On Mon, May 16, 2005 at 09:10:48PM -0400, Brian Kelley wrote: > Hey, you were the one that noticed the leaks! Just realized that... > What is your platform?
heh, I think I was actually wrong but I didn't post a followup to the list. Until recently compiling metakit under a debug build of python would cause a coredump. This is no longer the case but I think it still isn't correct - it just doesn't coredump any more. The interaction of the C++ destructors and the python refcounter and garbage collection (I think the wrappers were written before mark & sweep collection came into python) is mystical. There are only a few objects that need exposing to python (storage, view, and rows) and they should be straight forward to wrap in a light python class with just a pointer to the real C++ class. Something in the current hybrid breaks sometimes in a way that is hard to find and the hybrid doesn't buy much in performance. If you are using metakit from python you already kissed optimizations in the 2% range. -jack My platform is linux (debian) with a 2.4.22 kernel, gcc 2.95.4 Nothing is certain, but when I started using metakit the only other change I made was the python version (2.2 to 2.4). I have two sets of apache processes running the same code, one to handle recording (uses metakit and MySQLdb) and one that does agregate reporting (uses MySQLdb and gdmodule). Only the metakit using one ever coredumps - and only after handling a few hundred requests. _____________________________________________ Metakit mailing list - Metakit@equi4.com http://www.equi4.com/mailman/listinfo/metakit
