Remove Git Commit and Clean up Repository

When using Git, sometimes we may find we are in situations that some changes (or files) were commited by mistake and we want to remove them permanently from the repository. Listed below are the steps to do that.

For example, we have following configuration. We committed three changes and we want to remove the second one (with hash bbbbbb).

HASH     COMMENTS
aaaaaa   Initial commit
bbbbbb   The commit we want to remove
cccccc   Some other changes

To remove the bbbbbb, we can use git rebase command. One way to use it is as follows.

git rebase --onto bbbbbb^ bbbbbb

Another (safer, I think) way is to use git rebase -i aaaaaa. The command will open up a VIM editor, in which each commit after aaaaaa lies in a single line.

pick bbbbbb   The commit we want to remove
pick cccccc   Some other changes

To remove commit bbbbbb, we can simply delete the line pick bbbbbb, save and exit VI (:wq).

Now, if you run git log you would not see bbbbbb any more. But the disk space occupied by the commit is not release yet. That is because Git has reflog feature, that will keep unreachable objects for a period of time (3 months initially?). The following command can be used to release the disk space at once. Think again before you run it!

git reflog expire --all --expire-unreachable=now
git fsck --unreachable
git gc --aggressive --prune=now

The are many other usages of git rebase -i. For example, change comments, combine two commits to one, etc. Please refer to its document for more details.