[Glade-users] liststore with GObjects

[Tristan Van Berkom]

I ran your program in gdb.

The bug is in GTK+, specifically gtk_builder_value_from_string_type()
returns false without
setting an error, this causes gtkliststore.c:list_store_text() to
crash a few lines later.

It seems the current implementation of
gtk_builder_value_from_string_type() doesnt
know how to lookup objects by reference (only to build pixbufs), and
that the object
referencing is done in subclasses or implicitly when the type is reffered to by
a property (in gtkbuilder.c properties are handled with special care
and references
are resolved in the parse finish state).

Sigh, I suppose bugs should be filed... maybe it would be better to just
bring this up on gtk-devel-list...


On Thu, Aug 20, 2009 at 12:51 PM, Alexey
Kurochkin<alexey.kurochkin at pathfinderlwd.com> wrote:
> On Thu, 2009-08-20 at 12:04 -0400, Tristan Van Berkom wrote:
>> On Thu, Aug 20, 2009 at 11:54 AM, Alexey
>> Kurochkin<alexey.kurochkin at pathfinderlwd.com> wrote:
>> [...]
>> >> Yes, note that GObjects in the liststore are be passed by
>> >> reference (i.e. the value of an object column is the id/name
>> >> of the referred object in the file... the same object can show
>> >> up in multiple rows and columns...).
>> >
>> > If you mean replacing <col id="0" /> which Glade generates for GObject
>> > column with something like this:
>> >
>> > <col id="0">adjustment1</col>
>> >
>> > then it does not work. It segfaults... :(
>> >
>>
>> Does it ? try manually placing the adjustment definition /before/ the
>> liststore definition.
>>
>> As a liststore can refer to many different types I doubt its a good
>> policy to force all
>> reffered objects to be before the liststore, so, if the above formula
>> successfully
>> works around the segfault, I would call it a bug in GTK+ (that the
>> liststore does
>> not wait for the "finished" signal to resolve referred objects by id).
>
> Nope, the same result. If you care to look, attached is a simple test.
>
> By the way, I have gtk+2.14.5 source here and do not see liststore
> implementing anything called "finished", only custom_tag_start and
> custom_tag_end.
>