Re: destroying an unparented widget

[Allin Cottrell <cottrell wfu edu>]

ping. Since the message below has not drawn an answer to date,
perhaps I should add: I'm well aware that reports of memory leaks
in the GTK stack are often spurious, and I'm using
G_SLICE="always-malloc", and mostly my GTK widget usage shows no
leaks via valgrind, but I'm still puzzled by the case of a widget
which is allocated "off-screen" and then nominally destroyed, but
which then shows up in valgrind as leaking.

On Thu, 5 May 2011, Allin Cottrell wrote:

> There's one thing I've never quite understood: the correct
> mechanism for destroying a gtk widget that is never parented or
> shown.
>
> The context is that I'm creating a temporary widget with
> gtk_label_new(), just for the purpose of getting hold of its
> PangoContext so I can figure something out about the font in use
> (whether it supports the UTF-8 minus sign, 0x2212). Once I've got
> that info I'm using gtk_widget_destroy on the label. And valgrind
> is showing me a leak, which I take to mean that I haven't managed
> to destroy the widget. If I try using g_object_unref() instead I
> get an error message from GTK saying I can't do that on an object
> which has only a floating reference.
>
> Would the correct sequence be
>
> label = gtk_label_new("");
> g_object_ref_sink(G_OBJECT(label));
> /* do pango stuff */
> g_object_unref(G_OBJECT(label));
>
> ? And if so, is there an equivalent in GLib < 2.10? Thanks.

-- 
Allin Cottrell
Department of Economics
Wake Forest University