Engineering Craft › Version Control (Git)

Revert

Creating a new commit that undoes an earlier one.

Also known as: git revert, undo a commit, revert commit

git revert undoes an earlier commit by creating a new commit that applies the opposite changes. The original commit stays in history.

git revert 3f9c2a1             # make a new commit undoing 3f9c2a1
git revert HEAD                # undo the last commit
git revert -m 1 <merge-hash>   # undo a merge commit (say which side to keep)
git revert --no-commit A..B    # undo several, as a single commit afterwards

Before:   A ─ B ─ C (main)
After:    A ─ B ─ C ─ C' (main)     C' undoes the changes from C

Why it’s the safe undo

Revert only adds a commit, so nobody’s history changes. That makes it right for anything already pushed or shared, such as a bad change on main. Teammates just pull the new commit.

Compare git reset, which moves a branch back and rewrites history. That’s fine for local commits, and problematic for shared ones.

Details

• Conflicts are possible if later changes touched the same lines. Resolve them, then git revert --continue.
• Reverting a merge needs -m 1 to choose the mainline. Re-merging the same branch later won’t bring back the reverted changes, so you have to revert the revert first.
• The message says “Revert …”. Add why you’re reverting.
• Revert changes the code, not the data. A bad migration or sent email isn’t undone.

In incidents

When a release breaks production, reverting the offending commit and deploying is often the fastest rollback (rollback). Fix the real problem afterwards, and reintroduce the change with a new commit.

See rewriting history for when rewriting is fine and when it isn’t.