"gnome-control-center display" crashes if no RANDR extension

"No RANDR extension" was the result of clicking "nvidia-settings" (installed automatically by Ubuntu installer) to use external monitor as "separate X screen" with "xinerama", no need to open a terminal to get into this.

StacktraceTop:
 gtk_widget_size_allocate (widget=0x0, allocation=0x7fff56391130) at /build/buildd/gtk+3.0-3.4.0/./gtk/gtkwidget.c:4872
 ?? () from /tmp/tmpLg5uJs/usr/lib/libgnome-control-center.so.1
 g_cclosure_marshal_VOID__BOXEDv (closure=0x7fb2903e1d30, return_value=<optimized out>, instance=0x7fb290845120, args=<optimized out>, marshal_data=<optimized out>, n_params=<optimized out>, param_types=0x7fb2903e1da0) at /build/buildd/glib2.0-2.32.0/./gobject/gmarshal.c:1160
 _g_closure_invoke_va (closure=0x7fb2903e1d30, return_value=0x0, instance=0x7fb290845120, args=0x7fff56391468, n_params=1, param_types=<optimized out>) at /build/buildd/glib2.0-2.32.0/./gobject/gclosure.c:840
 g_signal_emit_valist (instance=0x7fb290845120, signal_id=<optimized out>, detail=0, var_args=<optimized out>) at /build/buildd/glib2.0-2.32.0/./gobject/gsignal.c:3207

The message "RANDR extension is not present" originates from the package "gnome-desktop3"

libgnome-desktop/gnome-rr.c::gnome_rr_screen_initable_init()

if (XRRQueryExtension (dpy, &event_base, &ignore))
{
...
}
else
{
   g_set_error (error, GNOME_RR_ERROR, GNOME_RR_ERROR_NO_RANDR_EXTENSION,
                   _("RANDR extension is not present"));

  return false;
}

libgnome-desktop/gnome-rr.c::gnome_rr_screen_new() says:

 * Returns: a unique #GnomeRRScreen instance, specific to the @screen, or NULL
 * if this could not be created, for instance if the driver does not support
 * Xrandr 1.2. Each #GdkScreen thus has a single instance of #GnomeRRScreen.

therefore in gnome-control-center panels/display/cc-diaply-panel.c::cc_display_panel_constructor() the following code is incorrect:

  self->priv->screen = gnome_rr_screen_new (gdk_screen_get_default (), &error);
  g_signal_connect (self->priv->screen, "changed", G_CALLBACK (on_screen_changed), self);
  if (!self->priv->screen)
    {
      error_message (NULL, _("Could not get screen information"), error->message);
      g_error_free (error);

if the call to gnome_rr_screen_new() returns NULL, as it will if no RANDR is available, the subsequent setup of the signal:

g_signal_connect (self->priv->screen, "changed", G_CALLBACK (on_screen_changed), self);

is done with self->priv->screen == NULL

So when that signal fires it passes a NULL screen pointer to on_screen_changed(). Passing the NULL pointer doesn't matter since it isn't used in the call-back, but self->priv->screen is used:

on_screen_changed (GnomeRRScreen *scr,
                   gpointer data)
{
  GnomeRRConfig *current;
  CcDisplayPanel *self = data;

  current = gnome_rr_config_new_current (self->priv->screen, NULL);
  gnome_rr_config_ensure_primary (current);

So current will be NULL after calling gnome_rr_config_new_current()

I've uploaded the potential fix to my PPA for Precise. Please test and report back.

https://launchpad.net/~tj/+archive/bugfixes

The bug hasn't seen any activity in years and there has been no recent similar reports, the code also changed quite a lot since and it's likely the issue doesn't exist anymore. Closing, feel free to open a new report if you still get problems in recent Ubuntu versions

Makes sense. Besides, I've been using XFCE for a long time now, so have not had any visibility whether this bug still occurs with Gnome.