/21/2002 03:54 PM
>
>To:    Russell Butek/Austin/IBM@IBMUS
>cc:    Sam Ruby/Raleigh/IBM@IBMUS, Sanjiva Weerawarana, Francisco
>        Curbera/Watson/IBM@IBMUS
>From:  Matthew Duftler/Watson/IBM@IBMUS
>Subject:    Re: WSDL4J import problem  (Document link: Russell Butek)
>
>Hi Russell,
>
>Ok, I've worked through what you've described, and I don't believe it's a
>bug.
>
>I've also discussed it with Sanjiva and Paco to make sure that we all have
>the same understanding of the import semantics in WSDL.
>
>Since all three files define items in the same namspace, you should
>theoretically be able to refer to any item in one file, from the items in
>the other files. Basically, it is legal WSDL to refer to the portType from
>the binding file, even though the binding file does not import the portType
>file. The same three WSDL files would also be perfectly legal if they
>contained no import statements whatsoever. However, a WSDLReader would have
>no way to resolve referenced items, unless they were defined in the same
>file. For any system to support referring to items contained in the same
>namespace as the referring definition, without explicit import statements,
>the system must have some out-of-band information.
>
><rjb>
>I agree with these statements when they refer to the Definitions in
>binding.wsdl.  The items in that Definition refer to items that are
>undefined.  Specifically, binding.wsdl's Definition contains a Binding
>object which refers to an undefined PortType.  That makes sense to me since
>there's no import in binding.wsdl for it to find the portType.
>
>But what about tutorial.wsdl's Definition?  All the necessary info is
>available.  The Definition contains a Binding.  That Binding refers to a
>PortType.  If both imports are in tutorial.wsdl, that PortType is
>undefined.  If the portType.wsdl import is moved to binding.wsdl, then that
>PortType IS defined.  Inconsistent.
>
>I understand what's happening in tutorial.wsdl's Definition.  Its Binding
>object is the one from binding.wsdl's Definition.  And it's PortType object
>is the one from portType.wsdl's Definition.  Unfortunately, what that means
>is that the Binding in the Definition refers to a PortType that is NOT in
>the same Definition.  Bad news.
></rjb>
>
>As it stands today, the WSDLReader provided with WSDL4J does not hand off
>one definition's imports to another to serve as out-of-band information.
>
><rjb>
>It certainly seems to when the imports are chained (tutorial.wsdl imports
>binding.wsdl imports portType.wsdl).
></rjb>
>
>  In the case you describe, the reader would have to hand off the imports
>from tutorial.wsdl to be used as out-of-band information when processing
>binding.wsdl.
>
>I suppose this is a feature that could be added to WSDL4J. If we were to
>add this, I think we would also have to make sure that the only imported
>documents which are handed off are ones which define items for the same
>targetNamespace as the definition the documents are being handed off to
>(since all namespaces being referred to must be explicitly imported, with
>the exception of the targetNamespace).
>
>Thanks,
>-Matt
>
>
>
> 
>
>                       Russell 
> Butek 
>
>                                                To:       Matthew 
> Duftler/Watson/IBM@IBMUS, axis-dev@xml.apache.org
>                       02/16/2002 
> 09:58         cc: 
>
>                       AM                       From:     Russell 
> Butek/Austin/IBM@IBMUS
>                                                Subject:  WSDL4J import 
> problem
> 
>
> 
>
> 
>
>
>
>
>Sam, this is the bug you found.
>
>Matt, we have an import scenario where a binding's portType is undefined
>when it shouldn't be: .
>
>tutorial.wsdl
>---------------
><import ... location="portType.wsdl">
><import ... location="binding.wsdl">
><service...>
>
>portType.wsdl
>-----------------
><portType ...>
>
>binding.wsdl
>---------------
><binding ...>
>
>When I run "java Test tutorial.wsdl"  I get:
>
>Retrieving document at 'tutorial.wsdl'.
>Retrieving document at 'porttype.wsdl', relative to 'file:tutorial.wsdl'.
>Retrieving document at 'binding.wsdl', relative to 'file:tutorial.wsdl'.
>definition = uri:diy:tutorial.wsdl
>Binding portType isUndefined?  true
>PortType isUndefined?  false
>
>definition = uri:diy:portType.wsdl
>PortType isUndefined?  false
>
>definition = uri:diy:binding.wsdl
>Binding portType isUndefined?  true
>PortType isUndefined?  true
>
>Notice that whereever the binding refers to a portType, that portType is
>undefined, even though that portType DOES exist.  I can understand that the
>binding.wsdl's definition's binding's portType is undefined, but the
>tutorial.wsdl's definition's binding's portType should not be undefined.
>
>If I change the wsdl files slightly, the portType isn't undefined:
>
>tutorial.wsdl
>---------------
><import ... location="binding.wsdl">
><service...>
>
>portType.wsdl
>-----------------
><portType ...>
>
>binding.wsdl
>---------------
><import ... location="portType.wsdl">
><binding ...>
>
> > java Test tutorial.wsdl
>Retrieving document at 'tutorial.wsdl'.
>Retrieving document at 'binding.wsdl', relative to 'file:tutorial.wsdl'.
>Retrieving document at 'porttype.wsdl', relative to 'file:binding.wsdl'.
>definition = uri:diy:tutorial.wsdl
>Binding portType isUndefined?  false
>PortType isUndefined?  false
>
>definition = uri:diy:binding.wsdl
>Binding portType isUndefined?  false
>PortType isUndefined?  false
>
>definition = uri:diy:portType.wsdl
>PortType isUndefined?  false
>
>Russell Butek
>butek@us.ibm.com
>
>#### import.problem.zip has been removed from this note on February 21 2002
>by Matthew Duftler




Thomas Sandholm <sandholm@mcs.anl.gov>
The Globus Project(tm) <http://www.globus.org>
Distributed Systems Laboratory
Mathematics and Computer Science Division
Argonne National Laboratory


Mime
View raw message