list-sink at trainedmonkeystudios.org
Tue Dec 1 08:13:23 CET 2009
On Sat, 2009-11-21 at 09:35 -0500, Gary Robinson wrote:
> OK, thanks for the confirmation that it's not working yet, and letting me know what it would take!
> I wish we had the resources to take on the task of converting it to
> RPython, but there's no way at this point. :) I'm using pyprocessing
> with unladen swallow now and it works fine but for the kinds of things
> we're doing (mathy floating point stuff) I'd expect more speed from
The C component of multiprocessing, at least in python 2.6 is tiny and
trivial: a read and write implementation for the underlying connection
class. It would probably be better off converted to plain python for
pypy in any case.
On Sat, 2009-11-21 at 15:06 +0100, Maciej Fijalkowski wrote:
> Oh, and by the way, personally I would prefer to kill the GIL, so we
> don't have to worry any more (but also noone from core devs is working
> on it).
What, roughly, would be required to remove pypy's gil? I have heard
rather nebulous tales of woe and tribulation on the cpython lists
referring to fine-grained per-object locking, re-entrant garbage
collectors, abi stability, and other horrors. I have no idea what bits
of this dubious information would be relevant to pypy's rather more
abstract internals, however.
More information about the Pypy-dev