David,

Can u please log a bug in bugzilla? patches are welcome as well.

Thanks,
dims

--- David Ekholm <[EMAIL PROTECTED]> wrote:
> > Hmm. 
> 
> > What do you mean by "share a common classpath"? They all refer to the same 
> > jar within an EAR? Or you added the class to the system classpath? 
> 
> In our system we don't pack each application to EARs, we leave them unpacked. Each
> application.xml refers to the same classpath, -a path to a directory
> 
> > I am not sure what you mean by classloader separation either. This is not 
> > mandatory, just one way to deploy applications. Therefore write access to 
> > statics is prohibited by EJB spec, because it would leave an application in 
> > an undefined state, as yours. 
> 
> Until the EJB specification matures we have found ourselves forced to rely on static 
> variables
> in some rare occations (write once, read many), knowing it is a bit unorthodox (can 
> cause
> problem with clustering for example if you do write many). I see no reason however 
> for AXIS to
> mix up the separation of classloaders encforced by Orion/Oracle. Of course having 
> separate
> classloaders is just one implementation, one could separate J2EE applications by 
> running them
> inside different VMs too, but that was not the way Orion/Oracle choose to do it. I 
> know we broke
> one rule by using a static variable, but the fact is that the EJB specification, (or 
> at least
> the implementation we use) doesn't well support some important things one needs to 
> be able to
> do, one of them being able to, from an EJB layer, look up the name/identifier of an 
> application
> that has been set in a way that is practical and safe from a deployment perspective. 
> We use a
> write once, read many static variable for this purpose. There may be a suitable 
> solution with a
> particuar EJB implementation, but we haven't found a good one for Orion/Oracle.
> 
> In short: AXIS is probably not breaking a rule here with its implementation, but 
> with an
> adjustment to how it uses classloaders, it would be more compatible with existing 
> J2EE
> solutions. -Solutions that may not be 100% orthodox, but solutions that look the way 
> they do for
> a good reason.
> 
> /David
> 
> 

> ATTACHMENT part 2 application/ms-tnef name=winmail.dat



=====
Davanum Srinivas - http://webservices.apache.org/~dims/

__________________________________________________
Do you Yahoo!?
Yahoo! Platinum - Watch CBS' NCAA March Madness, live on your desktop!
http://platinum.yahoo.com

Reply via email to