Stable releases of Bazaar?

[Martin Pool]

On Mon, Jul 28, 2008 at 4:23 PM, David Ingamells
<david.ingamells at mapscape.eu> wrote:

> I thought 'rock solid' too until:
> 1) I moved our repositories over to bzr+ssh:// because of the horrible
> performance with file:// to an NFS filesystem (problem is with NFS, not
> bzr). Everyone had been saying such good things about the smart server that
> I felt confident to make this move.
> 2) I upgraded a couple of repositories to pack-0.92. Everyone has been
> saying that this format is much better all-round so I felt confident making
> the upgrade.
>
> Suddenly bzr v 1.5 was less that 'rock solid'. (i.e. someone has got sloppy
> :-\ )
>
> I don't know if the problems I've encountered have been solved in v 1.6, but
> I do know that v 1.6beta2 has terrible performance with dirstate-tags.

For 1.6 we're going to ask people to upgrade those branches and give a
warning if they don't, so that people aren't unknowingly suffering bad
performance.  We will keep the ability to read from them but the main
performance work is on current formats.  packs are ten months old now.
 (Which doesn't sound like very long I know.)

> My advice is to make sure that your regression tests cover all combinations
> of usage (not an easy thing to achieve) since it is this above all else that
> keeps developers from getting sloppy. Until you are sure that your tests are
> 100% coverage it will give users confidence if one particular release will
> continue to be supported and kept stable so that they can choose when to
> move forward to a newer release.
>
> If the developer who caused the bug were required to fix it then there would
> be MORE motivation to get it right first time if he knew that he might later
> have to fix the problem in multiple releases.

I agree, but as you say it is easier said than done.  One thing we
need to improve is having more of what we call 'deterministic effort
tests' which can run automatically, on any machine, and reasonably
quickly, and give a red/green indication of whether performance is
good or has regressed.

-- 
Martin <http://launchpad.net/~mbp/>