Question

What is the simplest way to undo a particular commit that is:

• not in the head or HEAD
• Has been pushed to the remote.

Because if it is not the latest commit,

git reset HEAD

doesn’t work. And because it has been pushed to a remote,

git rebase -i

and

git rebase --onto

will cause some problem in the remotes.

More so, I don’t want to modify the history really. If there was bad code, it was there in the history and can be seen. I just want it out in the working copy, and I don’t mind a reverse merge commit.

In other words, what is the Git equivalent of the following svn commands:

svn merge -r 303:295 http://svn.example.com/repos/calc/trunk

which removes all changes from 295 to 302 by reverse merging all changes in those revisions, as a new commit.

svn merge -c -302 ^/trunk

which undoes the 302 commit, of course by adding another commit that reverse merges the changes from that respective commit.

I thought it should be a fairly simple operation in Git and a fairly common use case. What else is the point of atomic commits?

We have staging stashing and all to ensure the commits are perfectly atomic, shouldn’t you be able to undo one or more of those atomic commits easily?

Answer 1

Identify the hash of the commit, using git log, then use git revert <commit> to create a new commit that removes these changes. In a way, git revert is the converse of git cherry-pick — the latter applies the patch to a branch that’s missing it, the former removes it from a branch that has it.