On 2011-09-06 17:51, Anthony Liguori wrote:
I'm still contemplating how we go about doing this. This series
introduces a couple new concepts like QMP class hinting anonymous IDs.
I'm concerned that we'll further complicate the need to support
backwards compatibility.
Anonymous IDs must not be considered stable. If you prefer, we could try
to filter them out for QMP use. Need to check though if there is a good
location to catch them, but the rule is trivial.
I will also happily drop anonymous IDs again once we have converted all
devices to stable QOM full path addressing. But that will simply take
too much time to wait for it I'm afraid.
Would the command be useful if you couldn't address devices? If it just
dumped the full machine state all at once? That would at least obviate
the need to add anonymous IDs.
Theoretically usable, but extremely unhandy. I would not like so see
such an interface exposed to users.