Undoing a git rebase

Viewed 1593649

How do I easily undo a git rebase? A lengthy manual method is:

  1. checkout the commit parent to both of the branches
  2. create and checkout a temporary branch
  3. cherry-pick all commits by hand
  4. reset the faulty rebased branch to point to the temporary branch

In my current situation, this works because I can easily spot commits from both branches (one was my stuff, the other was my colleague's stuff). However, my approach strikes me as suboptimal and error-prone (let's say I had just rebased with two of my own branches).

Clarification: I am talking about a rebase during which multiple commits were replayed, not only one.

19 Answers

The easiest way would be to find the head commit of the branch as it was immediately before the rebase started in the reflog...

git reflog

and to reset the current branch to it (with the usual caveats about being absolutely sure before reseting with the --hard option).

Suppose the old commit was HEAD@{2} in the ref log:

git reset --hard HEAD@{2}

In Windows, you may need to quote the reference:

git reset --hard "HEAD@{2}"

You can check the history of the candidate old head by just doing a git log HEAD@{2} (Windows: git log "HEAD@{2}").

If you've not disabled per branch reflogs you should be able to simply do git reflog branchname@{1} as a rebase detaches the branch head before reattaching to the final head. I would double check this, though as I haven't verified this recently.

Per default, all reflogs are activated for non-bare repositories:

[core]
    logAllRefUpdates = true

Actually, rebase saves your starting point to ORIG_HEAD so this is usually as simple as:

git reset --hard ORIG_HEAD

However, the reset, rebase and merge all save your original HEAD pointer into ORIG_HEAD so, if you've done any of those commands since the rebase you're trying to undo then you'll have to use the reflog.

Charles's answer works, but you may want to do this:

git rebase --abort

to clean up after the reset.

Otherwise, you may get the message “Interactive rebase already started”.

Resetting the branch to the dangling commit object of its old tip is of course the best solution, because it restores the previous state without expending any effort. But if you happen to have lost those commits (f.ex. because you garbage-collected your repository in the meantime, or this is a fresh clone), you can always rebase the branch again. The key to this is the --onto switch.

Let’s say you had a topic branch imaginatively called topic, that you branched off master when the tip of master was the 0deadbeef commit. At some point while on the topic branch, you did git rebase master. Now you want to undo this. Here’s how:

git rebase --onto 0deadbeef master topic

This will take all commits on topic that aren’t on master and replay them on top of 0deadbeef.

With --onto, you can rearrange your history into pretty much any shape whatsoever.

Have fun. :-)

I actually put a backup tag on the branch before I do any nontrivial operation (most rebases are trivial, but I'd do that if it looks anywhere complex).

Then, restoring is as easy as git reset --hard BACKUP.

git reset --hard origin/{branchName}

is the correct solution to reset all your local changes done by rebase.

For multiple commits, remember that any commit references all the history leading up to that commit. So in Charles' answer, read "the old commit" as "the newest of the old commits". If you reset to that commit, then all the history leading up to that commit will reappear. This should do what you want.

If you don't want to do a hard reset...

You can checkout the commit from the reflog, and then save it as a new branch:

git reflog

Find the commit just before you started rebasing. You may need to scroll further down to find it (press Enter or PageDown). Take note of the HEAD number and replace 57:

git checkout HEAD@{57}

Review the branch/commits, and if it's correct then create a new branch using this HEAD:

git checkout -b new_branch_name

Another way that doesn't require doing a hard reset is to create a new branch with your desired starting point.

As with the other solutions, you use the reflog to find the correct starting point.

git reflog

(you can also use git log -g here for more detail)

Then you note the reference to the commit SHA (ex: e86a52b851e).

Finally, you use the git branch command.

git branch recover-branch e86a52b851e

Reference: https://git-scm.com/book/en/v2/Git-Internals-Maintenance-and-Data-Recovery#_data_recovery

If you are on a branch you can use:

git reset --hard @{1}

There is not only a reference log for HEAD (obtained by git reflog), there are also reflogs for each branch (obtained by git reflog <branch>). So, if you are on master then git reflog master will list all changes to that branch. You can refer to that changes by master@{1}, master@{2}, etc.

git rebase will usually change HEAD multiple times but the current branch will be updated only once.

@{1} is simply a shortcut for the current branch, so it's equal to master@{1} if you are on master.

git reset --hard ORIG_HEAD will not work if you used git reset during an interactive rebase.

It annoys me to no end that none of these answers is fully automatic, despite the fact that it should be automatable (at least mostly). I created a set of aliases to try to remedy this:

# Useful commands
#################

# Undo the last rebase
undo-rebase = "! f() { : git reset ; PREV_COMMIT=`git x-rev-before-rebase` && git reset --merge \"$PREV_COMMIT\" \"$@\";}; f"

# See what changed since the last rebase
rdiff = "!f() { : git diff ; git diff `git x-rev-before-rebase` "$@";}; f"

# Helpers
########
# Get the revision before the last rebase started
x-rev-before-rebase = !git reflog --skip=1 -1 \"`git x-start-of-rebase`\" --format=\"%gD\"

# Get the revision that started the rebase
x-start-of-rebase = reflog --grep-reflog '^rebase (start)' -1 --format="%gD"

You should be able to tweak this to allow going back an arbitrary number of rebases pretty easily (juggling the args is the trickiest part), which can be useful if you do a number of rebases in quick succession and mess something up along the way.

Caveats

It will get confused if any commit messages begin with "rebase (start)" (please don't do this). You could make the regex more resilient to improve the situation by matching something like this for your regex:

--grep-reflog "^rebase (start): checkout " 

WARNING: not tested (regex may need adjustments)

The reason I haven't done this is because I'm not 100% that a rebase always begins with a checkout. Can anyone confirm this?

[If you're curious about the null (:) commands at the beginning of the function, that's a way of setting up bash completions for the aliases]

What I usually do is git reset #commit_hash

to the last commit where I think rebase had no effect.

then git pull

Now your branch should match exactly like master and rebased commits should not be in it.

Now one can just cherry-pick the commits on this branch.

I tried all suggestions with reset and reflog without any success. Restoring local history of IntelliJ resolved the problem of lost files

Related