Re: [Qemu-devel] [RFC 00/11]: QMP feature negotiation support

[Luiz Capitulino <lcapitulino@redhat.com>]

On Tue, 26 Jan 2010 15:57:46 +0000
Jamie Lokier <jamie@shareable.org> wrote:

> Daniel P. Berrange wrote:
> > On Tue, Jan 26, 2010 at 12:57:54PM +0000, Jamie Lokier wrote:
> > > Luiz Capitulino wrote:
> > > > capability_enable [ "foo", "bar" ]
> > > > 
> > > >  Now, only one command is not terrible difficult, but we would
> > > > have to accept an array of objects, like:
> > > > 
> > > > [ { "name": "foo", "enabled": true }, { "name": "bar", "enabled": true } ]
> > > 
> > > That looks like XML-itis.
> > > 
> > > Why not { "foo": true, "bar": true }?
> > 
> > It depends on whether we think we're going to need to add more metadata
> > beyond just the enabled/disabled status. If we did want to add a further
> > item against foo & bar, then having the array of hashes makes that 
> > extension easier becaue you add easily add more key/value pairs to
> > each.
> 
> Sure, extensibility is good, and I personally don't care which
> format/function are used.  Just wanted to question the padded
> structure, because sometimes that style is done unintentially.
> 
> Look at the argument leading up here - Luiz says let's use separate,
> non-extensible enable/disable commands taking a list, because if it
> were a single command it'd be important to make it extensible.  Does
> that make sense?  I don't understand that reasoning.

 I didn't consider extensibility in my first format, but we could also
have:

capability_enable [ { "name": "foo" }, { "name": "bar" } ]

> On that topic: In the regular monitor, commands are often extensible
> because they take command-line-style options, and you can always add
> more options.  What about QMP - are QMP commands all future-extensible
> with options in a similar way?

 Yes, command input is done through a json-object as does output.