2.0.2 and GConf

[Mark Johnson]

Mark Johnson wrote:

>Derek Atkins wrote:
>
>  
>
>>Mark Johnson <mrj001 at shaw.ca> writes:
>>
>> 
>>
>>    
>>
>>>Mark Johnson wrote:
>>>
>>>   
>>>
>>>      
>>>
>>>>This does not look like the sort of thing that occurs when debugging
>>>>optimized code, but I am going to try a build without optimization
>>>>anyway.
>>>>
>>>>
>>>>
>>>>     
>>>>
>>>>        
>>>>
>>>Obviously, the optimization was interfering with debugging.  Here is the 
>>>gnucash terminal output:
>>>mj at ds9:~$ gnucash --g-fatal-warnings
>>>
>>>GConf-CRITICAL **: file gconf-listeners.c: line 444 (ltable_remove): 
>>>assertion `node != NULL' failed
>>>aborting...
>>>   
>>>
>>>      
>>>
>>This would seem to imply that we're trying to remove a listener
>>a second time, when the listener has already been removed.  At
>>least that's my reading of the message.
>> 
>>
>>    
>>
>I've been working on this in my spare time. (You'd think I'd have more 
>of that since I'm on leave from work, but....)
>
>Setting a breakpoint at gnc-gconf-utils.c:867 indicates that this is not 
>being called twice. I also set a breakpoint at gnc-main-window.c:2692 
>where the notification is added. I noted the id that was stored in the 
>GObject under notify_tag (in gnc_gconf_add_notification). When gnucash 
>tried to release this listener (in gnc_gconf_remove_notification), the 
>id value had changed (I confirmed that the object pointer was the same 
>as before).
>
>This raises two possibilities:
>1. Gnucash is changing it. (I expect this would be a bug.)
>2. "Something else" is trashing it. (Someone else's bug)
>
>  
>
In file src/gnome-utils/gnc-main-window.c, we have the following lines:
2692 gnc_gconf_add_notification(G_OBJECT(window), GCONF_GENERAL,
2693 gnc_main_window_gconf_changed,
2694 GNC_MAIN_WINDOW_NAME);
2695 gnc_gconf_add_notification(G_OBJECT(window), DESKTOP_GNOME_INTERFACE,
2696 gnc_main_window_gconf_changed,
2697 GNC_MAIN_WINDOW_NAME);

Now gnc_gconf_add_notification, stores an id in the object under the key 
notify_tag. notify_tag is made up of two components - the whoami 
argument to the function and a fixed string NOTIFY_TAG 
("GncMainWindow-notify_id" in both cases). This triggered my watch point 
for finding out when the notify-tag was overwritten.

Note in the above two calls to gnc_gconf_add_notification that the 
whoami and object parameters are the same. This causes the second call's 
storing of the id value in notify_tag key of the object to overwrite 
that of the first call. Thus, when the notify removal is later attempted 
in src/gnome-utils/gnc-main-window.c:
1815 gnc_gconf_remove_notification(G_OBJECT(window), 
DESKTOP_GNOME_INTERFACE,
1816 GNC_MAIN_WINDOW_NAME);
1817 gnc_gconf_remove_notification(G_OBJECT(window), GCONF_GENERAL,
1818 GNC_MAIN_WINDOW_NAME);

The second call fails and prints my GConf fatal warning, since it has 
just tried to remove the same notification that the first call removed. 
Naturally, this has lead to one notification being removed and the other 
being lost.

There were no further triggers of my watchpoint after line 2695 above. 
So nothing else changed the id value. So it looks like it was 
possibility #1 after all.

Comments?

Mark