The "concrete_class" field of InterfaceClass is only ever written, and as far
as I can tell is not particularly useful when debugging either; remove it.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 include/qom/object.h | 5 ++++-
 qom/object.c         | 1 -
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/include/qom/object.h b/include/qom/object.h
index 95d6e064d9b..f28ffea9a64 100644
--- a/include/qom/object.h
+++ b/include/qom/object.h
@@ -573,12 +573,15 @@ struct InterfaceInfo {
  *
  * The class for all interfaces.  Subclasses of this class should only add
  * virtual methods.
+ *
+ * Note that most of the fields of ObjectClass are unused (all except
+ * "type", in fact).  They are only present in InterfaceClass to allow
+ * @object_class_dynamic_cast to work with both regular classes and interfaces.
  */
 struct InterfaceClass
 {
     ObjectClass parent_class;
     /* private: */
-    ObjectClass *concrete_class;
     Type interface_type;
 };
 
diff --git a/qom/object.c b/qom/object.c
index b4c52d055d9..e9dfad854bd 100644
--- a/qom/object.c
+++ b/qom/object.c
@@ -314,7 +314,6 @@ static void type_initialize_interface(TypeImpl *ti, TypeImpl *interface_type,
     g_free((char *)info.name);
 
     new_iface = (InterfaceClass *)iface_impl->class;
-    new_iface->concrete_class = ti->class;
     new_iface->interface_type = interface_type;
 
     ti->class->interfaces = g_slist_append(ti->class->interfaces, new_iface);
-- 
2.47.1

On 7/1/25 12:13, Paolo Bonzini wrote:
> The "concrete_class" field of InterfaceClass is only ever written, and as far
> as I can tell is not particularly useful when debugging either; remove it.
> 
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> ---
>   include/qom/object.h | 5 ++++-
>   qom/object.c         | 1 -
>   2 files changed, 4 insertions(+), 2 deletions(-)

Patch queued, thanks.

On Tue, 7 Jan 2025 at 11:14, Paolo Bonzini <pbonzini@redhat.com> wrote:
>
> The "concrete_class" field of InterfaceClass is only ever written, and as far
> as I can tell is not particularly useful when debugging either; remove it.
>
> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

I think the reason to have the field would be if we supported
"given a pointer to an interface class, get the concrete
class"; e.g. supposing i have a ResettableClass *rc it would
be nice to be able to do
  DeviceClass *dc = DEVICE_CLASS(rc);
(asserting if your rc isn't actually one provided by a
DeviceClass, obviously). This doesn't currently work because
we don't have the code to actually do this, but apparently
GObject's object model does support this. See this thread from
a few years back:

https://lore.kernel.org/qemu-devel/CAFEAcA8Q3XZrdLJhJ4Uj1CX6J04ON9f91p=NuJXxA+cnEfdz6Q@mail.gmail.com/

But we've lived without this forever (and the thing I wanted
it for back then I ended up not needing), so it seems fine to
remove the field now; easy enough to put it back in future
if somebody does decide to implement cast-to-concrete-class.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>

thanks
-- PMM