Re: deprecating gdk threads

[Pavel Holejsovsky <pavel holejsovsky gmail com>]

On Mon, 06 Aug 2012 13:42:35 -0400, Colin Walters wrote:
> This raises the question "what main context"?  I'd suggest that GTK
> should store the result of g_main_context_get_thread_default() when
> gtk_init() is called.
>
> Also, how does the object slot get set?   Add it to _constructed for
> each object in GTK+?  And then we need to patch the bindings to check
> for it?

The idea was to have some flag in the class definition signaling that the 
objects of the class are 'context-sensitive', and gobject construction 
guts would assign the slot automatically using 
g_main_context_thread_default().  So no code would be needed inside gtk/
gdk/clutter, just one additional flag in class definition.
 
> I'm thinking a cleaner solution is to just patch gtk+ like this:
> 
> 1) Capture the context at gtk_init() time 2) Patch each finalize method
> that calls thread-unsafe functions (e.g.
>    libX11) to use g_main_context_invoke (gtk_get_main_context (),
>                                          gdk_x11_cursor_real_finalize,
>                                          object)

This is nice, much more lightweight, and does not use per-object slot.  
The only drawback is that it must be done manually in the whole gdk/gtk/
clutter.  But it might not actually be that big drawback.

> The thing to note here is that we don't need to bounce every
> g_object_unref(), we just need to ensure the finalizers are run in the
> default main context.

Yep, thats exactly why I think that this should be done inside framework 
and not on the bindings level.

pavel