Undoing Changes

Undoing Changes#

Learning Objectives#

  • Unstage file changes that were staged accidentally using git reset HEAD <file>

  • Undo the last commit while keeping the changes in the working directory using git reset -soft HEAD~1

  • Revert a commit by creating a new commit with the opposite changes using git revert <commit>

  • Perform a hard reset to a previous commit and discard all changes after that commit using git reset --hard <commit>

  • Understand the implications and warning associated with pushing after a reset

Removing files from the staging area#

We’ve got quite a bit of outstanding stuff we could add to our cheatsheet and good practice guide. Let’s make a note of these things in a TODO.txt file, which we’ll put in the root folder of our repository:

TODO:

Add note about staging multiple files with `git add` and `git diff`

Add note to commit good practice about perils of `git add .`

Add cheatsheet entries for `git rm` and `git mv`

Add cheatsheet entries for pushing and pulling

We might as well tick off the first two items, so let’s add the following content to Git-cheatsheet.md:

## Specifying multiple files

`git <command> path/to/directory` — Apply `<command>` to all files in and descended
                                    from `path/to/directory`

Examples:

`git add .` — Stage all changes to files in current directory or descended from
              current directory.

`git diff foo/` — Show diffs of all files in directory `foo` or descended from
                  `foo`.

And we add the following to Good-practice-guides/Commit-good-practice.md:

Make sure you know what you’re committing#

Take care when staging multiple files with e.g. git add . that you don’t stage changes you didn’t mean to. Always check what you’re committing with git diff or git status.

We go ahead and stage these, using our recently learnt syntax for staging changes to multiple files (our working directory is the repository root folder, as usual):

$ git add .

Wait! Something doesn’t feel right… Let’s check the status:

$ git status
On branch main
Your branch is up to date with 'origin/main'.

Changes to be committed:
  (use "git restore --staged <file>..." to unstage)
        modified:   Git-cheatsheet.md
        modified:   Good-practice-guides/Commit-good-practice.md
        new file:   TODO.txt

Ah, no! We don’t want to commit our TODO.txt file. This was just to help us keep track of our work and it doesn’t belong in the repository. Fortunately we can remove changes to a file from the staging area. In fact, git status tells us how to do this. The general command to use is:

git restore --staged <files>

So let’s use this on our TODO.txt file. We run the command, then check that the only differences staged are for the cheatsheet and good practice guide

$ git restore --staged TODO.txt

$ git diff --name-only --staged
Git-cheatsheet.md
Good-practice-guides/Commit-good-practice.md

That’s better, now we can go ahead and commit. (Good thing we checked before committing the first time round!)

$ git commit -m "Add material on basic pathspec usage (directories)"
[main 0984d2b] Add material on basic pathspec usage (directories)
 2 files changed, 21 insertions(+)

Undoing commits#

What if we’d gone ahead and actually committed our TODO.txt file by accident? Git offers a couple of ways to address this:

  • Reverting: Create a new commit that undoes the old commit

  • Resetting: Move HEAD back to a previous commit, so that all the later commits are removed from the commit history.

Reverting (undo a commit by making a new one)#

Let’s suppose we’ve ‘accidentally’ made a new commit which puts TODO.txt under version control, which we want to undo:

$ git log --oneline -3
cc01bda (HEAD -> main) Add TODO.txt
0984d2b Add material on basic pathspec usage (directories)
92b2ac2 (origin/main, origin/HEAD) Create general good practice guides directory

The command

git revert <commit>

can be used to create a new commit that undoes a previous <commit>. In our case, we want to undo the commit where we added TODO.txt, i.e. commit cc01bda. We’ll run that shortly, but first we need to make sure to make a temporary copy of TODO.txt and store it outside the repository. This is because the revert will return the repository to the state before we’d added TODO.txt, which will involve deleting the file. Having done this, we now perform the reversion:

$ git revert cc01bda

Because git revert is actually making a new commit, our text editor pops into life for us to write a commit message. It’s been pre-populated with a helpful message, telling us which commit is being reverted:

Revert "Add TODO.txt"

This reverts commit cc01bdaf30a98d5bfaf5e43838d90522695f251e.

# Please enter the commit message for your changes. Lines starting
# with '#' will be ignored, and an empty message aborts the commit.
#
# On branch main
# Your branch is ahead of 'origin/main' by 2 commits.
#   (use "git push" to publish your local commits)
#
# Changes to be committed:
#       deleted:    TODO.txt
#

We could edit this if we wanted to, but the default is fine, so we just close the file to complete the commit.

$ git revert cc01bda
[main bc9d190] Revert "Add TODO.txt"
 1 file changed, 9 deletions(-)
 delete mode 100644 TODO.txt

Now our commit history includes our new ‘reverting’ commit. Note also that TODO.txt has been deleted.

$ git log --oneline -4
bc9d190 (HEAD -> main) Revert "Add TODO.txt"
cc01bda Add TODO.txt
0984d2b Add material on basic pathspec usage (directories)
92b2ac2 (origin/main, origin/HEAD) Create general good practice guides directory

$ ls
Git-cheatsheet.md  Good-practice-guides/  README.md

Note on git revert#

The revert command will only work if the working tree and staging area have no changes in them.

Resetting (move back to a previous commit and lose later ones)#

Reverting a commit is often considered a ‘safe’ way to undo a commit, because the original, offending commit is not actually lost. This way, if we decide we did in fact want to make that commit, we can easily recover it (by doing git revert on it). There is a more destructive way to undo commit history that will get rid of the commits.

The git reset command is used to move back our HEAD to an earlier commit. The default form of the command is:

git reset <commit>

In effect, this will ‘rewind’ the commit history back to finish at the given <commit>, dropping later commits as if they’d never happened. However, it will put all changes since <commit> in the working tree, giving us a chance to work with the files as they currently are before the reset.

For example, let’s create an empty file foo.txt. We can do this using the touch command within the root folder of our repository:

$ touch foo.txt

We’ll make a commit of this new file and remove it via a reset. First, we make the commit:

$ git add foo.txt

$ git commit -m "Add foo.txt to practice resetting"
[main fcecec0] Add foo.txt to practice resetting
 1 file changed, 1 insertion(+)
 create mode 100644 foo.txt

$ ls
foo.txt  Git-cheatsheet.md  Good-practice-guides/  README.md

Let’s now reset the commit history back to how it was before we committed TODO.txt. We check the log to get the commit identifier, perform our reset, then check the status of the repository:

$ git log --oneline -5
fcecec0 (HEAD -> main) Add foo.txt to practice resetting
bc9d190 Revert "Add TODO.txt"
cc01bda Add TODO.txt
0984d2b Add material on basic pathspec usage (directories)
92b2ac2 (origin/main, origin/HEAD) Create general good practice guides directory

$ git reset 0984d2b

$ git log --oneline -5
0984d2b (HEAD -> main) Add material on basic pathspec usage (directories)
92b2ac2 (origin/main, origin/HEAD) Create general good practice guides directory
5cf8321 Remove rubbish.txt
d26a698 Add some rubbish to try out 'git rm'
910bb79 Add note about '--name-only' option to 'git diff'

We can see that our new HEAD, i.e. our new ‘current commit’, is what we reset to, namely 0984d2b. What state will our working tree be in? The answer is that it will contain the changes that would need to be made to recover the state of the repository as it was at HEAD just before the reset (at the now-removed commit fcecec0). In other words, we expect to see just the change that adds foo.txt. We can verify this with git status:

$ git status
On branch main
Your branch is ahead of 'origin/main' by 1 commit.
  (use "git push" to publish your local commits)

Untracked files:
  (use "git add <file>..." to include in what will be committed)
        foo.txt

nothing added to commit but untracked files present (use "git add" to track)

We don’t want to keep this foo.txt file, so let’s just delete it with the usual Unix command rm to get a nice, clean working tree again.

$ rm foo.txt

After all that, we can move that TODO list back into the repository. We’ll finish by removing the tasks we’ve completed and add some new tasks, leaving our list like so:

TODO:

Add cheatsheet entries for `git rm` and `git mv`

Add cheatsheet entries for pushing and pulling

Add cheatsheet entries about undoing changes

Hard reset#

If you’re really sure you don’t need to keep the changes from later commits when resetting, then you can automatically discard them by using the --hard option: git reset --hard <commit>. This will set your working tree to the exact same state as the commit you’re resetting to. It thus deletes the work you did after <commit>; be sure that this is what you want before running a hard reset!

Warning: pushing after a reset#

Something to be aware of when using git reset is that you can’t by default push to a remote repository if the tip of your local main branch is pointing to a commit that is behind where the remote repository (origin/main) is.

For example, suppose we had reset our local repository to the commit before origin/main, i.e. commit 5cf8321, “Remove rubbish.txt”. If we then tried to push our local repository to the remote one, Git would not complete the request and complain to us with the following message:

$ git push origin
To https://github.com/jbloggs9999/git-good-practice.git
 ! [rejected]        main -> main (non-fast-forward)
error: failed to push some refs to 'https://github.com/jbloggs9999/git-good-practice.git'
hint: Updates were rejected because the tip of your current branch is behind
hint: its remote counterpart. Integrate the remote changes (e.g.
hint: 'git pull ...') before pushing again.
hint: See the 'Note about fast-forwards' in 'git push --help' for details.

The main reason for this is to avoid a situation where commit history in the remote repository gets lost. This is especially important when collaborating with others, which we’ll cover in later episodes in this course.

You can read more on the topic of resetting on Atlassian’s tutorial.

Important note: sensitive data#

The methods for undoing commits discussed here are appropriate when the commit you made in error contains changes you didn’t want, but don’t fundamentally matter if they’ve been recorded in the repository. In fact, even with a reset, it is possible to recover the commit, with some advanced Git use that involves something called the reflog.

This means that, if you accidentally commit sensitive data, such as passwords or confidential information, you cannot work on the basis that git reset or git revert has removed the data from the repository. Moreover, if you pushed the commits to a remote repository, the information will be stored there.

In these cases, you need to use specialist tools to remove all traces of the sensitive data from the repository. See the GitHub documentation for information on this topic.

Summary Quiz#