Re: Reentrancy in dispose

[Ben Pfaff]

John Darrington <address@hidden> writes:

> I'm not convinced that some of the dispose routines that we have are
> very reliable at catching reentrancy.
>
> if (thing->obj != NULL)
>   g_object_unref (thing->obj);
> thing->obj = NULL;
>
> The problem is that g_object_unref can cause its argument's dispose 
> routine to run.  Which in turn, could unref a reference it holds on 
> us, and hence run our dispose routine.  Since we haven't yet set thing->obj
> to NULL, it will call g_object_unref on it again.
>
> Have I analysed this correctly?

Yes, I think so.  If that's a possibility (I think that it really
is not in many cases) We either have to do something like this:

    tmp = thing->obj;
    thing->obj = NULL;
    if (tmp)
      g_object_unref (tmp);

or use a "dispose_has_run" variable.  The latter makes me nervous
because it's possible that some (arguably buggy) dispose routine
calls back into us and causes us to allocate something new.