Have git-merge-base support multiple IDs

Hi,


I am using git merge-base as sort of a hack to determine where to start 
rebasing.
Suppose this is the commit log (git log --oneline), of course, all 
unpublished, which is why rebase comes in:

  98683793  Fix For faae2553
  3365a01b  Fix For ab80794f
  62943a23  Feature Baz
  ab80794f  Feature Bar
  faae2553  Feature Foo

To determine the rebase point (i.e. first commit in a series),
one can (ab)use git-merge-base:

  p=$(git merge-base ab80794f faae2553)
  git re -i ${p}^

And then reorder ab80794f, faae2553 to squash the fixes into the 
appropriate commits. This practice works well somewhat.
The twist is that merge-base in git 1.6.3.3 happens to ignore any 
further arguments following two IDs. In short:

  git merge-base A B C...

Only yields the merge-base of A and B, and ignores C...
Perhaps this missing feature could be added in a future version?

Jan Engelhardt venit, vidit, dixit 31.07.2009 17:51:

Hi,


I am using git merge-base as sort of a hack to determine where to start 
rebasing.
Suppose this is the commit log (git log --oneline), of course, all 
unpublished, which is why rebase comes in:

  98683793  Fix For faae2553
  3365a01b  Fix For ab80794f
  62943a23  Feature Baz
  ab80794f  Feature Bar
  faae2553  Feature Foo

To determine the rebase point (i.e. first commit in a series),
one can (ab)use git-merge-base:

  p=$(git merge-base ab80794f faae2553)
  git re -i ${p}^

And then reorder ab80794f, faae2553 to squash the fixes into the 
appropriate commits. This practice works well somewhat.
The twist is that merge-base in git 1.6.3.3 happens to ignore any 
further arguments following two IDs. In short:

  git merge-base A B C...

Only yields the merge-base of A and B, and ignores C...

Uhm, are you sure about this?
The first argument is special.
merge-base computes the merge base between two commits:
- the first argument
- a (hypothetical) merge between all other arguments.

It may look a if C was ignored, though.

Michael

On Friday 2009-07-31 18:22, Michael J Gruber wrote:

Jan Engelhardt venit, vidit, dixit 31.07.2009 17:51:

quoted

To determine the rebase point (i.e. first commit in a series),
one can (ab)use git-merge-base:

  p=$(git merge-base ab80794f faae2553)
  git re -i ${p}^

The twist is that merge-base in git 1.6.3.3 happens to ignore any 
further arguments following two IDs. In short:

  git merge-base A B C...

Only yields the merge-base of A and B, and ignores C...

Uhm, are you sure about this?
The first argument is special. merge-base computes the merge base between two commits:
- the first argument
- a (hypothetical) merge between all other arguments.
It may look a if C was ignored, though.

Hm indeed. Is there a better way to find the common ancestor of commits?

Jan Engelhardt venit, vidit, dixit 03.08.2009 15:39:

On Friday 2009-07-31 18:22, Michael J Gruber wrote:

quoted

Jan Engelhardt venit, vidit, dixit 31.07.2009 17:51:

quoted

To determine the rebase point (i.e. first commit in a series),
one can (ab)use git-merge-base:

  p=$(git merge-base ab80794f faae2553)
  git re -i ${p}^

The twist is that merge-base in git 1.6.3.3 happens to ignore any 
further arguments following two IDs. In short:

  git merge-base A B C...

Only yields the merge-base of A and B, and ignores C...

Uhm, are you sure about this?
The first argument is special. merge-base computes the merge base between two commits:
- the first argument
- a (hypothetical) merge between all other arguments.
It may look a if C was ignored, though.

Hm indeed. Is there a better way to find the common ancestor of commits?

I haven't tested thorougly, but at least for the standard example

git show-branch --merge-base A B C

seems to do what you want. Note that for this command, the order of
arguments is irrelevant, whereas for git merge-base it makes a huge
difference. Also, git show-branch documentation is a bit outdated. I
expect to look into this...

Michael