Re: Subclassing question

[Torsten Schoenfeld <kaffeetisch gmx de>]

Martin Schlemmer wrote:
> I have a class that overrides Gtk2::TextBuffer's insert_text signal to disallow
> adding newlines to the buffer.  The problem however seems to be that when
> the signal is overridden, the iterator is no longer updated causing it to be
> invalid - I assume its because its no longer pass as reference.

This appears to be a bug in the bindings.  The problem is that 
Glib::Object::signal_chain_from_overridden ends up passing a copy of the iter to 
the original C signal handler.  This copy is then correctly revalidated, but 
that doesn't affect the original iter, of course.

Technically, this is due to gperl_value_from_sv using g_value_set_boxed instead 
of g_value_set_static_boxed.  The former copies the boxed object.  I can't think 
of a reason not to use g_value_set_static_boxed there (other than "being 
defensive").  When I do this change, Glib's and Gtk2's test suites still pass.

muppet, do you remember why gperl_value_from_sv uses g_value_set_boxed instead 
of g_value_set_static_boxed?

> Is there any way to do this with the current Glib::Object::Subclass interface
> or perl?

Other than overriding every insert() method of Gtk2::TextBuffer with your own 
variant (with normal method overloading), the only workaround I can think of is 
not to rely on the iter-revalidating behavior of insert() in user code.