Peter Xu <pet...@redhat.com> writes:
> On Wed, Mar 02, 2016 at 10:47:39AM +0100, Markus Armbruster wrote:
>> Peter Xu <pet...@redhat.com> writes:
>> > If temporarily not considering kernel_irqchip=split case:
>> >
>> > [{
>> > "version": 2,
>> > "emulated": true,
>> > "kernel": true
>> > },
>> > {
>> > "version": 3,
>> > "emulated": false,
>> > "kernel": true
>> > }]
>> >
>> > To use array rather than dict so that we do not need to change qapi
>> > schema again when GICv4 comes.
>>
>> Drive-by shooting without sufficient context: we may *want* to change
>> the QAPI schema, because that makes the change introspectable with
>> query-schema.
>
> Failed to catch the point. :(
>
> What's "query-schema"? Is that a QMP command?
Yes.
More than you ever wanted to know:
http://events.linuxfoundation.org/sites/events/files/slides/armbru-qemu-introspection.pdf
https://www.youtube.com/watch?v=IEa8Ao8_B9o&list=PLW3ep1uCIRfyLNSu708gWG7uvqlolk0ep&index=28
> What I meant is that, we can define the following (for example):
>
> { 'struct': 'GICCapInfo',
> 'data': [
> 'version': 'int',
> 'emulated': 'bool',
> 'kernel': 'bool'] }
>
> And:
>
> { 'command': 'query-gic-capability',
> 'returns': ['GICCapInfo'] }
>
> So we can keep this schema as it is when new versions arrive. We
> can just push another element in.
To answer questions of the sort "can this QEMU version do X?", it's
often useful to tie X to a schema change that is visible in the result
of query-schema.
Often != always. I'd rather not mess with the schema in unnatural ways
just to make something visible in query-schema.
Note that "can this *board* do X" is a different question.
