Checkout old commit and make it a new commit

Question

On Git, say I mess up my commits, and I want to make the version 3 commits ago as the new version. If I do git checkout xxxx, it creates a new branch and it seems like I can only merge it? Could I make this the new "master version"?

I want:

A-B-C-D-E

to become

A-B-C-D-E-F

where F has exactly the same content as C

If I use git revert xxxx instead, it seems like it definitely will have conflicts and I need to manually resolve it.

What I really want is just make the old commit at some point the new commit, regardless of what's in my working directory or the latest commit.

How would I go about doing this?

just `git checkout . ` don't miss the final dot in the command — Ibrahim Tayseer, Mar 12 '20 at 10:20
@IbrahimTayseer Yes, but `git rm -r .` is fairly necessary prior to that command, otherwise if there is any file that is present in the newer version but not in the older version is still kept around. — huggie, Mar 13 '20 at 00:24
Not a duplicate, IMO, since this question wants to keep the previous commits in linear history for future reference. I had the same need, and svick's answer below worked for me, whereas the "duplicate" question was less helpful. — ultraGentle, Nov 02 '21 at 13:46

score 244 · Accepted Answer · answered Aug 01 '10 at 13:46

244

git rm -r .
git checkout HEAD~3 .
git commit

After the commit, files in the new HEAD will be the same as they were in the revision HEAD~3.

answered Aug 01 '10 at 13:46

svick

236,525
50
385
514

This seem to do what I want. Thanks. If I want to refer to the version by SHA-1 hash, should I be doing git checkout HEAD~XXXXX? – huggie Aug 04 '10 at 12:09
7

No, just `git checkout XXXXX`. `rev~n` means *n* revisions before `rev`. – svick Aug 04 '10 at 17:39
105

The dot after HEAD~3 is important. – bbuser Feb 21 '12 at 16:54
@bbuser: is it what, i missed it just now and found another sweet feature. – danjah Apr 03 '12 at 04:06
18

this works but I feel like there should be a more elegant way to do this. – Gabriel Southern Apr 10 '12 at 22:16
2

@Gabriel, what's unelegant about this solution? If you want to completely replace the working directory with content from another commit, this looks to me like a straightforward way to do it. – svick Apr 10 '12 at 22:28
8

with a large repo it takes awhile to delete all of the files and then check them all back out. I feel like git should have a way to revert to an old revision without having to delete everything. But that's just my opinion, this worked well for me and I didn't see a better way when I was looking for alternatives. – Gabriel Southern Apr 10 '12 at 22:39
1

what does 3 represent? 3 from the beginning or most recent? – dezman Sep 18 '13 at 23:08
1

@watson `HEAD~3` means “parent of parent of parent of `HEAD`”. – svick Sep 19 '13 at 09:17
3

Is it possible to use a specific commit ID? – Joseph Oct 11 '13 at 15:09
4

@svick so what would `HEAD~99` represent? – mraaroncruz Oct 15 '13 at 11:30
2

@Joseph Yes, you can use a commit ID. – Brian Kyckelhahn Feb 11 '14 at 23:55
9

Possible to explain in answer what `rm -r .` does? – 2540625 May 06 '16 at 22:41
I noticed that this command does not manage files that are moved by `git mv`. File after the rename still exists. – huggie Aug 01 '16 at 09:03
1

If you use `git checkout XXXXX .` don't forget the `.` at the end. Same thing with `HEAD~3 .` as @bbuser suggests. – adriaan Sep 13 '18 at 13:26
What is the dot after `HEAD~3` represents ? is this the same thing as BASH current directory ? – azzamsa Dec 18 '18 at 23:31
1

@azzamsa Yes, exactly, it's the current directory. – svick Dec 21 '18 at 17:00
Does this loose the history (commits) in between? I think preserving those is desired. – Pete Alvin Dec 23 '19 at 18:29
2

@PeteAlvin No, it won't lose any history. – svick Dec 24 '19 at 21:49
I used this to keep my log in tact so I could go back to my WIP commit in case I needed to. Using a "dead" branch would also work but this is a feature branch and will be squashed into dev so it's of no consequence beyond that – Jacksonkr Jun 01 '20 at 14:40
@svick nice cheat, thanks – Partizanin Jun 02 '20 at 16:45
I had to `git stash` after `git rm -r .` -- otherwise, this successfully set my working state to a previous commit while still preserving all the intervening commits in linear history, in the same branch -- so useful! – ultraGentle Nov 02 '21 at 13:54
I'm hesitant to recommend users run a `git rm -r .`. – Winny Dec 17 '21 at 16:19
does merged commit counted as `n` in `HEAD~n` – Irvan Hilmi Apr 04 '23 at 04:46

score 34 · Answer 2 · answered Aug 01 '10 at 05:05

34

It sounds like you just want to reset to C; that is make the tree:

A-B-C

You can do that with reset:

git reset --hard HEAD~3

_{(Note: You said three commits ago so that's what I wrote; in your example C is only two commits ago, so you might want to use HEAD~2)}

You can also use revert if you want, although as far as I know you need to do the reverts one at a time:

git revert HEAD     # Reverts E
git revert HEAD~2   # Reverts D

That will create a new commit F that's the same contents as D, and G that's the same contents as C. You can rebase to squash those together if you want

answered Aug 01 '10 at 05:05

Michael Mrozek

169,610
28
168
175

4

Hi thanks for the answer. But what if I don't want to do the hard reset since it's already pushed onto a public repository? – huggie Aug 01 '10 at 11:02
2

@huggie Ah. Then you probably want to use the revert method – Michael Mrozek Aug 01 '10 at 18:05
1

then git push -f to force push – radtek Nov 30 '14 at 05:39
1

Does this maintain commit "E" and create a new commit, "F"? – Choylton B. Higginbottom Jun 02 '16 at 16:36
@meetalexjohnson The revert method does; revert makes a new commit that is the opposite of the commit you specified, so they cancel out. The first revert would make F that's the opposite of E, and the second would make G that's the opposite of D, so you end up with the same code you had at C. That's why the second revert is `HEAD~2` instead of `HEAD~1` -- F is the head commit after the first one, so you need to go 2 back to refer to D – Michael Mrozek Jun 02 '16 at 17:42
Can I use the commit key (SHA) instead "HEAD~N"? – Peter Krauss Nov 08 '19 at 13:21
And about `git push`? ERROR "fatal: You are not currently on a branch" – Peter Krauss Nov 08 '19 at 13:31
1

I don't think this is what the OP desires because it looses D-E. – Pete Alvin Dec 23 '19 at 18:31
@PeteAlvin The first version wasn't; the second was. This was discussed in the comments 9 years ago – Michael Mrozek Dec 23 '19 at 20:00
FYI Using `reset` will remove your logs for any comments between HEAD and commit you reset to. – Jacksonkr Jun 01 '20 at 14:34

score 32 · Answer 3 · answered May 30 '13 at 12:49

32

eloone did it file by file with

git checkout <commit-hash> <filename>

but you could checkout all files more easily by doing

git checkout <commit-hash> .

answered May 30 '13 at 12:49

Nick Weisser

910
1
9
21

2

This causes rejection when pushing to origin. – 2540625 May 06 '16 at 22:43
2

This works for me. Pushing to origin wasn't a problem. – Noffls Mar 02 '17 at 08:47
2

so `git checkout ` will detach `HEAD` (push gets rejected), `git checkout .` should checkout `.` (all changes) from the commit to your working-tree, which you can apply as a new commit. You can also detach `HEAD` and branch off that commit. It should then be at `HEAD` for the new branch and you can commit there. The `.` is important. – Brandon G Jul 11 '18 at 17:57

eloone · Answer 4 · 2013-01-16T16:03:03.023

This is exactly what I wanted to do. I was not sure of the previous command git cherry-pick C, it sounds nice but it seems you do this to get changes from another branch but not on same branch, has anyone tried it?

So I did something else which also worked : I got the files I wanted back from the old commit file by file

git checkout <commit-hash> <filename>

ex : git checkout 08a6497b76ad098a5f7eda3e4ec89e8032a4da51 file.css

-> this takes the files as they were from the old commit

Then I did my changes. And I committed again.

git status (to check which files were modified)
git diff (to check the changes you made)
git add .
git commit -m "my message"

I checked my history with git log, and I still have my history along with my new changes made from the old files. And I could push too.

Note that to go back to the state you want you need to put the hash of the commit before the unwanted changes. Also make sure you don't have uncommitted changes before you do that.

score 12 · Answer 5 · answered Jan 14 '13 at 21:36

12

git cherry-pick C

where C is the commit hash for C. This applies the old commit on top of the newest one.

answered Jan 14 '13 at 21:36

Adrian Parsons

171
1
4

20

`git cherry-pick C` takes the *changes* introduced by C and applies them on top of E. This is not what the OP asked for. He wants to have the files in the exact state that they were in C, which `git checkout` provides. – gotgenes May 30 '13 at 13:06

score 2 · Answer 6 · answered Aug 01 '10 at 16:32

The other answers so far create new commits that undo what is in older commits. It is possible to go back and "change history" as it were, but this can be a bit dangerous. You should only do this if the commit you're changing has not been pushed to other repositories.

The command you're looking for is git rebase --interactive

If you want to change HEAD~3, the command you want to issue is git rebase --interactive HEAD~4. This will open a text editor and allow you to specify which commits you want to change.

Practice on a different repository before you try this with something important. The man pages should give you all the rest of the information you need.

Vote down. 1. the question explicitly asks for keeping older commits, so that might be a reason why other answers attempt to do that. 2. as you said, the commits might have been published and there's no need to introduce comments on what that means with rebasing. — tishma, Feb 22 '16 at 10:57

Checkout old commit and make it a new commit

6 Answers6

Linked