Il 22/09/2014 18:06, Andreas Färber ha scritto:
> Am 22.09.2014 um 17:57 schrieb Paolo Bonzini:
>> Il 22/09/2014 15:15, Laszlo Ersek ha scritto:
>>> $ git grep -E '\<qemu_register(_pc)?_machine\>' | wc -l
>>> 115
>>>
>>> Even if we just count the clusters, they're way too many:
>>>
>>> $ git grep -E '\<machine_init\>' | wc -l
>>> 66
>>
>> There are just a couple of multi-machine clusters (well, three: pseries,
>> pc, q35). So the default clusters can just default to the machine type.
>
> I would've gone even simpler and special-cased pc/q35 and "none" (like
> we did for -cpu "host") in the comparison function, sparing us any big
> interface work.
>
> Therefore from my point of review, this was merely about how we
> implement the comparison function. No reason to get a heart attack IMO.
Yeah, what I'm proposing is something like
MachineClass *a = pa, *b = pb;
cl1 = a->cluster ? a->cluster : object_class_get_name(a);
cl2 = b->cluster ? b->cluster : object_class_get_name(b);
res = strcmp(cl1, cl2);
if (res)
return res;
return strcmp(object_class_get_name(a),
object_class_get_name(b));
Plus a
mc->cluster = qm->cluster;
in vl.c's machine_class_init.
Paolo
