The git history command deserves more attention
摘要
文章介绍 Git 2.55 新增的 experimental git history 命令，包含 fixup、reword、split 三子命令。fixup 阶段提交内容到目标旧提交（类似 git commit --fixup + autosquash rebase），并自动重写所有分支顶点（可选限当前分支），但不处理 merge commits；reword 编辑旧提交消息（纯 commit graph 操作，无需改 index 或工作树），split 将一个提交按 hunk 分成两个（交互选 hunk）。所有操作均为原子性，拒绝可能导致冲突的场景，避免半破树状态。文章对比 jj，指出 git history 仅部分实现 jj 的好处（如分支同步），未提供操作日志、undo 或冲突承接，但已集成核心 Git，无需额外安装。结尾乐观认为未来 Git 将支持 first-class conflicts，扩展此限制。
荐读理由
git history fixup 在项目阶段重写老 commit 后自动同步所有分支，重写 hash 后其他 commit 和分支 tip 自动重建，满足你想在动手前判断“什么值得深入”的高信噪需求
原文
The git history command deserves more attention
Working with lots of changes in parallel on git can be painful. You end up juggling branches and commits, and running scary rebase -i commands that can leave your tree in a half-broken state if you so much as sneeze.
jj, an alternative to git, gets discussed a lot these days (1, 2, 3, 4) and is often pitched as a solution. While I’m very sold on the problems jj is trying to solve, the way it solves them hasn’t quite hit home with me. Every 3 months, for the last 1.5 years, I try it out for a few days, really trying to make it part of my workflow but eventually I give up and go back to git.1
That’s where git history comes in. It’s an experimental command that arrived across two releases, 2.54 (April, reword and split subcommands) and 2.55 (June, fixup subcommand). It got a flurry of attention on each release day, and then, as far as I can tell, not much community discussion since. Which is a shame, because IMO it already delivers several of the benefits people tout for jj without needing to switch your whole workflow. And the cool thing is that it’s part of the core git distribution, so you can try it without installing anything.
There are three subcommands: fixup, reword and split.
fixup
git history fixup fixes an old commit that has something wrong in it, then autorebases all your branches to match.
You stage the fix as usual with git add, then run git history fixup <commit> to fold those staged changes into the target commit. It’s like a git commit --fixup plus an autosquash rebase but with the extra magic that it also updates any other branch which contained that commit.
That last part goes further than git rebase --update-refs, which only moves refs sitting inside the range you’re actively rebasing. git history instead finds and rewrites every local branch descended from the commit (while also having an option to limit it to only the current branch). On the other hand it does not work in the presence of merge commits which, for some usages of git, is going to be a dealbreaker.
Here’s how it works in practice:
Before, with a fix staged for B:
After git history fixup B:
B* is B with the fix folded in. Rewriting a commit gives it a new hash, so C and D are automatically re-created on top as C* and D*, and the feat-1 and feat-2 branch tips move with them.
The most important property, common to all three commands, is that it’s atomic: it never leaves your tree in a half-broken state. It manages this by refusing any operation that could produce a conflict.
To be clear, this is strictly less powerful than jj. jj treats conflicts as first class so it can carry a conflicted state through a rebase and let you sort it out later. git history doesn’t do this yet but the docs leave the door open:
“This limitation is by design as history rewrites are not intended to be stateful operations. The limitation can be lifted once (if) Git learns about first-class conflicts.”
So basically, this limitation may change in the future; excited to see if it does!
reword
git history reword updates the commit message on an old commit and automatically rebases everything on top. This is very useful for going back and fixing commit messages when the design shifts as you iterate.
git history reword <commit> opens your editor with that commit’s existing message. You edit it, save, and the rest of the stack is rebuilt on top with the branches following along. It’s exactly like fixup but for commit messages instead of the tree contents.
Because it only changes a message, reword (like split later) never touches your index or working tree at all; it works purely on the commit graph. So both let you rewrite a commit on a branch you don’t have checked out without disturbing whatever you’re in the middle of.
Before:
After git history reword B:
Only B’s message changes, but that still gives it a new hash, so C is rebuilt on top as C* and feat-1 follows along.
split
git history split takes one commit and splits it into two, interactively picking what you care about from each. It’s the equivalent of git add -p, but without needing gymnastics with git rebase. I’ve found this to be the most specialized of the three, but invaluable when I need it.
Specifically, git history split <commit> drops you into a hunk-by-hunk prompt over that commit’s diff. The hunks you keep make up the first commit, the rest fall into the second.
Before, with B bundling two unrelated changes:
After git history split B:
B becomes B1 and B2, and C is rebuilt on top of the pair as C*.
Conclusion
Judging by how many people are using jj, I do think there’s still some key mental shift which I’m not yet making. And to be clear, git history doesn’t close the full gap: jj still gives you an operation log with easy undo, models your working copy as a commit, and can carry conflicts through a rebase, none of which this is trying to do.
But for now, git history is a big step forward in adopting many of the pieces that attract people to jj, and it’s already in the tool I use every day. And the way the documentation is written makes me hopeful that more improvements will be coming in upcoming releases!
If you enjoyed this post, consider subscribing to my newsletter or following via RSS. You can also share it on Hacker News or Lobsters.
- If there’s interest, happy to write up a post on my experience on this. ↩︎
这条对你有帮助吗？