[Qemu-devel] Re: [PATCH 01/25] Introduce QEMU dictionary data type

[Avi Kivity <avi@redhat.com>]

On 07/29/2009 05:22 PM, Luiz Capitulino wrote:
>
>    
>> I meant QObject as a base type, so it is a lower layer than QDict; QDict
>> implements the QObject methods, as do QString, QNumber, etc.
>>      
>
>   Ok, I'm failing at imagining "QDict implements the QObject methods"
> in C, can you elaborate more and/or sketch something?
>    

It's really no different than in other languages, except that you have 
to write a ton of boilerplate.  The kernel is riddled with object 
hierarchies:

typedef struct QType {
     const char *name;
     void (*destroy)(QObject *);
     QObject *(*clone)(QObject *)
} QType;

typedef struct QObject {
    QType *type;
} QObject;

typedef struct QDict {
     QObject base;
     // hash stuff
} QDict;

static QType qdict_type = {
   .destroy = qdict_destroy,
   .clone = qdict_clone,
};

QObject *qobject_clone(QObject *obj) { return obj->type->clone(obj); }

QObject *qdict_clone(QObject *obj)
{
     ...
}

>> The problem with void *, beyond requiring the user to know what the
>> object type is, is that it is impossible to control object lifecycle.
>> When you destroy a QDict containing void *, you cannot destroy the
>> contained objects.  On the other hand if QDict values are all QObjects,
>> then qdict_destroy() can call qobject_destroy() on all of them
>> (qobject_destroy might end up calling qdict_destroy() is a value
>> happened to be a QDict).
>>      
>
>   Right, but as you said in other email, can we get this merged first
> and then improve or should I do the change for this series?
>    

I think we should merge first, doing this sort of change on a 25-patch 
series is no fun.

-- 
error compiling committee.c: too many arguments to function