Re: GTK+ 2.10.7 released

[Owen Taylor <otaylor redhat com>]

[ Removing gnome-announce-list and gnome-hackers from the Cc ]

On Sat, 2007-01-06 at 04:44 -0800, Sergei Steshenko wrote:
> My memory, of course, is not getting better as years go by, but it still
> remembers some peculiar things.
>
> OK, on the official gtk site one can read:
>
> http://www.gtk.org/faq/#AEN703 :
>
> "
> 5.28. Why does GTK+/GLib leak memory?
>
> It doesn't. 
> ...
> ".

The FAQ answer could be a little more expansive, but I think the sense
is pretty obvious.

People *Frequently* (F) mail gtk-app-devel-list and 
Ask the Question (AQ):

 "I wrote this program that creates a window then destroys it
  and the program's memory size doesn't go back to its original
  value"

And the answer to that is:
 
 "GTK+ isn't leaking memory, you are seeing GLib or the C library
  caching allocated memory"

It's a not a statement "there has never been and never will be any bugs
in GTK+ that can cause a memory leak". Use common sense!

> I read this in the times of, I think, gtk+-2.8.18, and it is still there.
>
> So are there or there are no memory leaks in gtk+-2.8.20 ?
>
> Did gtk+2.10.* introduce more or less memory leaks ?

There may still be memory leaks (bugs) we haven't discovered in GTK+-2.8
and GTK+-2.10, so it's pretty hard to say. But for your purposes:

 GTK+-2.8 doesn't leak significant amounts of memory
 GTK+-2.10 doesn't leak significant amounts of memory

Really. 

> If one reads on http://www.gtk.org about absence of memory leaks, should
> he/she consider this to be cheap propaganda at best ?
>
> Or the above claim should be reformulated as:
>
> * for gtk+2.8.20 there are no known to developers occurences of memory leaks;

GTK+-2.8 is no longer maintained, so many memory leaks that have been
fixed in the GTK+-2.10 series will be present in GTK+-2.8.20.

> * gtk+2.10.* is effectively a developmnet release, so expect memory leaks and stay
> away from it for production code.

GTK+-2.10 is not a development release.

> I am asking these particular questions because of the
>
>  387170 Fairly large leak in gtk+
>  360350       leak in gtk_radio_button_focus
>  362439       gtkicontheme::pixbuf_supports_svg leaks GList
>  364514       gtk leaks GDI objects on the win32 classic look and feel
>  364868       GDI resource leak in GtkStatusIcon on win32
>  370395       leak in gtk_rc_parse_icon_source
>  382314       gtkpagesetup leaks when setting new paper size
>  389194       mem leaks in gtkpagesetupunixdialog
>  348108       Refleaks in gtk-demo

If you are interested, you can read all of these bug reports, you can
look at the CVS history and you can determine if they were present in
GTK+-2.8 or not. You can also determine the amount of memory that would
be leaked in a typical program.

But I don't understand the logic that takes you from "Some memory leaks
were fixed between GTK+-2.10.6 and GTK+-2.10.7" to "GTK+-2.10.7 is
likely to be very leaky and I should avoid it" 

> bugs mentioned in the original announcement of gtk+2.10.7.
>
> FWIW, 'konqueror' WEB browser doesn't appear to leak, while gtk+-based
> Mozilla, Firfox, Seamonkey leak a lot, and I'm not sure whether it's
> 'Gecko' or gtk+ issue.

The memory usage of Gecko, whether leaks or just using lots of memory,
has basically zero to do with GTK+.
                                        - Owen