Hallo,
I was asked to debug memory corruption problem in some sip-based Python
bindings, and what I found leads me to believe that sip is to blame.
I'll try to explain the problem:
Let's say you have a C++ class that looks like this:
struct C { M m; };
And then let's assume there is some Python code that accesses m after
its containing c object is no longer referenced by Python, for example:
C().m
The Python wrapper for m will then point inside the *deleted* c object,
leading to memory corruption.
I have a more complete example at
http://molb.org/~jk/siptest.tar.gz
First of all, I think this should be considered a bug in sip, since it
easily can result in memory corruption, while there is not even a
warning or even a mention of this problem in the documentation. It
might even be better for sip to refuse generating code for accessing
members.
What I would like happen in this case is that when the m wrapper is
created, the reference count for the c wrapper would be increased, and a
reference to c is stored in m, so that c can be dereferenced when m is
deleted.
However, I'm not sure how to achieve this. I looked at the ownership
stuff but I don't think this will help me. So I'm asking for help on
how to fix or work around this problem.
-- Jonathan Kleinehellefort
_______________________________________________
PyQt mailing list PyQt@riverbankcomputing.com
http://www.riverbankcomputing.com/mailman/listinfo/pyqt
