Re: freeze break request: fix pyatspi.getState().getStates()

[Tomeu Vizoso <tomeu vizoso collabora co uk>]

On 04/04/2011 05:58 AM, Frederic Peters wrote:
> Mike Gorse wrote:
>
> > AT-SPI has a method to return the list of states contained in a
> > state set as an array, but this sometimes fails with pyatspi on
> > 64-bit architectures because of what I think is a bug in pygi.  In
> > any case, pygi is copying an argument from an array (an int in this
> > case) and later assuming that it has a valid long value, so it might
> > read eight bytes where only four were initialized.  I've filed the
> > issue as BGO#646581.  I would like to apply the attached patch to
> > pyatspi for the time being; it works around the issue by not making
> > the problematic C call.
>
> Mmmm, I'd certainly prefer to have the pygobject patch applied; adding
> Tomeu for comment.

Hi,

I certainly would like to see that fix in at least PyGObject's stable 
branch.

Mike, could you come up with a test case? I don't have a 64bit machine 
around, so I'm CC'ing the python-hackers mailing list in case someone 
can give a hand. The test would go in the code linked below, you can run 
"make check.valgrind" to make sure the out-of-bounds access is not there 
any more:

http://git.gnome.org/browse/pygobject/tree/tests/test_gi.py#n961

J5 is rewriting that part of PyGObject and it will land soon in master, 
so having a test case will assure that we won't regress there.

It's great to hear that pyatspi is moving to introspection, btw.

Thanks,

Tomeu