axis-java-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Matt Bowersox" <mbowe...@rochester.rr.com>
Subject Re: Could we please take a second and discuss AxisClassloader (3rd attempt)
Date Fri, 01 Feb 2002 03:04:31 GMT

>
> Actually, the question is precisely the reverse - i.e. I should be asking
you that about Dynamo.
>
> Here's JRun's classloader structure:
>
>         System Classloader
>                 |
>         Server Classloader
>         /               \
>    Webapp1 CL        Webapp2 CL
>
> Axis gets loaded by the Server classloader, but the Webapp classloaders
are the ones who have the WEB-INF/classes dirs on the classpath.  So
clearly, all the classes loaded by the Server classloader are available in
all the Webapps, but NOT vice versa.  And new
Axis*().getClass().getClassLoader() will return the Server classloader for
us.
>
> Sooo... your turn. :)  How do you think this works in Dynamo such that the
current thread's classloader can't find the classes in WEB-INF?  In other
words - if a new Axis*() works for you, one would think the current WEB-INF
(since that's where your axis.jar lives) would be somewhere in the
classloader hierarchy starting at the current thread one.... no?
>
> --Glen
>

I admit that my understanding of dynamos internals is not sufficient enough
to do anything but guess,  but if pressed I would say it is due to the fact
that j2ee is just one modality for deploying servlets in dynamo  and the
classloader heirarchy is deaper in atg's framework. When I debugged this
stuff  the classname of the classloader I retrieved from the thread context
was an instance of an atg.somepackagename.CachingReusableClassloader,
whereas  the one retrieved from a newly created axis objects was a
com.sun.applauncher type classloader implementation. Dynamo's architecture
is very copmplicated. j2ee development modalities  coexists with their
nucleus bean/servlet component framaework (which is very very cool albeit
proprietary...  fully typed indexed and bound bean properties from jhtml/jsp
to nucleus servlet beans  etc ). A given dynamo instance can have multiple
j2ee containers each with multiple enterprise  applications etc. The end
result being that the webapp classloader was the only one I could use load
new axis classes.

Dynamo's unorthadox architecture not withstanding,   I did a medium
intensity perusal through the various  j2ee/ servlet specs and determined
that while the semantics of the classloader is discussed,  I did not find
specific guidance as to how a reference to the web containers classloader
should be obtained. and as I previously stated atg is j2ee certified further
emphasising that sun is not testing for classloader compliance to a
particular implementation specification at this granularity.

Do you think I should poke deeper into the specs  or talk to sun/atg?  or
would a discusions of a path forward be more apropriate? I am ok with
continuing to patch axis while it nears a more stable release version
however I think it would be nice if axis could be made to run unaltered on
all certified app servers. I am also curious if this going to be an atg only
problem. I would hate to see  the need for an appserver interop matrinx in
addition too to the soap interop matrix ;-)




Mime
View raw message