[Python-Dev] Fwd: Removal of GIL through refcounting removal.

Giovanni Bajo rasky at develer.com
Sun Nov 2 14:26:47 CET 2008


On Sun, 02 Nov 2008 10:21:26 +1000, Nick Coghlan wrote:

> Maciej Fijalkowski wrote:
>>> ...
>> 
>>> We know it is the plan for PyPy to work in this way, and also that
>>> Jython and Ironpython works like that (using the host vm's GC), so it
>>> seems to be somehow agreeable with the python semantics (perhaps not
>>> really with __del__ but they are not really nice anyway).
>>>
>>>
>> PyPy has a semi-advanced generational moving gc these days. It's not as
>> well tuned as JVMs one, but it works quite well. Regarding semantic
>> changes, there is a couple which as far as I remember are details which
>> you should not rely on anyway (At least some of the below applies also
>> for Jython/IronPython):
>> 
>> * __del__ methods are not called immediately when object is not in a
>> cycle
>> 
>> * all objects are collected, which means objects in cycles are broken
>> in arbitrary order (gc.garbage is always empty), but normal ordering is
>> preserved.
>> 
>> * id's don't always resemble address of object in memory
>> 
>> * resurrected objects have __del__ called only once in total.
> 
> Yep, I'm pretty those are all even explicitly documented as
> implementation details of CPython (PEP 343's with statement was largely
> added as a cross-implementation way of guaranteeing prompt cleanup of
> resources like file handles without relying on CPython's __del__
> semantics or writing your own try/finally statements everywhere).

Though there is a fair difference from "explicitly documented as 
implementation details" and the real-world code where programmers have 
learnt to save code lines by relying on the reference-counting semantics.

[[ my 0.2: it would be a great loss if we lose reference-counting 
semantic (eg: objects deallocated as soon as they exit the scope). I 
would bargain that for a noticable speed increase of course, but my own 
experience with standard GCs from other languages has been less than 
stellar. ]]
-- 
Giovanni Bajo
Develer S.r.l.
http://www.develer.com



More information about the Python-Dev mailing list