[BUG] apply_changeset / generate_changeset use wrong base

[Aaron Bentley]

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

Hi all,

When applying a changeset, apply-changeset uses the changeset base as
its base.  But the changeset base isn't always the right choice.

If a changeset is generated with reference to the target branch, the
latest revision in the target branch is selected as the base.

For a concrete example...

With Robert's integration (as of 1227) and my bzr.24 (as of 1086):
$ cd bzr.integration
$ bzr changeset ../bzr.24 > ~/integration.patch

This produces a changeset whose base is
abentley at panoramicfeedback.com-20050926150316-ed498650ab07a6ce

When this changeset is applied, it transforms my working tree into
robert's, losing the changes I've committed.

However,
$ bzr find-merge-base . ../bzr.24
merge base is revision
robertc at robertcollins.net-20050923085416-c24f9ae97742896f

Either we need to use common-ancestor to select the changeset base, or
we need to determine merge base independently of the changeset base.  I
prefer the second option.

Aaron
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDOVjA0F+nu1YWqI0RAqPaAJ0eMg+x70Rw0zs93E4eZ6Go3+Y3oQCgiduj
rrGBCvmecygit7gHFJWvNPE=
=BdNl
-----END PGP SIGNATURE-----