Launchpad by mistake marks branches as merged

[Dmitry Tantsur]

Here:
https://code.launchpad.net/pkibd?field.lifecycle=ALL
I have several branches, parent for all of them is lp:pkibd.
Any time I push to lp:pkibd the following two branches are marked as merged though they were not:
lp:~divius-inside/pkibd/move-viewer-into-library
lp:~divius-inside/pkibd/2.2.0-win32-build
I change their status manually but after next push to lp:pkibd they become "merged" again.

This problem may be connected with the fact that these child branches have no own changes (i.e. they are copies of some revision of trunk).

[Dmitry Tantsur (divius)]

I've made a commit to each child branch and now it works o.k. Maybe we need smarter detection of whether branch is already merged or not. Or maybe a way to forbid automatic status change.

[Martin Pool (mbp)]

2009/8/14 Launchpad Bug Tracker <email address hidden>:
> This problem may be connected with the fact that these child branches
> have no own changes (i.e. they are copies of some revision of trunk).

That's correct - if there are no new revisions Launchpad assumes
they've been merged. This may be a bit oversimplified and not reflect
the user's real intentions.

Possibly they should only be marked merged if they were observed to
have new revisions that are then merged into the trunk. (Though that
might have races if we don't observe all the updates.)

Or possibly there should be some way for the user to say a branch is
still live and important even if there are no new revisions.

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

[Tim Penhey (thumper)]

This is because the algorithm that marks it as merged checks to see if the tip of the branch is in the ancestry of the trunk branch. If the tips are the same then it "knows" that they are actually the same branch. We should change the check to see if the tip is in the ancestry but not the history (the history is the mainline of the branch), as this would be able to determine that the branch was the same as trunk at some stage, and hence not really "merged".

[Aaron Bentley (abentley)]

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

Tim Penhey wrote:
> We should change the check to see if the tip is in the ancestry
> but not the history (the history is the mainline of the branch), as this
> would be able to determine that the branch was the same as trunk at some
> stage, and hence not really "merged".

However, this will break in the case where the branch is "merged" by
pushing it into trunk.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAkqKsmgACgkQ0F+nu1YWqI2GQgCcDZZlNChQ59D5/yTrApohAEiC
TO4An0VZDD7tsH0sSsKhNEE1tITQthcq
=iYEa
-----END PGP SIGNATURE-----

[Tim Penhey (thumper)]

On Wed, 19 Aug 2009 01:53:48 Aaron Bentley wrote:
> Tim Penhey wrote:
> > We should change the check to see if the tip is in the ancestry
> > but not the history (the history is the mainline of the branch), as this
> > would be able to determine that the branch was the same as trunk at some
> > stage, and hence not really "merged".
>
> However, this will break in the case where the branch is "merged" by
> pushing it into trunk.

Yes. I don't think we are going to get it right absolutely every time.

Perhaps it is better to err on the side of not marking something merged when
it is rather than marking something merged that isn't.