Andi Vajda <va...@apache.org> wrote: > > On Mar 5, 2010, at 13:48, Bill Janssen <jans...@parc.com> wrote: > > > The problem Python has with multiple cores and multi-threading, > > pointed > > out by Dave Beazley last year (http://www.dabeaz.com/python/GIL.pdf), > > looks like it will still persist for the next three or four years, at > > least, on OS X. Incredible but true. I'm presuming that PyLucene > > also > > suffers from it, though the interaction with Java threads makes me > > wonder a bit. > > > > Anyone know whether this is so? > > Any thread that enters java releases the GIL. When the execution > returns (or enters) python again, the GIL is reacquired. > > In other words, the answer to your question depends on how much time > your threads are spending in the JVM.
So, yes, that Python will still suffer from the issue if it has lots of CPU-bound threads fighting to get scheduled, but time spent in the JVM counts as I/O time, not computation time. OK, that's useful. > > And, a follow-up question: if one were to build a version of Python > > 2.6 > > with the "newgil" patch in http://bugs.python.org/issue7753 applied, > > would it break jcc? > > If the GIL and threading APIs didn't change I wouldn't expect any > issues. The whole point of that patch is to backport Antoine Pitrou's new GIL from Python 3.2 to Python 2.5 - 2.7. See http://www.dabeaz.com/python/NewGIL.pdf for an overview. So the GIL changes. But perhaps not in a way that affects jcc. Bill
