Rewriting Git History

Iron rule: only rewrite commits not yet pushed to a shared branch. Every rewrite changes the SHA of the commit and all descendants.

Amend

git commit --amend              # edit message + staged changes
git commit --amend --no-edit    # fold staged changes, keep message
git commit --amend --author="Name <email>"

Interactive Rebase

git rebase -i HEAD~N
git rebase -i --root
git rebase -i --autosquash      # auto-order fixup!/squash! commits

Commits appear oldest-first (replay order, opposite of git log).

Verb	Effect
pick	use as-is
reword	keep diff, edit message
edit	stop to amend diff/message
squash	fold into previous, combine messages
fixup	like squash, discard this message
drop	remove entirely
exec <cmd>	run cmd between commits (e.g. exec make test)

Splitting

Mark edit, then at the stop:

git reset HEAD^
git add -p && git commit
git rebase --continue

Bulk Rewrite: filter-repo

Prefer git-filter-repo over deprecated git filter-branch.

git filter-repo --path secrets.env --invert-paths   # purge file from all history
git filter-repo --path-rename old/:new/
git filter-repo --mailmap mailmap.txt               # rewrite authors
git filter-repo --subdirectory-filter trunk

After rewrite: force-push with --force-with-lease.

Recovery

git rebase --abort
git reflog && git reset --hard HEAD@{N}

Pitfalls

• --autosquash needs commits made via git commit --fixup=<sha>.
• exec make test between commits pinpoints which rewritten commit breaks the build.
• filter-branch silently corrupts edge cases; filter-repo refuses unsafe rewrites.