ABI stability of internal DBs

[Jeremy Allison]

On Fri, Jul 09, 2010 at 10:52:39AM +0200, Volker Lendecke wrote:

> From my standpoint we can promise to do our best. But giving
> a fixed guarantee that we will never break rolling code
> upgradeability within a 3.x version is a very strong
> statement that can hurt us badly. It will eventually put a
> huge workload on us for when we find a bug that requires
> data fields to change. 

If it's that severe a bug, could we push a new .x release
for it ? :-). Sorry, that's a sneaky way around the problem.

I think we promise, and if we hit exceptional circumstances
we ask our users. I think that's the best we can do.

> But if you are 100% certain that these bugs never happen,
> I'm fine with your guarantee.

As I said. I've only ever seen one that really required
this within a .x release (the tdb signed/unsigned hash
mess). So I think they're pretty rare.

Jeremy.