Move existing, uncommitted work to a new branch in Git

I started some work on a new feature and after coding for a bit, I decided this feature should be on its own branch.

How do I move the existing uncommitted changes to a new branch and reset my current one?

I want to reset my current branch while preserving existing work on the new feature.


Update 2020 / Git 2.23

Git 2.23 adds the new switch subcommand in an attempt to clear some of the confusion that comes from the overloaded usage of checkout (switching branches, restoring files, detaching HEAD, etc.)

Starting with this version of Git, replace the checkout command with:

git switch -c <new-branch>

The behavior is identical and remains unchanged.


Before Update 2020 / Git 2.23

Use the following:

git checkout -b <new-branch>

This will leave your current branch as it is, create and checkout a new branch and keep all your changes. You can then stage changes in files to commit with:

git add <files>

and commit to your new branch with:

git commit -m "<Brief description of this commit>"

The changes in the working directory and changes staged in index do not belong to any branch yet. This changes the branch where those modifications would end in.

You don't reset your original branch, it stays as it is. The last commit on <old-branch> will still be the same. Therefore you checkout -b and then commit.


Alternatively:

  1. Save current changes to a temp stash:

    $ git stash

  2. Create a new branch based on this stash, and switch to the new branch:

    $ git stash branch <new-branch> stash@{0}

Tip: use tab key to reduce typing the stash name.


If you have been making commits on your main branch while you coded, but you now want to move those commits to a different branch, this is a quick way:

  1. Copy your current history onto a new branch, bringing along any uncommitted changes too:

     git checkout -b <new-feature-branch>
    
  2. Now force the original "messy" branch to roll back: (without switching to it)

     git branch -f <previous-branch> <earlier-commit-id>
    

    For example:

     git branch -f master origin/master
    

    or if you had made 4 commits:

     git branch -f master HEAD~4
    

Warning: git branch -f master origin/master will reset the tracking information for that branch. So if you have configured your master branch to push to somewhere other than origin/master then that configuration will be lost.

Warning: If you rebase after branching, there is a danger that some commits may be lost, which is described here. The only way to avoid that is to create a new history using cherry-pick. That link describes the safest fool-proof method, although less convenient. (If you have uncommitted changes, you may need to git stash at the start and git stash pop at the end.)


The common scenario is the following: I forgot to create the new branch for the new feature, and was doing all the work in the old feature branch. I have commited all the "old" work to the master branch, and I want my new branch to grow from the "master". I have not made a single commit of my new work. Here is the branch structure: "master"->"Old_feature"

git stash 
git checkout master
git checkout -b "New_branch"
git stash apply