Home Original page

[Python-Dev] Py_CLEAR to avoid crashes

Nick Coghlan ncoghlan at gmail.com
Mon Feb 18 13:36:53 CET 2008 
Greg Ewing wrote:
> Martin v. Löwis wrote:
>> when some caller of PyEval_EvalFrameEx still carries
>> a pointer to some object that got deleted, and then still some code can
>> get hold of the then-deleted object.
> 
> I seem to have missed the beginning of this discussion.
> I don't see what the problem is here. If a caller needs
> a pointer to an object, shouldn't it be holding a
> counted reference to it?

The problem is calls to Py_DECREF(self->attr) where some of the code 
invoked by __del__ manages to find a way back around to reference 
self->attr and gets access to a half-deleted object.

Amaury fixed a few of these recently by replacing the Py_DECREF calls 
with Py_CLEAR calls (and added the relevant pathological destructors to 
the test suite), but was wondering if there was a way to be more 
systematic about fixing them. About the only idea I have is to grep the 
source for all calls to Py_DECREF that contain a pointer deference and 
manually check them to see if they should use Py_CLEAR instead.

Cheers,
Nick.

-- 
Nick Coghlan   |   ncoghlan at gmail.com   |   Brisbane, Australia
---------------------------------------------------------------
             http://www.boredomandlaziness.org
More information about the Python-Dev mailing list