I have a git repository which holds a Drupal site. I spent the last day trying to build a feature using several different modules. I have given up on my current approach and have decided to try a different combination of modules. However, my repository has several commits on the master branch that contain this feature development process (I understand that I did not branch in an effective manner.) I want to get rid of the last three or four commits and set master to that point in my history (I don't want to merge my current work with anything, I just want it to go away.) How do I do this?
-
Have you already pushed your changes to the `remote`? – merlin2011 Apr 22 '14 at 19:00
-
I am not currently using a remote. I will probably set that up in a few days – Hoytman Apr 22 '14 at 19:42
-
1possible duplicate of [Revert to previous Git commit](http://stackoverflow.com/questions/4114095/revert-to-previous-git-commit) – UpAndAdam Apr 22 '14 at 19:46
-
Looks like the same answer. – Hoytman Apr 22 '14 at 19:54
4 Answers
In order to do it locally, you can do the following commands to go to master and move it to the old commit.
git checkout master
git reset --hard <old_commit_id>
If you then want to push it to the remote, you need to use the -f
option.
git push -f origin master

- 71,677
- 44
- 195
- 329
-
Could I ask for a good place to find a list of options that git uses and their meaning? I don't know what --hard or -f does. – Hoytman Apr 22 '14 at 19:28
-
Actually, I just found this documentation link : https://www.kernel.org/pub/software/scm/git/docs/git.html – Hoytman Apr 22 '14 at 19:40
-
@Mr JF, That is true. One must unprotect to rewrite history. There's no other way. – merlin2011 Mar 16 '18 at 03:29
-
Before pointing main
to a previous commit, I recommend backing up your current main
:
$ git checkout -b main_backup
Then you can safely point main
some number of commits back, e.g. 3:
$ git reset --hard main~3
After moving main
, see your tree of commits:
$ git log --graph --abbrev-commit --decorate --format=format:'%C(bold blue)%h%C(reset) - %C(bold green)(%ar)%C(reset) %C(white)%s%C(reset) %C(dim white)- %an%C(reset)%C(bold yellow)%d%C(reset)' --all
Now, I would recommend keeping the backup until you're absolutely sure it's unnecessary (for instance, once your new approach is fully implemented), but when you're sure, you can clean up by deleting main_backup
:
$ git branch -D main_backup

- 5,170
- 3
- 33
- 40
You can always do a git reset <commit>
. Perhaps the easiest way to do this is use a graphical frontend, i.e. gitk
.
You should perhaps first do a git branch branch-for-failed-experiment
so the work on the experiment isn't lost forever.
Be careful, if you published the branch (i.e., if others could have work based on your to-be-deleted commits), they will be left stranded. Make sure they sync up with you.

- 11,412
- 8
- 32
- 52
-
1I don't think I want anyone ever to know how bad I messed that one up :P – Hoytman Apr 22 '14 at 19:19
-
It sounds like what I really want to do is make a new commit in which all the mistakes are removes (which would look identical to a commit made 3 or 4 commits ago.) How would I do that? – Hoytman Apr 22 '14 at 19:21
-
You can revert the commits. Or you could do something like `git diff
.. – vonbrand Apr 22 '14 at 19:27`, stash the result away in a file, and then `git apply -R` this to undo all changes, then `git commit -a` the result. But I'd prefer to create a clean history (even if it is completely fake).
Note that at any given time you can change where a branch points to by using git update-ref refs/heads/branch id
, but before you do this, you must give a name to the tip of the tree, otherwise your work will unaccessible. So these two commands may do the job
git update-ref refs/heads/newfeature HEAD
git update-ref refs/heads/master XXYYY
But make sure that you do not have any uncommited changes otherwise all hell will break loose

- 591
- 6
- 8