RE: [gtk-list] Re: style usage + memory leak

[smolitor insight-tech com (Stephen Molitor)]

Well, I stand corrected.  I'm new to Gtk+, and I'm still trying to get a
handle on the reference counting stuff.

After reading refcounting.txt (in the docs directory of gtk+), here's my new
hypothesis:

When objects are created, their reference count is indeed 1 (not 0), but
they are flagged as
"floating."  When an object is adopted by another widget, it's floating flag
is cleared, but
the reference count is still 1.  So the effect is the same as if the initial
ref count were 0 - ordinarily, you do not need to (and should not) call
unref explicitly for objects you create, because ordinarily your objects
will be adopted (via gtk_container_add(), etc.) by another object.  If you
had to call unref() on every object you created, the reference counting
scheme would be worthless, since the whole purpose of reference counting
schemes is to avoid having to explicitly manage references.

Here's some quotes from refcounting.txt:

<< start quotes >>>>>>>>>>>>>
"When a GtkObject is created, it starts out in a special state called
'floating' (this is the twist).  This means that it is alive and has a
reference to it, but the `owner' of this reference is not known.  There are
certain `potential owners' that will adopt a floating GtkObject.  For
GtkWidgets the most common adopters are the parent widget."

<code excerpt from refcounting.txt>

 option_menu = gtk_option_menu_new ();
  /* option_menu->ref_count == 1 and it is flagged as `floating'.
   */

  gtk_container_add (window, option_menu);
  /* option_menu->ref_count still == 1, but it is no longer `floating'.
   */

<< end quotes >>>>>>>>>>>>>>>>


I'm not sure of any of this, though!!!  I'm trying to understand by writing
about it, and having people correct my posts (selfish, I know).  Also, I'm
not at my Linux box write now, so I can't try any of this out.

Are any of the Gtk+ developers/documentors reading this thread?  This seems
like a very important topic not covered fully by the documentation.



-----Original Message-----
From: Eric Harlow [mailto:linuxgeek@yahoo.com]
Sent: Thursday, October 29, 1998 10:05 AM
To: gtk-list@redhat.com
Subject: [gtk-list] Re: style usage + memory leak



Hmm.. it does appear that all objects have a reference count of one
when created.  I suppose the clean way is to do an unref on each
object you create (ug..) during the program.  Otherwise, this
certainly could cause leaks if you don't keep track of the memory.

To let GTK keep free the objects, could just unref the style after
assigning it to the window.  Don't know what kind of damage that might
do.  <g>

Don't know about the memory hogging though.  Might try replacing
g_malloc with something more informative that keeps track of who's
allocating the memory.  Might show what's going on.

 -Eric

---Robert Roebling <roebling@sun2.ruf.uni-freiburg.de> wrote:
>
> Eric Harlow wrote:
>
> > You should not be calling unref unless you are also making the ref
> > call.  If you just call the unref (as in your example), this should
> > end up freeing the style you just created.
>
> No. When I create the style, its ref count is 1. When I then
> assign it to a widget, I assume the widgets will raise the
> ref count to 2. Without unref after assigning it the widget
> would unref the style upon its own destruction leaving
> the ref count at 1 -> memory leak.
>
> > GTK keeps track of memory pretty well and you should rely on GTK
> > to do that work for you.
>
> Daring thesis. Also GTK+ cannot handle memory on its
> own, you'd still have to know when to unref unless you
> use C++ which can do that itself. wxWindows does it for
> you - as an aside...
>
> And I'm still wondering about the 1100 memory leaks reported
> by dmalloc. It could be a cashing function. I hope it is.
>
>   Robert
>
> --
> ------------------------------------------------------------------
> Robert Roebling              "Write once, compile, run everywhere"
>
> wxWindows/GTK       http://wesley.informatik.uni-freiburg.de/~wxxt
>
> --
> To unsubscribe: mail -s unsubscribe gtk-list-request@redhat.com <
/dev/null
>
>

_________________________________________________________
DO YOU YAHOO!?
Get your free @yahoo.com address at http://mail.yahoo.com

--
To unsubscribe: mail -s unsubscribe gtk-list-request@redhat.com < /dev/null