git - remove commits from merged branch

Question

I accidentally pushed a commit with some huge files, and then reverted it. But it causes anyone pulling this branch to fetch these files in history, so I decide to remove or squash these two commits. However, some branch has been merged in. I don't know how to make "git rebase -i" keep the branches structure.

The history now looks like:

H - new commits
|
G - merge
| \
|  F - commits on another branch
|  |
E  | - some other commits
|  |
D  | - corrected B
|  |
C  | - revert B
|  |
B  | - huge files
| /
A - early commit

Can I change it to following?

h - new commits
|
g - merge
| \
|  F - commits on another branch
|  |
e  | - some other commits
|  |
d  | - corrected B
| /
A - early commit

Interactive rebase and remove `huge files` and `revert B`. Then redo the merge. Finally, cherry-pick the new commits coming after that. — Tim Biegeleisen, Sep 16 '16 at 12:21
@TimBiegeleisen Thanks. Is there any scripts to do it? Because there some new branches and merges after that point, replaying by hand is difficult. — Xiao, Sep 16 '16 at 12:39
Given that so many commits and merges have happened since you committed the large file, you might want to look [here](http://stackoverflow.com/questions/2100907/how-to-remove-delete-a-large-file-from-commit-history-in-git-repository) to see how to prune away this large file. — Tim Biegeleisen, Sep 16 '16 at 12:41
Doing an interactive rebase by hand is risky now, and might be error prone. — Tim Biegeleisen, Sep 16 '16 at 12:42
@TimBiegeleisen @nonsensickle Thanks. `git filter-branch` helped saves space. Though haven't figured out how to completely remove the commits with it yet, it's enough for this case. — Xiao, Sep 16 '16 at 13:12

score 2 · Accepted Answer · edited May 23 '17 at 12:32

2

Yes you can but you shouldn't.

It will require you to rewrite the history and replace the already pushed history on the server (which requires force push and most often results in everyone yelling at you).

But if you really want to then git filter-branch is what you want to use, much like in this SO answer. So you would do something like this:

git filter-branch --commit-filter '
    if [ "$GIT_COMMIT" = "<your commit to remove here>" ]
    then
        skip_commit "$@";
    else
        git commit-tree "$@";
    fi'  HEAD

There are a few more examples here.

edited May 23 '17 at 12:32

Community

1
1

answered Sep 16 '16 at 12:55

nonsensickle

4,438
2
34
61

Great, this is what I'm looking for. It's an internal repo and not a problem to have others do a forced-pull. – Xiao Sep 16 '16 at 13:15

score 0 · Answer 2 · answered Sep 16 '16 at 12:45

No, you cannot.

The hash of each commit includes not just the hashed contents of all the files in the index, but also the hashes of the commit's parents.

"Corrected B" would have a different parent than it has now. That would change the hash.

It is possible to fix this, but it is not possible to avoid force-pushing the fix, and have everyone force-pull it, once that's done.

The only real way to fix this would be the following process:

 git checkout A

Check out the parent commit, before the branch forked. Then, create two work branches:

 git checkout -b corrected-mainline
 git checkout -b corrected-fork

You're now on the corrected-fork branch. Now, on this branch, you should be able to git cherry-pick all the commits up to the F commit, skipping the commit with the huge files, and its revert. If any of these commits are a merge, do the same merge.

Now, do the same on the corrected-mainline branch

 git checkout corrected-mainline

Now, git cherry-pick all the commits on the mainline branch, up to the last commit before the G merge. Then, do a merge with the corrected-fork branch. This is now your corrected G merge commit.

Finish cherry-picking any commits that were originally on top of G in your diagram, finishing with the last commit on your original branch.

At this point, the contents of your work branch should be the same as the contents of your original branch. Verify that this is so.

Then, delete your original branch, the rename your work branch to its name. Or, use git reset --hard to reset your original branch to your corrected work branch.

Then do a git push --force, to push your corrected branch up. Of course, anyone that pulls the branch will end up getting a force-pull.

I'm using `git filter-branch`. Thanks for your answer. – Xiao Sep 16 '16 at 13:19 — Xiao, Sep 16 '16 at 13:19

git - remove commits from merged branch

2 Answers2