Lab: Reset, Revert and Restore — Choose the Right One

Time20 minutes

LevelBeginner → Intermediate

You needLocal Git only — no GitHub account needed

Scenario

Three commands, all of which “undo” something, all of which people confuse. The confusion is expensive in one specific direction: reset --hard is the only one that can destroy work Git never saw, and it is the one people reach for first because the name sounds decisive.

This lab puts you in five situations and asks you to pick. Each one is a thing that happens weekly on a real team.

Objective

Choose correctly between restore, restore --staged, reset --soft and revert; see why reverting a commit in the middle of history can conflict; and see exactly what reset --hard throws away.

Prerequisites

• Git installed
• The working tree and the index — the three places a change can live are the whole subject here

Starting state

mkdir -p /tmp/lab-rrr && cd /tmp/lab-rrr
git init -q -b main .
git config user.email "lab@example.com"
git config user.name "Lab User"

printf 'line1\nline2\n' > config.txt
git add . && git commit -q -m "Initial config"
printf 'line1\nline2\nline3\n' > config.txt
git add . && git commit -q -m "Add line3"
printf 'line1\nline2\nline3\nline4\n' > config.txt
git add . && git commit -q -m "Add line4"

git log --oneline

189b403 Add line4
685027c Add line3
5b94860 Initial config

Caution

Everything here happens in /tmp/lab-rrr. Nothing touches a real repository.

Tasks

For each situation, decide which command applies before running it.

1. Situation A — you edited a file and want the edit gone. Nothing staged, nothing committed.

   echo "oops" >> config.txt
   git status --short

   Which command discards the edit and leaves everything else alone?

2. Situation B — you staged a file by mistake. The edit should stay; it just should not be in the next commit.

   echo "staged" >> config.txt
   git add config.txt
   git status --short

   Which command unstages without touching the working tree? (Then discard the edit as in A.)

3. Situation C — you committed too early. The last commit should not exist, but its changes should stay staged so you can recommit properly.

   git reset --soft HEAD~1
   git status --short
   git log --oneline

   Then recommit: git commit -q -m "Add line4".

4. Situation D — a commit is already pushed and must be undone. Rewriting is not an option. Undo the tip commit:

   git revert HEAD --no-edit
   git log --oneline
   cat config.txt

5. Situation E — undo a commit in the middle of history. Reset the tip revert first so line4 is back, then revert Add line3:

   git reset --hard HEAD~1          # remove the revert from D; line4 returns
   git revert HEAD~1 --no-edit

   It does not go cleanly. Read the output, then cat config.txt. Resolve it so the file contains line1, line2, line4, and continue the revert.

6. The destructive one. Make an uncommitted edit, then:

   echo "unsaved" >> config.txt
   git status --short
   git reset --hard HEAD
   git status --short
   git reflog | head -3

   Where did the edit go?

Practice Mode (in your browser)

Educational Git simulator. Supports the commands needed for this exercise; it is not a complete Git installation. Nothing runs on your machine and nothing is sent anywhere — progress is kept in this browser only.

3 situations: Discard an unstaged edit · Unstage a file without losing the edit · Undo a commit that is already shared. Estimated 10 minutes. Hints, a worked solution and a restart are available; success is checked against the repository state, so any correct route counts.

Open Practice Mode

Loading the simulator…

Validation

After step 5, git log --oneline shows a Revert "Add line3" commit on top of the three originals and config.txt has three lines. After step 6, git status is clean and the word unsaved appears nowhere — not in the file, not in any commit, not in the reflog.

Hints

A. The edit is only in the working tree. git restore <file> overwrites the working tree from the index.

B. git restore --staged <file> copies the index entry back from HEAD. The working tree is untouched, so the edit survives.

C. --soft moves the branch pointer and leaves both the index and working tree as they were. --mixed (the default) would also unstage; --hard would also discard.

D. revert creates a new commit whose changes are the inverse. History only grows.

E. line4 was added on the line after line3. Removing line3 while line4 stays touches adjacent lines, and Git cannot tell whether line4 depended on line3. That is a conflict, and it is a judgement call — which is why Git stops.

6. It went nowhere. The reflog records commits; this edit was never committed.

Solution

A

git restore config.txt

B

git restore --staged config.txt     # status: " M config.txt" — unstaged, edit kept
git restore config.txt              # then discard it

C

M  config.txt
685027c Add line3
5b94860 Initial config

The commit is gone; its change is staged (M in the first column).

D

[main 399dfa2] Revert "Add line4"
 1 file changed, 1 deletion(-)

config.txt now has three lines. The Add line4 commit is still in history.

E — the conflict:

Auto-merging config.txt
CONFLICT (content): Merge conflict in config.txt
error: could not revert 685027c... Add line3
hint: After resolving the conflicts, mark them with
hint: "git add/rm <pathspec>", then run
hint: "git revert --continue".
hint: You can instead skip this commit with "git revert --skip".
hint: To abort and get back to the state before "git revert",
hint: run "git revert --abort".

line1
line2
<<<<<<< HEAD
line3
line4
=======
>>>>>>> parent of 685027c (Add line3)

Resolve by writing the file you want, then continue:

printf 'line1\nline2\nline4\n' > config.txt
git add config.txt
git revert --continue --no-edit
git log --oneline

b32d4d0 Revert "Add line3"
189b403 Add line4
685027c Add line3
5b94860 Initial config

(git revert --abort at any point returns you to before the revert began.)

6

 M config.txt
HEAD is now at b32d4d0 Revert "Add line3"

Status is clean. The reflog’s top entry is the revert — there is no entry for the edit, because the reflog records commits and this was never one.

Explanation

Three places, three commands. A change can be in the working tree, the index, or a commit. restore operates on the first two. reset moves the branch pointer and optionally resets the first two to match. revert creates a commit. Choosing correctly is a matter of asking where is the change now, and where should it be?

Where the change is	Where you want it	Command
Working tree only	Gone	git restore <file>
Staged	Working tree only	git restore --staged <file>
Last commit	Staged	git reset --soft HEAD~1
Last commit	Working tree	git reset HEAD~1
Any shared commit	Undone, history kept	git revert <commit>
Working tree, staged, or unshared commits	Gone	git reset --hard — unrecoverable for uncommitted work

Revert of a middle commit is a merge. Git applies the inverse patch on top of everything that came after. If later commits touched the same region, that patch does not apply cleanly. The conflict markers show HEAD’s version (with both lines) against the reverted parent’s version (with neither), and you decide.

Danger

git reset --hard is the only command in this lab that destroys work with no recovery path. Committed work is always recoverable from the reflog for a while; uncommitted work that --hard discards is not in Git at all. Before any hard reset: git status. If it is not clean, git stash first.

Troubleshooting

Step 5’s revert applied cleanly with no conflict. You skipped the reset --hard HEAD~1 that brings line4 back. Without line4 present there is nothing adjacent to conflict with.

git revert --continue opens an editor. Add --no-edit, or save and close the editor.

Step 3 left the working tree with the change but nothing staged. You ran git reset HEAD~1 (mixed) rather than --soft. Both are valid; --soft keeps it staged.

Clean up

cd /tmp && rm -rf lab-rrr

Related lessons

The Working TreeOne version of each file, no history, no safety net.

The IndexWhat staging actually stores.

Lab: Undo a bad mergeThe same reset-or-revert decision, applied to a merge.

Git TroubleshootingSymptom-first decision tree.

Next lab

Parallel work with worktrees — fix a production bug without stashing, switching, or losing your place.