Re: [Fwd: dereferencing void *]

[David Marceau <uticdmarceau2007 yahoo ca>]

On 03/23/2014 09:52 PM, Colomban Wendling wrote:
> Le 24/03/2014 02:11, David Marceau a écrit :
> > On 03/23/2014 08:36 PM, Joël Krähemann wrote:
> > > The following will compile but is it still correct?
> > >
> > >   g_atomic_pointer_set(&(returnable_thread->safe_data),
> > >                        NULL);
> > http://developer.gimp.org/api/2.0/glib/glib-Basic-Types.html#gpointer
> > typedef void* gpointer;
> >
> > http://developer.gimp.org/api/2.0/glib/glib-Atomic-Operations.html
> > void g_atomic_pointer_set (volatile gpointer *atomic, gpointer newval);
My bad:  volatile void* *
Colomban is correct.
> That's not the signature in current GLib, but that's actually correct,
> and means:
>
> void g_atomic_pointer_set (volatile void* *atomic, gpointer newval);
>
> hence, pointer to pointer.
>
> > struct _AgsReturnableThread
> > {
> >   AgsThread thread;
> >   volatile guint flags;
> >   volatile void *safe_data;  // volatile gpointer safe_data
> > };
> >
> > safe_data holds the address pointing to a void value(unassigned type).
> > By putting &(blah->safe_data), you are asking for the address of the
> > address pointing to a void value.  That's not what you want.
>
> Yes it is.  The atomic pointer functions expect a pointer to the pointer
> to atomically set, not the pointer to set itself.
>
> > Just put (blah->safe_data, NULL).
> > The compiler shouldn't complain.
>
> it shouldn't using a real function, but then it would crash at runtime.
>  Anyway, GLib has various compiler-specific implementation of the atomic
> operations, and the GCC one uses macros, which will detect the
> dereference of that pointer to void, which is invalid C.
>
> The macros does basically (a lot simplified):
>
> #define g_atomic_pointer_set(ptr, valure) \
>       ({ \
>               *(ptr) = (value); \
>               __sync(); \
>       })
>
> > If it does, you could cast blah->safe_data with
> > (volatile void*)(blah->safe_data)
> >
> > or
> >
> > (volatile gpointer)(blah->safe_data)
>
> That's not a wise idea in case it'd hide some real issue, nor of any
> use, in C any pointer is a valid void*.  And it wouldn't help here.
>
> Cheers,
> Colomban
> _______________________________________________
> gtk-app-devel-list mailing list
> gtk-app-devel-list gnome org
> https://mail.gnome.org/mailman/listinfo/gtk-app-devel-list