New Bazaar Branches on Launchpad don't work

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Martin Albisetti wrote:
> On Mon, May 18, 2009 at 9:20 AM, Jelmer Vernooij <jelmer at vernstok.nl> wrote:
>> I think the two are unrelated; lp:gant is a vcs-import and Russel is
>> working with a bzr-svn-created branch. Bazaar (or launchpad?) should not be trying to stack in this situation.
> 
> Well, the way it currently works is that all branches for a project
> stack on it's development focus (trunk), no matter if they have common
> parents or not.
> I agree this isn't great, and there are a bug or two filed about it.
> 
> 
> Maybe jml can give us some insight in this  :)
> 

I would mention that stacking on a branch with unrelated history still
should stack, it just has to copy all of the history. Which makes it
*inefficient* but shouldn't make it *fail*.

My guess is that the "auto-upgrade branch/repository format" code uses a
check to see whether the format is rich-root or not, and that is failing
with a Remote branch.

The other possibility is that the Branch + Repository combination is
unexpected, and when 'auto-upgrading' one of them it is actually
downgrading. I ran into this when pushing from 1.9-rich-root repository
with a 0.92 format branch. It saw the branch was too old, and then
auto-"upgraded" both to 1.6 format. That only really broke when using
- --dev6-rich-root because that causes stacking to cross serialization
formats. (Same problem as with crossing-rich-roots, it goes boom.)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkoRdqAACgkQJdeBCYSNAAOhFQCfZioc8cuwWEE530zOVOho4n68
09YAnj/oatAWxqV4Jt3GbkopC1fnmY84
=SzTj
-----END PGP SIGNATURE-----