Re: a GList of g_object_add_weak_pointer()s

[David Nečas (Yeti) <yeti physics muni cz>]

On Thu, Mar 01, 2007 at 06:38:48PM -0800, Rick Jones wrote:
> I am in the midst of fitfully trying to convert netperf4 
> (http://www.netperf.org/svn/netperf4/trunk/ to use GObject.  In netperf4 
> there are these things (soon to be objects) called "tests" and a test 
> can be dependent upon another test.  Several tests can depend on a given 
> test.  I figured that a GList (into glib for a penny, in for a pound I 
> figure) of g_object_add_weak_pointer() references would then allow the 
> test on which the other N tests are dependent to send each of them a 
> signal via a "foreach" when it was setup and they could proceede.
>
> The problem is that g_list_append() wants a pointer value, but 
> g_object_add_weak_pointer() wants a place to stuff the pointer - the 
> GList node won't really exist until g_list_append() completes, which 
> leaves me a chicken and egg situation.  I suppose I could just pass 
> g_list_appeand() NULL, but (yes, I should look at the source or write a 
> test program :) would g_list_append() actually like that?
>
> I don't really want to allocate another bit of memory to hold the 
> reference, I'd like to have it in the data field of the GList node.
>
> Suggestions?
>
> I was thinking that I could have a dummy area that I pass to 
> g_list_append(), and then I could pass &(node->data) to 
> g_object_add_weak_pointer().

Exactly, except for the `dummy area' which reveals some
misunderstanding.  The pointer passed to g_list_append() is
just a value.  If you for some reason don't know the value
at the time you are creating the item for it, just assign
the right value to item->data later, pass it to
g_object_add_weak_pointer(), or whatever.

However I don't understand why you can't do just this:

  p = some_object_new();
  item = g_list_append(NULL, p);
  g_object_add_weak_pointer(p, &item->data);
  list = g_list_concat(item, list);
  /* or g_list_concat(l, i), but that's inefficient, it's
   * better to construct the list in reverse order and then
   * do g_list_reverse() */

However, if the data structures do not change (or seldom
change) after construction, it would not use lists, they
bring no advantage in this case and they have bad data
locality, pointer arrays would be better.

Yeti


--
Whatever.