How to rebase a branch off a rebased branch?

Question

So my history looks like this:

o---o---o---o master \ o---o---o A \ o B

So, to explain:

• I have branch A which was started from master
• I have branch B (with only 1 commit) which was started from A

What I want is this:

o---o---o---o master \ o---o---o A \ o B

---

What I did was:

1).

git checkout A git rebase master

This resulted in a lot of conflicts which, after some significant time spent fixing, the following history emerged:

o---o---o---o master \ o---o---o A

Which is exactly what I wanted.

(I don't know where B is right now)

---

2).

After this I did a lot of squashes and changed the order of commits on A, to make the history look like I want.

---

3).

Now, what I also want to do is:

git checkout B git rebase A

However this doesn't seem to work and I don't know why. If I do git log I see the commits which were there before I did step 1.

Also, I get the same huge number of conflicts which I already solved at step 1. I spent significant time doing it, don't want to do it again.

This example suggested to use --onto, which I did:

git checkout B git rebase --onto A

But this deletes the commit on B entirely and makes A and B point to the same commit i.e. the last one on A.

---

My question is: How can I effectively rebase B off A so that it looks like B started from A ? (which was actually true in the beginning).

My best guess is that I'm using --onto wrong. Or that I should use something else (like cherry-pick).

Answer 1

Short answer to How can I effectively rebase B off A so that it looks like B started from A? _{Assuming you want to move exactly one commit:}

git rebase --onto A B~ B

_{If you want to move more then one commit use:}

git rebase --onto A old_A B

The rest of the answer.

Your branch of B is still around (you can check it out), but its parent is still the exact commit object that A was before.
to see a graphical representation of this I use:

git log --graph --decorate --all

to see all branches and where they are with respect to each other.

What you originally had:

o---o---o---o  master
     \
      o---o---o  A
               \
                o B

What you have now:

o---o---o-----------o  master
     \               \
      o---o---o(B~)   o---o---o A
               \
                o B

In terms of using --onto, you need to have a starting point and an ending point. use:

git rebase --onto [target] [rebasing stops] [rebasing head]
git rebase --onto A B~ B

And what you get:

o---o---o----------o  master
     \              \
      o---o---o      o---o---o A
            (old_A)           \
                               o B

[branch_name]~ indicates the parent commit of the branch.

The B~ is the branch that you do not want to change. (It happens to be the old A)

Alternatively, if B was the only commit that had A as a parent, (i.e., B is the end of a chain of commits that branch off master) you could do

git checkout B
git rebase master
git checkout B~   # this is the commit before B (the A commit)
git branch -d A   # remove the old A branch (it was rebased, and so is now invalid
git branch A      # recreate the A branch on the commit that is based on the original A

Answer 2

I have the same issue of my git flow, and I found a best and fast way to do it.

(1) Project history in the beginning:

    master ---A---B---C
                \
                 D---E---F feature1
                          \
                           G---H feature2

(2) Rebase feature1 onto master and force pushed:

    master ---A---B------------------------C
                \                           \
                 D---E---F feature1(old)     D---E---F feature1
                          \
                           G---H feature2

(3) Rebase feature2 onto featrue1 (the new one)

    master ---A---B------------------------C
                                            \
                                             D---E---F feature1
                                                      \
                                                       G---H feature2

The most confused part is how to do (3) .. but no one has a clear answer of doing it.

I believe many people encountered the same issue as I did, when we were trying to do "rebase --onto", we found that feature1(old) is actually not existing!

    git rebase --onto feature1 feature1(old) feature2

Solution is to use below instead:

    git rebase --onto feature1 feature1@{1} feature2

The syntax feature1@{1} means "the last known state of feature1 before the rebase", answer is refereed from https://coderwall.com/p/xzsr9g/rebasing-dependent-branches-with-git

Answer 3

If you have already re-based A. It should be the case that B is exactly where you left it. The branch (a pointer) that was A has simply moved to it's new location.

What I would recommend to effectively rebase B onto A is, as you suggested, to use 'cherry-pick'. This command attempts to apply the changes made in a commit to the branch on which you run it.

So if the commit IDs of the commit to which B originally pointed was '123456' then I would recommend moving your current 'B' to the same place as the new 'A' with git branch -f B A then run git cherry-pick 123456 which will apply the changes onto A.

I believe the --onto flag is used to set the target location from which to apply the commits.It defaults to "upstream" (source: http://git-scm.com/docs/git-rebase).

The way I like to think of the rebase command is as follows:

git rebase --onto <Starting here> <Apply all commits from HERE> <TO HERE>

Using this, it would probably have been simpler to rebase B onto master, then point A to the commit preceding B.

git rebase master B

(as the start point (--onto) is implicitly 'master')

then to use git branch -f A B^ (the ^ means 'the parent of')