Call for testing of colocated branch support in bzr.dev

[John Arbash Meinel]

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

...

> mbp at joy% bzr push lp:~mbp/launchpad/test-push bzr: ERROR: Server 
> sent an unexpected error: ('error', 'KeyError', "'Bazaar meta 
> directory, format 1 (with colocated branches)\\n'") HPSS calls: 2 
> (0 vfs) SmartSSHClientMedium(bzr+ssh://mbp@bazaar.launchpad.net/)
> 
> (which is not so surprising, until support is landed there.)
> 
> But, if I try to push into a newly created branch, it seems to get
>  stacking wrong and to transfer an overly large amount of data. 
> Digging in to this a bit more, I think it is because Launchpad (or
>  Bazaar?) doesn't autoconfigure stacking properly if you create a 
> branch, then separately push to it, which is probably a bug 
> regardless of colo stuff.
> 
> It will be nice when we can use this for repos on lp and get away 
> from stacking.
> 
> I'm pretty sure this is a bug in init, perhaps an old latent bug: 
> <https://bugs.launchpad.net/bzr/+bug/936772>
> 


I'm pretty sure it is a known bug that an interrupted push creates a
.bzr directory, but no branch, such that a second push creates a full
branch without stacking.

IMO, the source of the problem is that 'push' creates a repository,
but not a branch until the push finishes.

At the time, I felt creating a Branch on disk (at NULL_REVISION,
probably), and then fetching into it was the way to go. (I believe
this dates all the way back to Robert and I both implementing 'bzr push'.)

I think the original problem was Robert didn't want to create a branch
at 'tip' revision, when the repo didn't have the data yet. Which is
fair enough, though I think creating the Branch at NULL_REVISION, and
then updating it to 'tip' when the fetch finishes is probably a good
way to go.

It would really simplify the fetch code, so that we wouldn't have
separate fetch logic for first creation vs updating.

I'm pretty sure it is already an open bug:
https://bugs.launchpad.net/bzr/+bug/818138

Note, some of the simplification stuff would affect a few bugs like:
https://bugs.launchpad.net/bzr/+bug/775523
https://bugs.launchpad.net/bzr/+bug/771255
maybe https://bugs.launchpad.net/bzr/+bug/397183


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

iEYEARECAAYFAk9CJu4ACgkQJdeBCYSNAAN3uACdGMquELdJOQsJyASWsxz97rY/
3JsAoL0ajqIUbHgWmcV2836lgdJvYBI1
=fGLM
-----END PGP SIGNATURE-----