Created attachment 15227 [details] [review]
patch

Created attachment 15228 [details] [review]
prev to be patch.gz, this seample plain text

Thanks for the bug report

That patch leaks the values.  It should look more like:

  if (!str_a)
    cmp = 1;
  else if (!str_b)
    cmp = -1;
  else
    cmp = g_ascii_strcasecmp (str_a, str_b);
 	
  g_value_unset (&val_a);
  g_value_unset (&val_b);

  return cmp;
 }
etc...

This is what the function looks like now. Does that mean this is wrong
too? I mean, it's callung g_value_unset() earlier than what's in the
comment above.

static gint
compare_func (GtkTreeModel *model, GtkTreeIter *a, GtkTreeIter *b,
gpointer data)
{
        GValue val_a = {0};
        GValue val_b = {0};
        gint cmp;
        const gchar *str_a, *str_b;
                                                                     
          
        gtk_tree_model_get_value (model, a, SORT_DATA_COLUMN, &val_a);
        str_a = g_value_get_string (&val_a);
        gtk_tree_model_get_value (model, b, SORT_DATA_COLUMN, &val_b);
        str_b = g_value_get_string (&val_b);
                                                                     
          
        cmp = g_ascii_strcasecmp (str_a, str_b);
                                                                     
          
        g_value_unset (&val_a);
        g_value_unset (&val_b);
                                                                     
          
        if (!str_a)
                return 1;
        else if (!str_b)
                return -1;
        else
                return cmp;
}

Memory wise the new code looks ok.  The main difference was the
handling of NULL.  g_ascii_strcasecmp will just g_warning on NULL if
either arg is NULL.  Which begs the question.  Why are we checking
NULL later ?

The code is also a smidge bogus in that a=NULL b=NULL will claim that
a > b

Can a or b == NULL ?

any news on this ?