[PATCH weston v2] compositor: reset surface's resource field on resource destruction

[Kristian Høgsberg]

On Wed, Sep 11, 2013 at 08:54:13PM +0200, Giulio Camuffo wrote:
> with the surface ref-count feature a surface may live on after its
> resource was destroyed. set it to NULL in that case, so that code
> like find_resource_for_surface() in input.c will act accordingly.
> ---
>  src/compositor.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/src/compositor.c b/src/compositor.c
> index 88df279..a79f911 100644
> --- a/src/compositor.c
> +++ b/src/compositor.c
> @@ -1071,6 +1071,9 @@ destroy_surface(struct wl_resource *resource)
>  	struct weston_surface *surface = wl_resource_get_user_data(resource);
>  
>  	weston_surface_destroy(surface);
> +	/* the surface may live so make sure nothing thinks
> +	 it still has a resourc*e */
> +	surface->resource = NULL;

This avoids the warning, but we might access freed memory here. And if
we did something like checking for ref_count > 1 before calling
weston_surface_destroy() and then setting resource to NULL after if
the ref_count was > 1, we'll run into a problem when we later destroy
the surface for real.  We'll end up in weston_pointer_set_focus(),
called from weston_surface_unmap() and there we'll send out the leave
event with the NULL resource.

So I think the right course of action is to not send out the leave
event if pointer->focus is NULL.  We discussed this back when we had
the leave event warning and I wasn't too happy about doing that, but
in the end I don't see a way around this.

Kristian

>  }
>  
>  static void
> -- 
> 1.8.4
> 
> _______________________________________________
> wayland-devel mailing list
> wayland-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/wayland-devel