On Wed, Dec 15, 2010 at 06:12:25PM +0100, Paolo Bonzini wrote: > On 12/15/2010 11:00 AM, Michael S. Tsirkin wrote: > >>Indeed, subsections are for data that is rarely needed so that > >>there's some chance (sometimes ~100%) of migration working > >>seemlessly. > > > >If a subsection arrives that qemu does > >not know about, won't migratin fail? > > Yes, that's why rarely needed => some high chance of migration > working (though no certainty). > > >>In this case it's either > >>no-bump-and-live-with-the-consequences, or changing the version id. > > > >This was discussed to death already. version ids have the problem > >that they don't play nicely with downstreams. > > Downstream version bumps don't play nicely with upstream, so > downstream does have a reason for always-necessary subsections. But > upstream can bump the version id as much as they care. > > Paolo
This assuming upstream developers do not care about downstreams. To give a chance for downstream to cherry-pick changes, upstream should use subsections instead of version ids too. -- MST
