Bug in merge-base

Bug in merge-base

[Petr Baudis]

  Hello,

  I've hit a strange bug in merge-base I don't want to debug now. ;-)
I've been doing my regular git pull test from my main branch, but
suddenly git merge wanted to do a tree merge instead of fast-forward.
What went wrong?

  The commits tree looks like:

  36c764 -- 808162 -- ...... -- 8e08e7
  5b53d3 -'

(see the rev-tree output at
http://pasky.or.cz/~pasky/dev/git/stuff/rev-tree)

  The problem is, merge-base hints me wrong:

  $ merge-base 8e08e7 808162
5b53d3

instead of 808162.

  Thanks,

-- 
				Petr "Pasky" Baudis
Stuff: http://pasky.or.cz/
C++: an octopus made by nailing extra legs onto a dog. -- Steve Taylor

Re: Bug in merge-base

[Daniel Barkalow]

On Wed, 20 Apr 2005, Petr Baudis wrote:

> Dear diary, on Wed, Apr 20, 2005 at 12:17:12AM CEST, I got a letter
> where Daniel Barkalow <barkalow@iabervon.org> told me that...
> 
> > I can think of. Are you sure there isn't another path to 5b53d3?
> 
> I'm not. Actually there well might be.
> 
> I think merge-base should never take a path which is effectively
> "upside-down" when a straight "upside" one is available.
> 
> Hmm. So what I depended on for merge-base was that when doing it on A
> and B and A is predecessor of B, then it will always just return A.  I
> will perhaps need to abuse rev-tree somehow for this then, it looks.

It is currently optimizing for the shortest longer path, but I guess it
should optimize for the shortest shorter path (i.e., the 0-length path
from A to itself always wins).

On the other hand, the date-based comparison (which you'd get with the
version I sent yesterday with [2/4] and [4/4]) would give you the most
recent common ancestor, which would necessarily avoid this situation.

Do you want to go with the date-based approach, or should I work out a 
shortest shorter path algorithm?

	-Daniel
*This .sig left intentionally blank*