On Sun, 2005-04-17 at 22:12 -0700, Brian Stansberry wrote: > I'm concerned about the memory leak problem with the > impl jar, but it looks like truly solving that will > have to wait for the addition of something like > Simon's ContextClassLoaderLocal to the JDK. (BTW, any > news on that?)
Well, Robert suggested I try to raise some interest in submitting it as an Apache suggestion rather than direct from me. I sent it on direct to Henri Yandell (Jakarta PMC chair) to see what he thought of the idea. If I don't get any response there, I'll post direct to the Jakarta pmc list. However it occurred to me that Robert's suggestion of splitting LogFactory into a true interface/implementation might give a way forward anyway. This would allow the interface to be deployed in a shared classloader, but allow the class which actually declares the static member to be deployed on a per-webapp basis. This would give us back per-webapp singleton behaviour. I need to actually check out the DOM_QUIXOTE branch of logging to see whether my understanding of Robert's emails is correct.... > > I also owe a patch to the user guide discussing best > practices re not caching Log in a static field. If no > one objects, I can also include comments on JDK 1.1 > support. That would be cool. However I would like to state my personal opinion here: that deploying *anything* in a shared classloader sucks, and that anyone who does such a thing deserves what they get! Doing this really seems to me to be optimisation-over-correctness, and I don't see any justification for this. Regards, Simon --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
