[Python-Dev] Removing the GIL (Me, not you!)
tonynelson at georgeanelson.com
Sat Sep 15 00:55:01 CEST 2007
At 3:30 PM -0400 9/14/07, Jean-Paul Calderone wrote:
>On Fri, 14 Sep 2007 14:13:47 -0500, Justin Tulloss <tulloss2 at uiuc.edu> wrote:
>>Your idea can be combined with the maxint/2 initial refcount for
>>> non-disposable objects, which should about eliminate thread-count updates
>>> for them.
>> I don't really like the maxint/2 idea because it requires us to
>>differentiate between globals and everything else. Plus, it's a hack. I'd
>>like a more elegant solution if possible.
>It's not really a solution either. If your program runs for a couple
>minutes and then exits, maybe it won't trigger some catastrophic behavior
>from this hack, but if you have a long running process then you're almost
>certain to be screwed over by this (it wouldn't even have to be *very*
>long running - a month or two could do it on a 32bit platform).
I don't think either of you understand what setting the initial refcount to
maxint/2 for global objects in a thread's refcount vector would do. It has
/no/ effect on refcounting. It only prevents the refcount from becoming
zero for objects that can never be released, but which would always have a
zero thread refcount on thread exit, which would cause a useless and
frequent thread count decrement for the object. As the object can never be
released, its thread count would be initially non-zero, so the thread count
won't be made zero when the thread refcount becomes zero. The thread count
is shared in the object. The thread refcount is per thread, and should not
be shared, even at the physical cache line level, if good performance is
When a new thread is created, part of the thread state would be the
refcount vector. Hopefully it would mostly be just VM magic, but the
initial part of the vector would contain the immortal objects' refcount,
and those would be set to maxint/2. Or 1, for that matter.
TonyN.:' <mailto:tonynelson at georgeanelson.com>
More information about the Python-Dev