82

I have a newbie question about Git:

I need to move back and forth in a history of a branch. That means, I need to get all the files to the state they were in in some old revision, and then I need to get back to the latest state in the repository. I don't need to commit.

With SVN, it would be

svn up -r800

to get to revision 800, and 

svn up

to get in sync with the repository.

I know the hash of the commit I want to get back to, so I tried

git reset <hash>

which seems to get me there. But then I tried

git pull

but that complains about conflicts.

So what's the proper way to move through the history of the branch?

I'm thinking in terms of SVN, so don't hezitate to point me to some nice tutorial. Note that I've already checked http://git.or.cz/course/svn.html and http://www.youtube.com/watch?v=8dhZ9BXQgc4 .

Thanks, Ondra.

dotancohen
  • 27,834
  • 33
  • 133
  • 191
Ondra Žižka
  • 40,240
  • 36
  • 196
  • 259
  • 1
    Side note: I got used to avoid `git pull` altogether. Instead, I use `git fetch --all` aliased to `gu` in bash, and have `gitk` open all the time, viewing all branches - see View -> edit -> check all 4 checkboxes. Then I move using `git reset` or `gist stash` + `git co`, depends on what I need. – Ondra Žižka Mar 16 '12 at 21:09

5 Answers5

74

Well, I'm a former svn user too, and now use git for all my projects.

When using git, you should change the way of thinking from the client-server architecture that's used in svn. In svn, every change needs a connection with server. Using git, your repo is in the working directory. You don't need a connection for every repo action.

Only use git push and git pull to synchronise with repo. Think of it like using rsync or any backup solution, to make two place have exactly same content. Just like you connect external backup hard disk, then make the content in it same with the content in your main. That's the usage of git pull and git push.

If you just want to go back and forth the history, do it using git checkout. See the revision id using git history. If you're using Linux, use gitk to see the revision tree. In Windows, tortoise git can display it using revision graph.

To get back to latest revision, use git checkout master. Before doing any command, always make yourself do git status. This command will display anything you need to know about current repo condition, and what action that you need to do to make it right. Before do git pull and git push, it's better to make sure that git status result is contain text working directory clean.

If you need to revert a file to it's previous revision, you can do it with git merge. Before doing it to a file, test it first with git diff. Ex: git diff rev1:rev2 filename. It will print out any different between two revision. Change in rev1 will be replaced by the changes in rev2. So to do revert, rev2 will be the older than rev1. After you satisfy with the diff result, do it with git merge, just replace diff with merge, all other parameters stay the same.

I hope this helps you. The main key is to see that your working dir is your repo. Understanding this will help you use git to it's full capability. Good luck.

Donny Kurnia
  • 5,177
  • 5
  • 33
  • 49
43

You can use git checkout to checkout any commit and then use it with a branch name to go back to a named branch.

git checkout with a commit id and not a branch name moves you off any named branch and on to what is known as a detached head.

If you use git reset then it will move your branch itself back to an old state, orphaning the more recent commits which probably isn't what you want.

CB Bailey
  • 700,257
  • 99
  • 619
  • 646
  • What the `checkout` really does? Does it change the files to the state of the given commit, by reversly applying the patches of the commits between current state and the given commit's state? – Ondra Žižka Jan 22 '10 at 23:43
  • I've found `git reset` and `git reset origin`. I will not do any commits, so it does not matter whether I loose the recent commits as far as it's easily recoverable (which means, by a single command). – Ondra Žižka Jan 22 '10 at 23:45
  • 1
    git checkout don't make the change permanent. To do it, use git merge. See my answers. – Donny Kurnia Jan 22 '10 at 23:46
  • 1
    checkout reset's the index/cache/staging area to the tree designated by the given commit and then updates the working tree to match the cache. It doesn't need to do any patch re-applying; it just uses the tree as at the given commit. (It also has error checking and updates the checked out branch if you pass it a branch name.) – CB Bailey Jan 22 '10 at 23:54
32

The other answers are informative, but I believe this is closest to what the OP wants:

Add these two functions to your ~/.bashrc:

# checkout prev (older) revision
git_prev() {
    git checkout HEAD~
}

# checkout next (newer) commit
git_next() {
    BRANCH=`git show-ref | grep $(git show-ref -s -- HEAD) | sed 's|.*/\(.*\)|\1|' | grep -v HEAD | sort | uniq`
    HASH=`git rev-parse $BRANCH`
    PREV=`git rev-list --topo-order HEAD..$HASH | tail -1`
    git checkout $PREV
}

Usage:

$ git_prev
Previous HEAD position was 7042c8a... Commit message2
HEAD is now at d753ecc... Commit message1

$ git_next
Previous HEAD position was d753ecc... Commit message1
HEAD is now at 7042c8a... Commit message2

Note: These commands always enter detached HEAD state. If you git_prev then git_next from a currently checked out branch, you will end up back at the latest revision but you will be in detached HEAD state. Do git checkout BRANCH_NAME to get back to normal.

Community
  • 1
  • 1
Raine Revere
  • 27,477
  • 4
  • 33
  • 48
  • Once you're doing such things, take a looking a look at git aliases may be worthwile. – rethab Oct 13 '15 at 16:07
  • 1
    I love this answer and it was working for a minute but now when I run git_next i get this ```usage: grep [-abcDEFGHhIiJLlmnOoqRSsUVvwxZ] [-A num] [-B num] [-C[num]] [-e pattern] [-f file] [--binary-files=value] [--color=when] [--context[=num]] [--directories=action] [--label] [--line-buffered] [--null] [pattern] [file ...] ``` – Squirrl Oct 30 '18 at 18:43
  • I had the same problem as above, which seems to be because `git show-ref -s -- HEAD~` is returning nothing. I had some luck replacing the first line of `git_next` with `BRANCH=git branch --contains HEAD | grep -v HEAD | sort | uniq`. – deltacrux May 03 '21 at 03:04
15

Try git reflog, this lists commits and checkouts you have done to switch between the commits, even the commits you have lost when checkout to a previous commit.

Then you can try git checkout <hash of a commit> to switch to that commit.

Hope this helps!

Ryan Le
  • 151
  • 1
  • 2
7

To checkout a different version of a file, use

git checkout rev -- filename

Where rev can be the ID of a commit, the name of a branch, the name of a tag, or a relative version.

Use git log, gitk to look examine versions to see which version of the file you want.

To make this version of the file permanent, you need to commit the file: git add filename; git commit filename

I would not recommend git pull to examine versions because it does a merge -- potentially modifying your current state.

You do not need to use git reset in this case, unless you git add a file you decide not to commit.

Jamey Hicks
  • 2,310
  • 1
  • 14
  • 20
  • sweet jeebus be careful when doing this. My coworker did this command to try out changes from my branch, wiped out about a weeks worth of my work and pushed it to the repo by doing this... – karina May 02 '16 at 22:37
  • 1
    As long as your coworker didn't do a push -f, you could always revert his commit and recover the missing work. If you have access to the repo he pushed to, even a force push is recoverable via git reflog. Also, if you still have your branch, it is recoverable. But yes, you have to be careful. – Jamey Hicks Jun 29 '16 at 00:18