Completely remove commit from git database

1. List all branches that contain the commit:

   git branch --contains COMMITSHA
   

2. Remove commit from these branches:

   git checkout BRANCH
   git rebase -i COMMITSHA^
   # delete line with commit and save
   

   If a changed branch is tracked in any remote, push it there with override:

   git push --force REMOTE BRANCH
   

   e.g:

   git push --force origin master
   

   (Note that, depending on your development process, the commit may appear in untracked remote branches as well.)

3. Purge the commit, so it can not be restored from your local repo:

   git reflog expire --all BRANCH1 BRANCH2 # list all branches you changed
   git prune --expire now
   

   Note that you must also run this command on all remote repositories that had this commit. If you have no access to the remote repo, you have to cross your fingers — commit will eventually expire by itself and will be purged by git gc.

   Be warned that above commands will remove all dangling objects from Git repo and all the history of branch changes — so you wouldn't be able to restore (by non-forensic means) anything lost prior to its run.

4. Tell all collaborators to fetch changed branches and update any work they might have based on them. They should do as follows:

   git fetch REMOTE      
   

   For each branch that is based on a branch you changed above (including the branch itself if they have it locally):

   git checkout BRANCH
   git rebase REMOTE/BRANCH
   git reflog expire --all BRANCH
   

   After they're done:

   git prune --expire now
   

1. Make sure that no refs need that commit (reset back to before it, or rebase it out)
2. Delete the object from .git/objects (it'll be in a folder named after the first two characters of the hash, and the filename will be the rest of the hash).

Note, however, that if you pushed this commit to a public repository, removing it locally won't remove it from the remote.