[pypy-svn] r56200 - pypy/extradoc/talk/ep2008
arigo at codespeak.net
arigo at codespeak.net
Tue Jul 1 17:25:01 CEST 2008
Author: arigo
Date: Tue Jul 1 17:24:59 2008
New Revision: 56200
Modified:
pypy/extradoc/talk/ep2008/gc.txt
Log:
Review attempt, but still not completely sure what we try to say in this
talk :-)
Modified: pypy/extradoc/talk/ep2008/gc.txt
==============================================================================
--- pypy/extradoc/talk/ep2008/gc.txt (original)
+++ pypy/extradoc/talk/ep2008/gc.txt Tue Jul 1 17:24:59 2008
@@ -2,6 +2,19 @@
Intro
=====
+::
+
+ From arigo to fijal: should we discuss this a bit? I still don't
+ know exactly what this talk is trying to achieve but I would
+ personally imagine that to give an intro to GC techniques we need
+ pictures with objects as boxes and references as arrows. Also I
+ think we should go for "CPython" slide, a "Jython / IronPython"
+ slide, possibly a "IronPython on Mono" slide to introduce Boehm, and
+ finally several PyPy slides starting from the simple mark-and-sweep
+ and progressing to moving GCs with some more motivation, finally
+ reaching generational moving GCs.
+
+
* Memory management model
* GC is *not* only the part that takes
@@ -54,7 +67,8 @@
==================================
* CPython - O(n^2) where n is number of objects
- allocated
+ allocated (XXX this is a temporary problem more than
+ a deep issue, it might be fixed soon in CPython)
* Any reasonable generational GC should do better
@@ -62,6 +76,7 @@
================
* Full collection - costly (always the case with cpython)
+ (XXX wrong, CPython uses 3 generations in its cycle finder)
* Nursery collection - quick
@@ -89,6 +104,10 @@
* in CPython, they never get collected
* interesting example - a module global with __del__
+ (XXX not sure I follow what you mean here. Python
+ was designed so that most objects don't have a
+ reference to the module they come from, but just
+ a module name, precisely to avoid cycles)
* in PyPy, cycle gets broken in random order
More information about the Pypy-commit
mailing list