Re: [Qemu-devel] [RFC v4 00/13] qmp: query-device-slots command

[Eduardo Habkost <ehabkost@redhat.com>]

On Tue, Aug 15, 2017 at 01:57:50PM -0500, Eric Blake wrote:
> On 08/14/2017 04:57 PM, Eduardo Habkost wrote:
> > Changelog
> > ---------
> > 
> > Changes v3 -> v4:
> > * New compact representation of slot sets.
> > * New generic code to automatically merge similar slots
> >   into a single entry in the command output while keeping
> >   implementations of the method simpler.
> > * Example implementation of IDE and USB bus enumeration
> 
> > 
> > Slot sets are represented by a list of option names and sets of
> > possible values for each of those options.  The command uses a
> > compact representation for the set of valid values for an option.
> > For example, the following set of 5 PCI functions:
> > 
> >       bus: pcie.0
> >       device-number: 31
> >       functions: 1,4,5,6,7
> > 
> > would be represented in the JSON data as:
> > 
> >   {"available":false,"count":5,
> >    "device-types":["pci-device"],"hotpluggable":false,
> >    "opts":[
> >       {"option":"function","values":[1,[4,7]]},
> 
> A list (and not just a single-type list, but a list that mixes scalar
> and sublist),
> 
> >       {"option":"device-number","values":31},
> 
> vs. a scalar.  Why not a one-element array?

It was just to keep the representation as compact as possible, in
the common case of single-value sets.  Probably we can drop that
feature as it saves only 2 bytes in the JSON representation.

> 
> >       {"option":"bus","values":"pcie.0"}],
> >    "opts-complete":true}
> > 
> > I planned to use QAPI alternates to model/document that in the
> > schema, but it would require implementing a few missing features
> > in QAPI alternate support.
> 
> Yeah, I can see how existing QAPI alternates do not yet support arrays,
> which becomes important to your representation.  Do you need help
> getting the QAPI generator improved to support a particular feature that
> you found to be lacking?

I think the lack of support for lists on alternates was the main
obstacle.

Probably we would also need to remove the restriction against
alternates with ambiguous string representations, to allow a
list/number/string/bool alternate to be defined.

Being able to set constraints on the number of elements of a list
would be nice to have, but not required.

-- 
Eduardo