Re: Non-atomicity in gtk_list_store_insert(), gtk_list_store_set()

[Christof Petig <christof petig-baender de>]

Jonathan Blandford schrieb:
> Federico Mena Quintero <federico ximian com> writes:
>
>
> > The idea is to call it like this:
> >
> > 	gtk_list_store_insert_and_set (list_store, &iter, N,
> > 				       0, "foo",
> > 				       1, "bar",
> > 				       -1);
> >
> > This would make the insertion+modification atomic; the append_and_set()
> > version would be analogous.  Only the "insert-row" signal would be
> > emitted.

A very good idea (also since iterator dereferences slow gtkmm down 
considerably, there are similar patches 
[http://bugzilla.gnome.org/show_bug.cgi?id=108816] floating around as a 
remedy).

IIRC you missed the parent argument ...

Isn't append_and_set simply passing -1 for position - a separate 
function would be overkill IMHO (though gtk+ tends to go that way)?

A GValue variant would be a good idea, too. Especially C++ people tend 
to put (boxed) objects into ListStores.

Perhaps insert_before/insert_after are far more efficient (I don't know 
the implementation details), so replacing them (too) would make it even 
faster.

> This function is probably fine to add.  A couple comments:
>
>  * I am not really sold on the name.  I think insert_with_values sounds
>    a bit better.

I like that name too.

>  * We also need a non-varargs version for language bindings.

GList?

> > Should we have this, and deprecate the old insert()/append() functions? 
> > set() can remain in place, as it is needed to modify existing rows.
>
>
> I'd rather not deprecate the old functions here.  It doesn't buy us anything.

But a comment should tell the users to prefer the new variant because of 
atomicity and speed.

    Christof