[gtkada] Heap corruption and other problems

[Dmitry A. Kazakov]

On Mon, 12 Aug 2013 09:37:45 +0200, you wrote:

>> 1. It seems that Gtk3 up to version 3.8 (Glib 2.36) has serious problem
>> which corrupts Ada's memory pool sooner or later. The bug is sporadic and I
>> cannot track it down, especially without building Glib from sources, which
>> is a problem itself.
> 
> I don't think we have seen that one ourselves. What is the symptom, a Storage_Error
> or something that could help recognize it better ?

Sometimes Storage_Error, even when GNAT debug pool is used and has 0 bytes
allocated in it. Sometimes application crashes silently (access violation).
I cannot track it down.

I discovered the problem first with text buffer. GtkAda implementation of
Insert is this:

   procedure Insert
      (Buffer : not null access Gtk_Text_Buffer_Record;
       Iter   : in out Gtk.Text_Iter.Gtk_Text_Iter;
       Text   : UTF8_String)
   is
      procedure Internal
         (Buffer : System.Address;
          Iter   : in out Gtk.Text_Iter.Gtk_Text_Iter;
          Text   : Interfaces.C.Strings.chars_ptr;
          Len    : Gint);
      pragma Import (C, Internal, "gtk_text_buffer_insert");
      Tmp_Iter : aliased Gtk.Text_Iter.Gtk_Text_Iter := Iter;
      Tmp_Text : Interfaces.C.Strings.chars_ptr := New_String (Text);
   begin
      Internal (Get_Object (Buffer), Tmp_Iter, Tmp_Text, -1);
      Free (Tmp_Text);
      Iter := Tmp_Iter;
   end Insert;

It allocates and frees each time when called and that causes the problem
relatively soon. When I replaced it with: 

   procedure Insert
      (Buffer : not null access Gtk_Text_Buffer_Record'Class;
       Iter   : in out Gtk.Text_Iter.Gtk_Text_Iter;
       Text   : UTF8_String)
   is
      procedure Internal
         (Buffer : System.Address;
          Iter   : in out Gtk.Text_Iter.Gtk_Text_Iter;
          Text   : Interfaces.C.char_array;
          Len    : Gint);
      pragma Import (C, Internal, "gtk_text_buffer_insert");
   begin
      Internal (Get_Object (Buffer), Iter, Interfaces.C.To_C (Text),
Text'Length);
   end Insert;

The issue was gone. Now it keep on crashing but elsewhere, usually in some
random Gtk_New.

> Are you able to run your application in valgrind, maybe there is a bug that can be found
> much earlier than waiting for the memory corruption...

I didn't test it under Linux yet. Maybe it is a Win32 problem.

>> 2. Another serious problem is in GDK which shows itself as:
>> 
>>   Gdk-CRITICAL **: gdk_device_get_source: assertion `GDK_IS_DEVICE
>> (device)' failed
>> 
>> on many occasions. E.g. in combo boxes after resizing them.
> 
> We have seen that one at some point, but it is gone in the development
> versions now.

Create a window with combo box in it:

with Gtk.Box;             use Gtk.Box;
with Gtk.Main;
with Gtk.Window;          use Gtk.Window;
with Gtk.Combo_Box_Text;  use Gtk.Combo_Box_Text;

procedure Simple is
   Window : Gtk_Window;
   Combo  : Gtk_Combo_Box_Text;
   Box    : Gtk_VBox;

begin
   Gtk.Main.Init;
   Gtk_New (Window);

   Gtk_New_VBox (Box);
   Window.Add (Box);
   Window.Set_Title ("Test");
   Gtk_New (Combo);
   Box.Pack_Start (Combo, False, False);
   Combo.Append_Text ("A");
   Combo.Append_Text ("B");
   Combo.Append_Text ("C");

   Window.Show_All;

   Gtk.Main.Main;
end Simple;

Start program. Resize window so that combo box resizes. Select something in
the box. Done.

> I think it is part of our own patches to gtk+, did you apply these ?
> They are in the contrib/ directory (of course, that's in case you recompile your
> own gtk+, I don't remember what system you are working on).

I don't know how to do it under MinGW. The Gtk I am using I took from
mingw32 Fedora packages (Gtk 3.8.2). I simply copied /bin /lib etc and
changed Path environment variable. That resolved some issues, e.g.
GtkColorButton crash.

>> So far Gtk3 looks barely usable for any serious application. Sorry for all
>> efforts AdaCore spent on GtkAda.
> 
> We do use it very successfully for the upcoming version of GPS in fact. The
> transition is indeed somewhat painful, but then things seem to work as expected
> for us. Of course, we may have made progress or fixes since the last GPL release.

I hope you are right. The problem is difficult is reproduce. In my
application I need a lot of clicking (creating and removing widgets) for
the problem to show. Which suggests a race condition somewhere.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de