How to amend older Git commit?

Question

I have made 3 git commits, but have not been pushed. How can I amend the older one (ddc6859af44) and (47175e84c) which is not the most recent one?

$git log
commit f4074f289b8a49250b15a4f25ca4b46017454781
Date:   Tue Jan 10 10:57:27 2012 -0800

commit ddc6859af448b8fd2e86dd0437c47b6014380a7f
Date:   Mon Jan 9 16:29:30 2012 -0800

commit 47175e84c2cb7e47520f7dde824718eae3624550
Date:   Mon Jan 9 13:13:22 2012 -0800

Answer 1

You could can use git rebase to rewrite the commit history. This can be potentially destructive to your changes, so use with care.

First commit your "amend" change as a normal commit. Then do an interactive rebase starting on the parent of your oldest commit

git rebase -i 47175e84c2cb7e47520f7dde824718eae3624550^

This will fire up your editor with all commits. Reorder them so your "amend" commit comes below the one you want to amend. Then replace the first word on the line with the "amend" commit with s which will combine (s quash) it with the commit before. Save and exit your editor and follow the instructions.

Answer 2

You can use git rebase --interactive, using the edit command on the commit you want to amend.

Answer 3

git rebase -i HEAD^^^

Now mark the ones you want to amend with edit or e (replace pick). Now save and exit.

Now make your changes, then

git add .
git rebase --continue

If you want to add an extra delete remove the options from the commit command. If you want to adjust the message, omit just the --no-edit option.

Answer 4

In case the OP wants to squash the 2 commits specified into 1, here is an alternate way to do it without rebasing

git checkout HEAD^               # go to the first commit you want squashed
git reset --soft HEAD^           # go to the second one but keep the tree and index the same
git commit --amend -C HEAD@{1}   # use the message from first commit (omit this to change)
git checkout HEAD@{3} -- .       # get the tree from the commit you did not want to touch
git add -A                       # add everything
git commit -C HEAD@{3}           # commit again using the message from that commit

The @{N) syntax is handy to know as it will allow you to reference the history of where your references were. In this case it's HEAD which represents your current commit.

Answer 5

I prepared my commit that I wanted to amend with an older one and was surprised to see that rebase -i complained that I have uncommitted changes. But I didn't want to make my changes again specifying edit option of the older commit. So the solution was pretty easy and straightforward:

1. prepare your update to older commit, add it and commit
2. git rebase -i <commit you want to amend>^ - notice the ^ so you see the said commit in the text editor

3. you will get sometihng like this:

   pick 8c83e24 use substitution instead of separate subsystems file to avoid jgroups.xml and jgroups-e2.xml going out of sync
   pick 799ce28 generate ec2 configuration out of subsystems-ha.xml and subsystems-full-ha.xml to avoid discrepancies
   pick e23d23a fix indentation of jgroups.xml
   

4. now to combine e23d23a with 8c83e24 you can change line order and use squash like this:

   pick 8c83e24 use substitution instead of separate subsystems file to avoid jgroups.xml and jgroups-e2.xml going out of sync    
   squash e23d23a fix indentation of jgroups.xml
   pick 799ce28 generate ec2 configuration out of subsystems-ha.xml and subsystems-full-ha.xml to avoid discrepancies
   

5. write and exit the file, you will be present with an editor to merge the commit messages. Do so and save/exit the text document

6. You are done, your commits are amended

credit goes to: http://git-scm.com/book/en/Git-Tools-Rewriting-History There's also other useful demonstrated git magic.

Answer 6

I've used another way for a few times. In fact, it is a manual git rebase -i. It is useful when you want to rearrange several commits including squashing or splitting some of them. The main advantage is that you don't have to decide about every commit's destiny at a single moment. You'll also have all Git features available during the process unlike during a rebase. For example, you can display the log of both original and rewritten history at any time, or even do another rebase!

I'll refer to the commits in the following way, so it's readable easily:

C # good commit after a bad one
B # bad commit
A # good commit before a bad one

Your history in the beginning looks like this:

x - A - B - C
|           |
|           master
|
origin/master

We'll recreate it this way:

x - A - B*- C'
|           |
|           master
|
origin/master

Procedure

git checkout B       # get working-tree to the state of the commit B
git reset --soft A   # tell Git that we are working before the commit B
git checkout -b rewrite-history  # create a new branch for alternative history

Improve your old commit using git add (git add -i, git stash etc.) now. You can even split your old commit into two or more.

git commit           # recreate commit B (result = B*)
git cherry-pick C    # copy C to our new branch (result = C')

Intermediate result:

x - A - B - C 
|    \      |
|     \     master
|      \
|       B*- C'
|           |
|           rewrite-history
|
origin/master

Let's finish:

git checkout master
git reset --hard rewrite-history  # make this branch master

Or using just one command:

git branch -f master  # make this place the new tip of the master branch

That's it, you can push your progress now.

The last task is to delete the temporary branch:

git branch -d rewrite-history