On Nov 27, 2007 1:59 PM, Robert Dunne <[EMAIL PROTECTED]> wrote: > Hi, > > Newton's been mentioned a couple of times in this thread, so I thought I'd > write a quick note to clear up what its approach to RMI does and does not > do. > > Thankfully, we haven't had to reimplement all of RMI. Under the covers all > we've done is change the way RMI deals with classloaders and codebases, to > make it more OSGi friendly. Also, our RMI work doesn't involve any bytecode > manipulation. > > Our early work made it possible to associate a codebase with a bundle using > an new manifest header. Newton's deployment process takes care of making the > codebase jars available, so that there's no separate stage in which you have > to put jars on an HTTP server etc. We still use this approach, although we > are gradually phasing it out. > > Our current approach is based on bundles only, rather than special codebase > jars. In this approach bundle and package integrity are never violated so, > for example, it doesn't involve transfer of classes independent of their > bundles or dynamic generation of new bundles. > > Keeping it very brief, we use codebase annotations to encode the identity of > the bundle or public package from which each class comes. During > deserialization this information is used to decide which classloader to use, > and to install any missing bundles. > > In our view an ideal solution would require a couple of changes to the way > OSGi works, so we've outlined the issues as we see them in an RFP submitted > to the OSGi EEG. Similar issues apply to other widely used technologies, > such as ORM based persistence, so we're hopeful that a future OSGi > specification will tackle them. > > Regards, > > Robert
Thanks for the precisions. I clearly need to look more into Newton to stop sounding stupid, now that I don't risk reproducing bits of it into SmartFrog anymore. My only excuse is that your EclipseCon '07 slides, without the voice, were a bit misleading :) Could you tell us a bit more about how you coped with the problem I outlined earlier ? As I understand it, you found a way to deal with installing needed bundles and their dependancies just by using a custom RMIClassLoaderSpi implementation ? -- Olivier Pernet We are the knights who say echo '16i[q]sa[ln0=aln100%Pln100/snlbx]sbA0D4D465452snlbxq'|dc _______________________________________________ OSGi Developer Mail List [email protected] http://www2.osgi.org/mailman/listinfo/osgi-dev
