ABI stability of internal DBs

[Jeremy Allison]

On Thu, Jul 08, 2010 at 10:18:21PM +1000, Andrew Bartlett wrote:
> On Wed, 2010-07-07 at 09:10 -0700, Jeremy Allison wrote:
> > On Wed, Jul 07, 2010 at 08:08:55AM -0400, simo wrote:
> > > 
> > > FWIW this goes way beyond what I proposed. I proposed only that 3.5.1 be
> > > compatible , in the cluster case, with 3.5.X.
> > > 
> > > I think it is ok to ask the cluter to be shut down and restarted if you
> > > go 3.6.x
> > 
> > This I think is perfectly reasonable, and is the
> > level of stability we (unofficially) strive for
> > today, and doesn't add any extra burdon to our
> > testing resources.
> > 
> > Thanks Simo - I think this is a very reasonable
> > proposal and I'd like to standardize on this
> > level of "stability" so long as everyone else
> > agrees. Votes ?
> 
> I'm a little concerned that 'silence means violent agreement', because

Silence doesn't mean violent agreement here, that's why I called
for votes. I'm voting +1 :-).

> I'm still a bit worried about the cost, but I guess the restricted
> nature of patches to the release series should mean we don't change
> things too franticly here. 
> 
> In terms of what we promise to our cluster users, should we just promise
> and test an upgrade from the exact previous version (but deliver broader
> compatibility in practice)?
> 
> Finally, how are we going to mark and version these structures, so that
> it's unambiguous as to what structures are part of both the long-term
> and short-term ABIs here?

I'm not sure we need to be so formal on this. The people
who care about this watch the code *very* closely (hi there
metze & vl :-) and anyone changing this usually discusses
it with them before pushing any patches.

Jeremy.