Bad base selection

[Aaron Bentley]

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

The merge base selection in bzr is usually pretty good, but here's an
example of how it can go wrong:
http://panoramicfeedback.com/opensource/bzr/wrong-base.svgz

The selected merge base is labeled 87734, and is coloured green.  Since
it's 1629 steps away from the graph root, it meets the criteria of
'common ancestor farthest away from graph root'.

But it's pretty clear that R1318 would have been a better choice, as it
is closer to both R1321 and r1155.  It would be nice to handle this
better, but such an algorithm should be

1. reasonably easy to comprehend
2. no more prone to ambiguous results than the current method.
3. not subject to the 'shortcut problem', where old revisions, merged
recently, are treated as good candidates.

On the whole, it may not be worth worrying about 'till after the stable
release, but it's useful to have that case documented.

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

iD8DBQFDV5400F+nu1YWqI0RAkv4AKCAKRrC0hZbBO/JxfhfEZHLnLCaDwCeNjVa
8DRSxrGKOssnVOdqWfvF+Ys=
=xhVU
-----END PGP SIGNATURE-----