On Tue, Jun 10, 2008 at 9:58 AM, Thomas Heller <[EMAIL PROTECTED]> wrote: > Martin v. Löwis schrieb: >> I just finished a PEP 3121 implementation, uploaded at >> >> http://codereview.appspot.com/1962 >> >> It contains all the API changes proposed in the PEP, >> and updates all modules to conform to a minimal level >> of the API. I'd like to commit the change later today, >> so that it still makes the first beta. >> >> There is more that can be done about using the API, i.e. >> changing modules to actually stop keeping global variables, >> so that proper garbage collection and separation of >> interpreters becomes possible. I'll look into changing >> some modules after the beta, but I expect that this will >> evolve throughout the entire 3.x series. At a minimum, >> xxmodule will be rewritten to work as a sample of how >> this is supposed to be used. > > So, when all modules are changed to properly support multiple interpreters, > it may be possible to implement a module combining the benefits of > both threading and multiprocessing - moving computation to a separate > interpreter (with its own GIL) without the need to marshal data > across processes?
You'd still need to marshal across threads, as well as the thread switch cost, and much of the extra memory of a process (due to duplicated python modules.) All you really save is the address space switching cost, which I doubt anybody has benchmarked. Multiple interpreters aren't worth the extra complexity. -- Adam Olsen, aka Rhamphoryncus _______________________________________________ Python-3000 mailing list Python-3000@python.org http://mail.python.org/mailman/listinfo/python-3000 Unsubscribe: http://mail.python.org/mailman/options/python-3000/archive%40mail-archive.com
