On Fri, Oct 30, 2020 at 09:41:46PM +0100, Paolo Bonzini wrote:
> Il ven 30 ott 2020, 21:03 Eduardo Habkost <ehabk...@redhat.com> ha scritto:
>
> > > OBJECT_CLASS_PROPERTY_ADD_STR(oc, MachineState, kernel_filename,
> > > "kernel", prop_allow_set_always);
> >
> > I like the idea of having an interface like this, but I would
> > like to avoid having to write even more boilerplate for each
> > property type to make this work.
> >
> > What would you think of:
> > OBJECT_CLASS_PROPERTY_ADD(oc,
> > PROP_STRING("kernel", MachineState, kernel_filename),
> > prop_allow_set_always);
> >
> > Then we could make the same PROP_STRING macro usable both as
> > object_class_property_add_static() argument and as initializer
> > for existing static Property arrays.
> >
>
> The name should be an argument to OBJECT_CLASS_PROPERTY_ADD though (which
> could be a function and not macro; perhaps
> object_class_property_add_field?). PROP_STRING would be
> DEFINE_PROP_STRING(NULL, etc.) and would not be entirely reusable in
> Property arrays.
>
> But even with that snag I agree with your less-boilerplate argument against
> my proposal.
>
> Since most if not all device properties would have to specify the same
> allow-set function, we would end up defining more macros
> DEVICE_CLASS_PROPERTY_ADD_STR, and so on. If the Property isbpassed a
> struct, instead, we can define just one wrapper
> device_class_property_add_field.
>
> So what about:
>
> 1) add new constructors without the DEFINE prefix and without the name
> argument
>
> 2) add object_class_property_add_field
>
> And later:
>
> 3) add device_class_property_add_field and remove dc->props
>
> 4) remove the name field from Property.
Sounds good, and I like the "field property" name. It is more
descriptive than "static property".
--
Eduardo
