[Patches] [ python-Patches-843455 ] Make weakref callbacks play nice in gc

[SourceForge.net]

Patches item #843455, was opened at 2003-11-16 22:18
Message generated for change (Tracker Item Submitted) made by Item Submitter
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=305470&aid=843455&group_id=5470

Category: Core (C code)
Group: Python 2.4
Status: Open
Resolution: None
Priority: 9
Submitted By: Tim Peters (tim_one)
Assigned to: Neil Schemenauer (nascheme)
Summary: Make weakref callbacks play nice in gc

Initial Comment:
See bug 839548.  Weakref callbacks attached to objects 
in dead cycles can see objects that have gone thru 
tp_clear() now, and can resurrect such objects too.  A 
wide variety of bad things can result, up to and including 
segfaults.

This implements a scheme I detailed on Python-Dev to 
teach gc about the bad things callbacks can do, still 
calling them, but ensuring that a callback can't reach 
any objects on which tp_clear has been invoked.  If a 
callback happens to resurrect an object, tp_clear won't 
get invoked on the latter object at all by gc.  This keeps 
Python-visible objects wholly sane, and even 
unsurprising.

If this patch is accepted, I'll also check in a plain-text 
file with a version of the Python-Dev msg explaining the 
scheme.

Neal, I most want your eyeballs on this, since you know 
the most about gc's design.  Please assign to Fred when 
you're done (whether or not you can make time), 
because I want his eyeballs on the weakref changes.

Two changes were made to the weakref 
implementation:  (1) a new private API function 
_PyWeakref_HasCallback(), so gc can determine which 
objects have associated weakref callbacks; and, (2) 
because objects in cyclic trash don't have refcount 0, I 
had to remove the refcnt==0 check from 
PyObject_ClearWeakRefs().  Maybe it would be better to 
introduce a workalike private function that skipped that 
one check, and leave the public function alone?

When reviewing this, note that it has to be backported 
to the 2.3 line too:  2.3.2 is the release in which Jim 
Fulton first saw segfaults in real life.


----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=305470&aid=843455&group_id=5470