Re: [Qemu-devel] [PATCH 3/3] qom: add unit test for Interfaces

[Anthony Liguori <aliguori@us.ibm.com>]

On 06/18/2012 08:51 AM, Andreas Färber wrote:
> Am 18.06.2012 15:46, schrieb Peter Crosthwaite:
>>>>> +#define HERBIVORE(obj) \
>>>>> +    INTERFACE_CHECK(Herbivore, (obj), TYPE_HERBIVORE)
>>>>> +
>>>>> +typedef struct Herbivore
>>>>> +{
>>>>> +    Object obj;
>>>>> +} Herbivore;
>>>>
>>>>
>>>> All this is doing is saying Herbivores are Objects right? A user cant
>>>> add anything to this struct given that interfaces are stateless so
>>>> could this be simplified to
>>>>
>>>> typedef Object Herbivore;
>>>
>>>
>>> This is admittedly a little wierd...
>>>
>>> Interfaces don't exist as Objects in QOM.
>>
>> Not in the sense that they used too, but all objects that implement an
>> interface are still OBJECTs just through the inheritance path of their
>> concrete class, which is the point of this struct yes?

Yes, they are objects, which is why the struct just contains Object.  It's 
basically just a type-safe wrapper for Object.  Nothing more than that.

>>
>>   They are just classes.  But it's
>>> very handy to be able to have a Herbivore type that you can cast objects to.
>>>
>>
>> Yes I agree in full, but the typedef definition is functionally
>> equivalent to what you have there, and removes the temptation to add
>> fields to the object type. Having a skeletal struct there gives the
>> illusion to readers that interface objects are in someway extensible.

typedef struct Object Herbivore;

is a weak alias.  IOW, it's not an error to pass Object * instead of Herbivore 
*.  Worse yet, passing Carnivore * also is not an error.

The struct definition provides a strict alias.  We could wrap it in a macro to 
avoid confusing the reader I guess.

> Why have a typedef at all then? You can just use Object directly. If
> that ever changes it leads to all kinds of problems (well, necessary
> adjustments), as seen in the pci_host series.

Type safety.

Regards,

Anthony Liguori



>
> Andreas
>