Re: python classes and signal connects

[Simon Feltman <s feltman gmail com>]

Hi Thomas,

I was able to verify the general callback closure mechanism within
pygobject and used by Gtk.Builder.connect_signals is leaking
references. See:
https://bugzilla.gnome.org/show_bug.cgi?id=685598

Also attached to this ticket is a unittest showing how to verify
reference counts in a unittest context which might be of use.

What is actually leaking in this case is the dictionary holding the
callback, which in turn keeps the callback around.

-Simon

On Thu, Oct 4, 2012 at 9:35 AM, Thomas Bechtold
<thomasbechtold jpberlin de> wrote:
> Hi Simon,
>
> thanks for the detailed answer! See comments bellow.
>
> On Mi, 2012-10-03 at 01:17 -0700, Simon Feltman wrote:
>> 1) Don't rely on __del__ for cleanup, __del__ should generally be
>> avoided as it doesn't work like a destructor in most languages, add an
>> explicit method to cleanup the AddressInfo instance, clear out any
>> held widgets by setting them to None, also issue the dbus disconnect
>> within this and even disconnect signals if possible.
>
> Disconnecting signals produces segfaults. See
> https://bugzilla.gnome.org/show_bug.cgi?id=685387 .
>
> Setting the Widget to None works fine in the simple example. In d-feet,
> I try to use GtkBuilder (with Glade) when possible. So attached is
> another example (button-signal-gtkbuilder.py) where the solution "set
> all Widgets to None" does no longer work. The example loads the widgets
> from an ui xml definition and connects the signals with
> gtk_builder_connect_signals().
> The instances are never collected by python's gc.
>
> Cheers,
>
> Tom