On May 31, 2011, at 1:44 AM, David Jencks wrote: > Locally I now have a server that runs eba-with-isolation using aries trunk > and fixes a lot of the tck problems that showed up with this version. > However due to a problem with aries proxies changing the SerialVersionUID if > it is not hardcoded on every serializable class I've had to add > SerialVersionUID to a few hundred classes across several projects. IMO most > of these classes shouldn't have hardcoded serialVersionUIDs because they are > not really intended for long-term compatibility and people will update the > classes and forget to update the SerialVersionUID. > > The new aries proxy code also seems to consume significantly more permgen > space. I didn't investigate in detail but a full server runs out of permgen > space around half way through starting with 128 MB permgen. My server > started OK with 1024 MB permgen.
I'd be interested to see how much more memory it's using... Do you have a new drop of source? In what cases will Aries be building a proxy? > > It's possible to get eba-with-isolation working with a patched 0.3 aries but > it looks like aries have made a backwards incompatible change to their > project structure that makes releasing bug fixes for 0.3 implausible. I read the aries list (pretty quickly). I would think that it would be possible to insert a 0.3.1 fix release and move their trunk to 0.3.2. But I've probably not thought enough about the problem? From the looks of things, I think we want to avoid their current trunk. > > To get the eba-with-isolation sample to work I had to add a lot of > import-packages for jasper, el, and other internal classes. This seems like > a bad idea.... I'm wondering if I missed something about how this is supposed > to work. I'd expect this to be a problem for jsps in a wab as well.... how > does the wab deployer deal with this? > > I'm not exactly sure how to proceed here... Is it possible to run without isolation? I'd very much rather see a more converged codebase. If we could run without isolation and minimize regression errors, that would be a big bonus, I think... --kevan
