In this post I will show how I sometimes recover wrong changes (commits) in a coding project, using git on the command line.
Why would I want to do this?
In my thesis, I’m working on a project that I develop in one environment, and then test in another environment composed of multiple virtual machines. So each important change that I do may have a significant impact on the functionalities of the project. Sometimes, the change I do might not have the result I expected. Then I have to see the changes and analyze the project’s behavior before and after the last commit.
How do you see the last commit?
To test a specific commit, you need the hash. To get the hash you can run git log, then you get this output:
root@debian:/home/debian/test-project# git log
commit <last commit hash>
Author: Isabel Costa <example@email.com>
Date: Sun Feb 4 21:57:40 2018 +0000
<commit message>
commit <before last commit hash>
Author: Isabel Costa <example@email.com>
Date: Sun Feb 4 21:42:26 2018 +0000
<commit message>
(...)
You can also run git log --oneline to simplify the output:
root@debian:/home/debian/test-project# git log --oneline
<last commit hash> <commit message>
cdb76bf Added another feature
d425161 Added one feature
(...)
To test a specific commit (e.g.: <before last commit hash>), that you think has the last working version, you can type the following:
git checkout <commit hash>
This will make the working repository match the state of this exact commit.
After you do this you’ll get the following output:
root@debian:/home/debian/test-project# git checkout <commit hash>
Note: checking out '<commit hash>'.
You are in 'detached HEAD' state. You can look around, make experimental changes
and commit them, and you can discard any commits you make in this state without
impacting any branches by performing another checkout.
If you want to create a new branch to retain commits you create, you may do so
(now or later) by using -b with the checkout command again. Example:
git checkout -b new_branch_name
HEAD is now at <commit hash>... <commit message>
After analyzing the specific commit, if you then decide to stay in that commit state, you can undo the last commit.
How to undo this commit?
If you wish to undo/revert the last commit you can do the following, using the commit hash that you get from the git log command:
git revert <commit hash>
This command will create a new commit with the “Revert” word in the beginning of the message. After this, if you check your repository status, you’ll notice that you have the HEAD detached at the commit you tested before.
root@debian:/home/debian/test-project# git status
HEAD detached at 69d885e
(...)
You don’t want to see this message, so to fix this and attach back the HEAD to your working repository, you should checkout the branch you are working on:
git checkout <current branch>
During the making of this post, I found this tutorial — Undoing Commits and Changes — by Atlassian, which describes very well this issue.
Summary
If you want to test the previous commit just do
git checkout <test commit hash>; then you can test that last working version of your project.If you want to revert the last commit just do
git revert <unwanted commit hash>; then you can push this new commit, which undid your previous commit.To fix the detached head do
git checkout <current branch>.
You can find me on Twitter, LinkedIn, Github, Medium and my personal website.
Latest comments (20)
The last time when I tried to undo the commit I failed, but this guide gives me enough knowledge about it.
Thanks!
MBOX Converter
This is the safest option if you want to preserve the changes you made but simply need to fix the commit message or add more changes before committing again.
Use the following command:
git reset --soft HEAD~
This command:
Uses git reset to modify the HEAD, which points to the latest commit.
The --soft flag tells Git to keep the changes made in the last commit as unstaged changes in your working directory.
HEAD~ refers to the parent commit of the current HEAD, effectively undoing the last commit.
Join the online community of geometry dash lite! Millions of players share their creations, challenges, and experiences, fostering a passionate atmosphere.
Works great! 👍😃
Have you considered the fact that the code is pushed to a remote repository like Github. I am sorry but the explanations are not very clear for me.
visual studio git tools you can select undo last commit
If it's in feature branch I just do hard reset and force push. Revert commit is best solution to revert merge pull request in main/master branch.
Great information that you presented.Commits and changes.Great reach out on your message.
That is bad. Do not do this.
Could you explain why?
You can undo the latest commit with
git reset --soft HEAD^too. But if you already pushed your changes a revert could be a better option.I use
git reset HEAD~1to undo the last commit from a local branch and keep the modifications, this is a way for easily fix the changesalso a really powerful command is interactive rebase, though also it changes history for example
it means you want to make changes in last 3 commits, you can squash them, delete them, change messages, by the way I use it a lot in my feature branches even when I already pushed some commits then I do forced push to rewrite history on the server, it is ok till I know that only I work on that branch
If you just have a local commit you can also use the
git reflogindexes to move back to a point in history even moving back from doing merges and pulls. Something likegit reset --hard HEAD@{<reflog index number>}Reflog is life
I do undo differently, in my
~/.gitconfigI have this alias:Works great!
Awesome :) more options! I did not know about alias on git.
For a quick revert you can use git reset HEAD^1 --hard and the you can commit it with additional changes. I thing this "prevents" the revert prefix in the git history. What do you think about it?
As long as the commit you are 'undoing' is local and is not part of any remote branch (you have not push-ed it yet), or if this is a one-dev pet project and you are sure nobody is 'consuming' your remote branch (and you can push to it with --force) - git reset is OK.
The thing is - it rewrites history. So if you'll make another commit on top of it but the 'undone' commit is pushed already, the history of your local branch and your remote tracking branch will diverge - and you'd have to push it either to another branch, or with --force, rewriting the remote branch. And if the remote branch is rewritten, anyone who had cloned it won't be able to pull from it any more that easily etc...
Thus the general advise is not to rewrite history of remote branches, and that's what git revert is for.
Funny side note - once seen a commit starting with a word Revert repeated 6 times o_O
Great explanation! In my case, I thought revert was better because I pushed the previous changes before.
Thanks for the suggestion. I used that before, but in the last problems where I had to undo the commits and then push the changes, that was the solution that worked for me. Also, I didn't mind having a commit dedicated to the revert (with the prefix) on the git history. I think it is a possible solution too.