Re: [Qemu-devel] [PATCH 07/18] qom: add link properties

[Anthony Liguori <anthony@codemonkey.ws>]

On 12/01/2011 05:21 AM, Avi Kivity wrote:
> On 11/30/2011 11:03 PM, Anthony Liguori wrote:
>> Links represent an ephemeral relationship between devices.  They are meant to
>> replace the qdev concept of busses by allowing more informal relationships
>> between devices.
>
> So, links are equivalent to pointers?

Yup.  Once we have qom inheritance (next stage), we can have a link<PCIDevice> 
property and you'll be able to set it to an E1000State with the appropriate 
casting and error checking taking place.

>> Links are fairly limited in their usefulness without implementing QOM-style
>> subclassing and interfaces.
>>
>>
>> +static void qdev_get_link_property(DeviceState *dev, Visitor *v, void *opaque,
>> +                                   const char *name, Error **errp)
>> +{
>> +    DeviceState **child = opaque;
>> +    gchar *path;
>> +
>> +    if (*child) {
>> +        path = qdev_get_canonical_path(*child);
>> +        visit_type_str(v,&path, name, errp);
>> +        g_free(path);
>> +    } else {
>> +        path = (gchar *)"";
>
> If gchar != char, this is wrong.  Also, you're converting a const
> pointer into a non-const pointer, discarding type safety.

I'll address this later in the thread.

>> +        visit_type_str(v,&path, name, errp);
>> +    }
>
> Shouldn't this be visit_type_link()?

link isn't a primitive type for visitors.   visit_type_link() would just call 
visit_type_str() so I don't think there's a ton of value in introducing the 
extra layer.

The accessors are the only places that should be marshaling links.

Regards,

Anthony Liguori

>