Hi Tom

I spent about a week playing around with axis to see why it wasn't working
for me, and simplifying test cases to find out the causes.  I did post a
couple of bug reports on bugzilla, but no-one has even commented on them
yet.  I really did want to stick with axis, but in the end I had to make a
decision which was expedient for my project, which was to generate the WSDL
using .NET.  I can't afford to spend the time investigating and writing bug
reports for axis when there is another solution which works perfectly for
me.

I didn't report any of the other bugs primarily because they were too
numerous, and it takes a significant amount of time to write up a bug report
that is truly useful to the developer.  They were also quite fundamental, so
would be easily detected by anyone else creating even the simplest of
services.  This also makes me think they may have already been addressed in
the nightly builds.

Nightly builds bring up a whole new area of gripes.  Firstly, it always
turns out to be a major pain, involving a fair amount of guesswork, to
upgrade to even a release version of axis.  This is because our project uses
jakarta projects such as torque (which btw we are now quite desperate to
abandon), and velocity, and each expects different versions of the commons
libraries, xerces and log4j.  The problem is that none of the jakarta jar
libraries contain any version information whatsoever, so it is always
guesswork to figure out which one is using the latest version, and very
often they are not even backwards-compatible.  Axis itself is no better in
this regard.  The jar file is always called axis.jar, with no version
suffix, and the manifest file never includes a version number either.  I am
not willing to use a nightly version of axis in production, and only use
beta versions of tools when there is no alternative.  What would be great is
if there were a CVS branch created after 1.0, which included only bug fixes
that are then released in "service packs".  I know this means more work for
you guys, but in the real world you can't expect companies to use nightly
builds unless they are heavily involved in the product's development like
Macromedia is.

Even writing a response like this takes up valuable development time (which
I can afford at the moment because the server just kicked the bucket!).  I
don't apologise for not taking more time to contribute to axis, it's just a
reality.

Martin



----- Original Message -----
From: "Tom Jordahl" <[EMAIL PROTECTED]>
To: <[EMAIL PROTECTED]>
Sent: Saturday, November 23, 2002 5:53 AM
Subject: RE: Document style web services


>
> Martin,
>
> Can you please try Java2WSDL with the latest nightly build and report any
bugs you find in Bugzilla?
> We want this to work!  I hope you will find that the latest source fixes
most (all?) of the major problems.
>
> Thanks
> --
> Tom Jordahl
> Macromedia
>
>
> -----Original Message-----
> From: Martin Jericho [mailto:[EMAIL PROTECTED]]
> Sent: Thursday, November 21, 2002 5:32 PM
> To: [EMAIL PROTECTED]
> Subject: Re: Document style web services
>
>
> Dennis,
>
> My experience is that Java2WSDL in Axis 1.0 has too many bugs to generate
> document/literal style WSDL, but if you can generate it by some other
means,
> the WSDL2Java and bean marshalling seem to work fine.
>
> The reason you can't have multireferencing in document style calls is
> because the document is validated against the schema.  If you define the
> schema to allow IDs and REFs on every element, you can implement the
> multirefs yourself, but this would make your schema virtually unreadable,
> very complicated, and probably less robust.
>
> Martin Jericho
>
> ----- Original Message -----
> From: "Dennis Sosnoski" <[EMAIL PROTECTED]>
> To: <[EMAIL PROTECTED]>
> Sent: Friday, November 22, 2002 7:03 AM
> Subject: Re: Document style web services
>
>
> > Hi Anne,
> >
> > Does Axis support automatic marshalling of document-style messages? I
> > was under the impression it does not, which was why I suggested a
> > DataBindingProvider might be useful to add this support. I agree that
> > document-style is a better approach for the future, though I'd hardly
> > call it a "predominant consensus" at this point. AFAIK document style
> > interfaces are not as widely supported as RPC style, though, and I'm
> > surprised to see your statement that most SOAP implementations support
> > automatic marshalling for document style. Can you give me any figures
> > for this?
> >
> > As for "no problem building automatic serializers" I have to disagree. A
> > Schema definition does not, in general, provide enough information to
> > directly map to Java data structures. If you use an approach where the
> > data structures are either pre-generated from the Schema or constrained
> > to obey a predefined mapping to and from the Schema you can get around
> > this, but that's hardly automatic.
> >
> > I'm also puzzled by your statement that it's difficult handle
> > multi-referencing object structures using document style. Is there a
> > reason this can't be handled with ID/IDREF or key/keyref links?
> >
> > Thanks,
> >
> >   - Dennis
> >
> > Dennis M. Sosnoski
> > Enterprise Java, XML, and Web Services Support
> > http://www.sosnoski.com
> >
> > Anne Thomas Manes wrote:
> >
> > >Dennis,
> > >
> > >This is a pretty antiquated view of document style. Document style is
no
> > >longer used just for XML messaging. Most SOAP implementations support
> > >automatic marshalling of both RPC-style and document-style messages. As
> long
> > >as you have a WSDL description of the message structure, there's no
> problem
> > >building automatic serializers.
> > >
> > >The predominant consensus in the industry at this point is to use
> > >document-style by default. Document style is much easier to validate,
> > >transform, and manipulate. The primary reason to consider using
> rpc/encoded
> > >is if you need to send multi-referencing object structures. SOAP
encoding
> > >does a really nice job marshalling these structures. It's much harded
to
> > >represent them using literal XML Schema. But if you're not using
> multi-refs,
> > >it's a better practice to use document-style.
> > >
> > >Regards,
> > >Anne
> > >
> > >
> > >
> > >>-----Original Message-----
> > >>From: Dennis Sosnoski [mailto:[EMAIL PROTECTED]]
> > >>Sent: Thursday, November 21, 2002 1:25 PM
> > >>To: [EMAIL PROTECTED]
> > >>Subject: Re: Document style web services
> > >>
> > >>
> > >>Hi Matt,
> > >>
> > >>The whole point of document style is that your application gets passed
> > >>the XML message payload as XML document fragments. See the "message"
> > >>sample for an example of this. With a document style interface your
> > >>class would look like:
> > >>
> > >>public class SomeXMLService {
> > >>    public Element[] someXMLMethod(Element[] elems) {
> > >>        ...
> > >>    }
> > >>}
> > >>
> > >>If you want to convert the XML into objects you need to do it
yourself,
> > >>perhaps using a framework such as Castor (http://www.castor.org). I
know
> > >>there's been some integration of Castor with Axis, though I think this
> > >>was for custom serialization with RPC style.
> > >>
> > >>This brings up an interesting point, though. Why not have a Java
> > >>DataBindingProvider as a replacement for the MsgProvider? This should
> > >>allow easy use of document style while converting seamlessly between
XML
> > >>and objects without the application needing any special code. I'm
> > >>looking into some data binding code currently, perhaps I'll see if I
can
> > >>work in this direction.
> > >>
> > >>  - Dennis
> > >>
> > >>Dennis M. Sosnoski
> > >>Enterprise Java, XML, and Web Services Support
> > >>http://www.sosnoski.com
> > >>
> > >>
> > >
> > >
> >
> >
>

Reply via email to