Every time I see one of these nifty git tricks or workarounds I find myself wondering, “why not just use jj?”
You get a nicer, significantly simpler interface. You don’t need any tricks. You don’t have to google how to work yourself out of a bad state, ever. And you get near-perfect git compatibility (ie you can use jj on a shared git repo, doing all the same things, and your teammates won’t know the difference).
I’ve wondered if there is a psychological thing here: someone who spent time memorizing all the git nonsense may have some pride in that (which is earned, certainly), that introduces some mental friction in walking away???
I'm one of the git users you describe who are resistant to jj. jj sounds great, Steve Klabnik's endorsement is very convincing, and I would probably love it, but here's the issue: I've used git for 17 years and have internalized its idiosyncracies sufficiently to practically never run into problems, and help anyone on my team who does.
jj is harder to adopt for people with a thorough mental model of git, because it's harder to accept jj commands at face value. I know it's modifying my git tree, so I feel compelled to grok exactly what it's doing, but that's distracting and time consuming.
People like me should probably trial jj exclusively for two weeks, to shed that antijjotic resistance and form a more clear-headed opinion.
> jj is harder to adopt for people with a thorough mental model of git
No, it really isn’t. I have used git since shortly after it was first released and I’ve written a git implementation.
I switched to jj in one day. And the amount of git arcana I have to keep in working memory is now basically nil. My VCS now works in almost a 1:1 mapping with how my brain wants to interact with my repo rather than having to go through a translation layer.
If you understand what git commands are doing, what jj does is essentially trivial to add to your mental model.
I also get the benefit of being able to use workflows that I always want to use in git but which are an enormous pain in practice. And I get access to wildly powerful new workflows I didn’t even consider because they would be outlandish in git.
What I will say is this: there is certainly an adjustment period, and I also totally hear you about how learning internals can be time consuming.
I think you can get a lot of the way there with at least the core concept with something like this, if you'll indulge me:
With git, you build up stuff on your filesystem. You then select which bits go into a diff in your index, and then when you're done, you stamp out a commit.
With jj, you instead start by creating a commit. In this case, it's empty. Then, every time you run a jj command, it does a snapshot, and this produces a new commit. However, because we want to have a stable identifier for a commit, we introduce a new one: the change id. This is basically an alias for the latest commit snapshot that was made. Your git history is just made up of each of these latest commits for each change.
... does that make any sense? Obviously there's more to all of the features than this, but that's sort of the core way that the histories map to each other.
It's not a trick or workaround. It's a very straightforward use of a command flag on one of the most used git commands. It's even conceptually very simple, you're just rebasing a subset of commits of a branch onto a different parent. Anyone who has ever rebased already has a working mental model of this operation. Framing this issue where knowing a single flag on a command every git user uses every day to perform an operation they already broadly understand as some arcane knowledge and 'nonsense' is ridiculous.
> introduces some mental friction in walking away???
I don't think it's just mental friction. Suppose you've learned git well enough that everything you do in it is automatic and fast, and the things which aren't fast by default you've built aliases and tooling for over the years. Yes, starting from ground zero you might want something like jj, but at the current point in your life you're not starting from ground zero. Switching to jj means learning another tool to achieve similar outcomes on your workflows.
If jj took many weeks of relearning, I might be right there with you. But the overwhelming majority of people I’ve personally seen who try the switch convert within a day, are barely slowed down by day two, and are effectively fluent within three days to a week at most.
It really just depends. I was very comfortable with the git cli. It didn't take long to learn jj's, and I'm faster with it now than I ever was with git, simply because a lot of things are easier to do and take less commands.
I don't think it's pride, more that I don't see the problem that it solves for me. I don't find that git gets in my way at all, and I don't find it confusing. It's a pretty transparent tool that I use every day and hardly notice.
I don't mind other people using jj, but I simply don't feel a need to try it. There's nothing prideful about that, it's just pragmatism.
Pragmatic is using the best tool for the job. Certainly subjectively, and arguably objectively, jj is the better tool in many ways.
My advice is to try it. You should like it, if you're in the majority of folks to give it a sincere shot. If you don't like it, cool. But then my advice changes. jj is probably the future. Adapt or become one of those old-timers who froze their learning in time.
The most common reason is that the git user has no idea it exists. I am in this bucket alongside 99% of git users.
There are an assortment of other potential reasons:
- It is a new tool that hasn't been battle tested as much as git has, which can decrease confidence.
- Git has inertia. People have learned it, it takes less effort to add a new git skill to your repertoire than learn a new tool from scratch, even if the new tool is easy to pick up.
- Due to its novelty, the auxiliary tooling ecosystem for jj is smaller (does it have plugins for all the popular editors? Lots of people like those, git's are high quality)
- Git is good enough. It's not perfect, but its popularity means that its shortcomings have readily available fixes or tweaks from users. It simply isn't bad enough and there are bountiful resources on how to use it effectively.
I think a core reason (besides not knowing jj exists), is the framing that there is a choice that has to be made, or a switch that has to occur. It is, instead, additive. I have Sublime Merge (GUI git client) and jj both looking at my git repo all day. Zed's git stuff is watching it too.
jj is sort of a bag of git tricks for me that I use when needed. It's no different than some things being easier with the git CLI vs others being easier in Sublime. I'll be at a stage where my committing/branching/rearranging wants are something that jj nails perfectly, and I do those there. As far at the other tools are concerned, I just did a bunch of sophisticated git operations.
The "colocated with git" capability of jj is probably it's most amazing feature tbh, and is key to any sort of adoption.
1. It's very new; I haven't had time to learn it properly yet.
2. It's very new and tooling doesn't support it well, e.g. VSCode. There aren't many GUIs yet.
3. I tried it once co-locating with Git and you definitely can't use both at the same time, even if it can use a `.git` directory. It ended up in a huge mess.
I'm definitely in favour of better-than-Git alternatives but I don't think it's reasonable to expect everyone to switch to JJ right now. It isn't so much better that abandoning the de facto standard is obviously worth it yet. (In contrast to things like the iPhone, Rust, SSDs, etc.).
Also I really wish they would focus on some of the bigger pain points of Git. I can deal with rebasing and whatnot. Sometimes it's painful but it's usually not that bad.
What I can't deal with are submodules and LFS. Both are awful and fixing them properly requires fundamental changes to the VCS which aren't going to happen in Git. JJ has an opportunity to do that. Imagine if JJ could say "we have submodules but they aren't awful!" or "you can check in large files!". Those are the sort of huge advantages that would mean you can say "why not just use JJ".
Colocating is the default now, and you should be able to use both at the same time, though running git commands that mutate can confuse jj. Ideally you only use read-only git commands.
Both submodules and LFS are things that jj wants to address, but they take time.
Oh, there's another stgit user! ^5
Coming from darcs, I couldn't use git until stgit came along, and today, it's one of those few tools I can't imagine working without. Nothing else matches my way of code hacking.
So often, I watch people making a big mess with git, and I always recommend stgit to them, so they can post proper and reviewable branches for merging. But in all these years, I could never convince anybody.
It’s a bit like qwertz. Sure, it is not optimal, there are better alternatives available. But it is good enough, and it is universal. That trumps a 5% typing improvement on my own custom keyboard layout at the cost of not being able to use my coworkers keyboard.
Also, I dislike all of the alternate git frontends I tried, because they are opinionated in a way they clash with my workflow.
Moreover, I don’t think the git CLI is that bad. Once you learn some basic concepts, it makes a lot of sense and is pretty consistent.
Most problems people report stem from a refusal to learn the underlying structure and models. That is on them. And when using a different frontend, they don’t disappear either. They are just abstracted, to allow you to avoid learning them. But they are still there, and you will probably still need to know them at some point.
> Most problems people report stem from a refusal to learn the underlying structure and models.
It's very easy to fall into the trap of believing this: git's implementation fits together neatly enough that it feels like the best you could do. Like, yes it's complex, but surely that's just intrinsic complexity of the problem? (Also, I think we all sometimes feel like someone with a different view must just not know as much as us.)
But if you have used other version control systems (I'm thinking particularly Mercurial here) you realise that actually some of that complexity is just totally made up by git.
Jujutsu is not a Git frontend, it is an independent VCS that is compatible with Git. This is important because it has concepts and ideas of its own. It is used entirely independent of Git within Google.
i went from being a "jj cli power user" to relying on jjui for all of my complex rebase needs so quickly that i now have to read the man page to recall basic commands
This particular tip is basic. It just describes how to copy changes from one branch to another with one kind of rebase. But there's always someone who is learning git for the first time and may not even know how it's used.
Git is simple enough and has features and capabilities that jj does not have. Contrary to popular belief, git is not hard to use. I refuse to use any "simpler" system that is slower or less feature-rich than git. I don't even want to learn another commit graph model, because git's model is very good. About 95% of what people like yourself call "git nonsense" consists of useful features that many people would be annoyed to not have.
I believe that a large number of git or general VCS users have no idea about commit hygiene. They have not had to cherry-pick or edit commits, and have no idea what to do about conflicts. To people like that, git's features and methods will appear especially foreign.
I looked over jj specifically many moons ago and concluded it would annoy me and not function at my job. I forgot what the reasons were. One reason was most likely because I need submodules and worktrees to work. I just looked at its FAQ, and saw a bunch of nonsensical new terms as well. Nothing is more compatible with git than git itself, and I am very satisfied with how well git works for me.
submodules aren't native with jj, you use git commands to manage them, and it works fine. Eventually there'll be some sort of native support.
worktrees are called "workspaces" in jj, but are the same.
A lot of people find jj easier to have good commit hygiene with, and find it simpler and more powerful, not less. But that said, if you're happy with git, you should continue to use it.
There is also mental friction with learning an entirely new tool. `jj` is different enough from `git` that one can't transfer knowledge. Currently the documentation is not good enough to assuage that issue.
Frankly, this is a rather poor, uninformed excuse. Most people who try it grok the concepts in days. And they feel right at home in weeks. That's when you start hearing, "I'll never go back."
Wish i could remember my issues with jj. I tried it, i wanted to stick with it because i loved the fact that i could reorder commits while deferring the actual conflicts.. but something eventually prevented me from switching. Searching my slack history where i talked about this with a coworker who actually used jj:
1. I had quite a bit of trouble figuring out a workflow for branches. Since my companies unit of work is the branch, with specifically named branches, my `jj ls` was confusing as hell.
`jj st` might have helped a bit, but there were scenarios where creating an commit would abandon the branch... if i'm reading my post history correctly. My coworker who was more familiar explained my jj problems away with "definitely pre-release software", so at the time neither of us were aware of a workflow which considered branches more core.
Fwiw, I don't even remember when the jj workflow had branches come into play.. but i was not happy with the UX around branches.
2. iirc i didn't like how it auto stashed/committed things. I found random `dbg!` statements could slip in more easily and i had to be on guard about what is committed, since everything just auto pushed. My normal workflow has me purposefully stashing chunks when i'm satisfied with them, and i use that as the visual metric. That felt less solid with jj.
Please take this with a huge grain of salt, this is 10 month old memory i scavenged from slack history. Plus as my coworker was saying, jj was changing a lot.. so maybe my issues are less relevant now? Or just flat out wrong, but nonetheless i bounced off of jj despite wanting to stick with it.
"creating a commit would abandon the branch" is certainly something lost in translation. There are other reasons you may have not liked the UX, largely that if you create branches and then add a bunch of commits after it, the branch head doesn't automatically move by default. There is a config setting you can change if you prefer that, or the `jj tug` alias some people set up.
I more or less use the method described [here](https://steveklabnik.github.io/jujutsu-tutorial/advanced/sim...) for branches. One thing I do change is that I set the bookmark to an empty commit that serves as the head of each branch. When I am satisfied with a commit on head and want to move it to a branch I just `jj rebase -r @ -B branch`. When I want to create a new branch it's just `jj new -A main -B head` and `jj bookmark set branch_name -r @`
Half my team switched to JJ this year, and I do find stacking PRs to be much more pleasant now. We had previously tried out Graphite but it didn't really stick.
I wrote up a little way to use JJ's revsets to make it easy to push an entire stack of branches in one command:
The main issue I kept having when trying to do this with just git is then managing all the branch names to be attached to the right moved commits, so that my stack could be reviewable on github's open PRs.
Does jj help with that at all?
I've experimented a bit with git-town.com (OSS) and now everyone at $DAYJOB uses graphite.com (SaaS) which does that part very well.
It’s one of the core features that rebases, including branch names (bookmarks in jj) work ‘correctly’. You can rebase whole dags, including merges, with multiple named heads with just one jj rebase -b.
`--update-refs` flag helps a lot in vanilla git. That and `--autosquash` should probably be default flags to `git rebase`. I also don't entirely trust rebase without `-i` (`--interactive`), personally. I hear there is talk about shaking up the out-of-the-box default flags in git 3, and I think rebase should especially get new defaults.
I came into the comments specifically to ask if this flag existed. I feel bad that the author developed this whole flow just because they didn't know about this, but that's pretty common with git.
discoverability is a big problem, especially for CLI tools which can't afford to show small hints or "what's new" popups. I myself learned it from someone else, not docs.
update-refs works only in a narrow case when every branch starts form the tip of a previous. Your helper might still be useful if it properly "replants" whole tree keeping its structure.
Running a git command on one branch and multiple branches being affected is really unusual for me! This really does look like it is designed for just this problem, though. Simple overview: https://blog.hot-coffee.dev/en/blog/git_update_refs/
I've settled on a workflow that reverses the situation. I simply commit all my work to the main branch and cherry pick commits into temporary feature branches only when submitting PRs.
This way I only need to worry about maintaining a single consistent lineage of commits. I've been using this workflow for about a year now and find it to be much easier than juggling and rebasing feature branches.
Interesting, I'll be sure to check it out. It sounds pretty similar to the tool I built which lets you edit a "plan" in a text editor to assign commits to feature branches - the plan is saved so it can be amended continuously.
I usually just `git rebase origin/main -i` after the base branch has been merged there, and this means I need to explicitly drop the merged commits, but I can inspect what's happening.
Add `--update-refs` to your interactive rebase and it will give you an easy line to know how many commits to drop because it will add an `update-ref` line for the old branch. You can just easily delete everything up to and including that `update-ref` line and don't have to manually pull up a git log of the other branch to remember which commits already merged.
(Plus, of course, if you have multiple branches stacked, `--update-refs` makes it easier to update all of them if you start from the outermost branch.)
I'm a heavy user of git-spice: https://abhinav.github.io/git-spice (created by a former coworker) and can't really go back to a time without it. While still not nearly as good as Facebook's Phabricator, it's probably the best workflow for small, focused stacked PRs you can achieve in a Github / Gitlab based repository.
Even if `--update-refs` didn't exist, my experience is that git can identify duplicate commits produced by rebase, and knows to skip them when rebasing the same commits to the same place again. Am I imagining that?
What I would really like is a Git equivalent to Mercurial's "fold" operation. I usually make a bunch of commits as I work, just as checkpoints, which I then want to turn into a single final commit when it's done, which could be quite some time later, e.g. "started on thing", "broke it", "broke it more", "Add thing to improve foo of bar".
Mercurial's "histedit" command offers two operations to combine the commits: "fold" and "roll", where "fold" brings the earlier commit into the later commit and "roll" does the reverse. The "fold" operation does exactly what I want in this case: It brings all the changes into the final commit from when I actually finished it, using the commit date of that commit.
Git's "rebase -i" command offers "squash" and "fixup" and "fixup -c" and "fixup -C", but they all do the same thing as "roll", i.e. they keep the earliest commit's date, the date when I started working on the thing and not the date when I finished working on it. (So I then cut and paste the correct date and update the commit afterwards. This may not be the best way to do it.)
That is an interesting desire to use a later commit date rather than an earlier one. So many prefer that commit date of when an effort started.
One way to accomplish that is to reorder the commmits in `rebase -i`: "pick" the last commit first and squash/fixup the rest after. That can produce some very weird merge conflicts, but it works well more often than you might think, too.
At the very least, you can do your hand editing of the commits during the rebase instead of after by switching the earliest commit from "pick" to "edit" to have the full power to amend the commit before it moves on (with `git rebase --continue` when you are satisfied). (Versus "reword" if you just want to change the commit message only.)
Also, instead of naming commits things like "broke it" and "broke it more" an option is to use `git commit --fixup={previous commit hash}`. That auto-generates a "fixup!" name based on the previous commit and auto-marks it as a fixup if you use the `--autosquash` flag to rebase.
I do use fixup commits as well, for fixing small issues discovered after I make the proper commit, and in that case it makes sense to use the date and message of the earlier commit.
But in the case I'm describing, the earlier commits are essentially just temporary snapshots on the way to making a proper commit. Just a more explicit undo history, basically. I usually don't even bother naming them anything as meaningful as "broke it", actually. Maybe most people wouldn't even bother making these commits – or maybe they would if Git supported "fold".
Even for 'normal' rebases of a multi-commit series (without named feature branches), I habitually use the --onto form. It is simply easier to conceptualise what is happening if one is explicit about the 3 references.
As such, using it for the situation described in the piece then becomes a trivial matter, especially if one also habitually runs a graph viewer - I generally have one or more instances of gitk running all the time.
This made good sense, though by the time I got to bits saying "that last step is critical. Without it, your next sync will break" and "Don't forget!" I was laughing out loud. I love git :-)
Stacked commits is awesome, but it sucks that with git you need all these workarounds, and that the servers (GitHub etc) are not designed with that workflow in mind.
I left Google a few months back to work on another project. I missed the internal version control so much that I dropped that other project and am now focused http://twigg.vc
It's very similar to what's used at Meta and Google. Atomic commits, linear history, auto rebase, code review compatible with stacked commits.
I feel that stacked PR in git should be common knowledge, however many of the documentations are scattered. I found stacked.dev but feels that its not exploring pure-git workflow that well. thats why I tried to collect all of those docs into a single website that's dedicated into that singular workflow. would love to get any feedback:
Stacking commits lets you do that without having to wait for each change to be reviewed/merged to the main branch before you iterate on top of those changes.
True. I find I rarely need it: standard rebase or merge do the trick. If they don't the review cycle or PR size may be too high. Super rare I need onto. So rare I look up how to do it when I do.
It's such a complicated way to work though, you start another set of changes then you go back addressing comments then you go back updating the stacked branch and you might need to do that few times...
Teams should focus on getting stuff merged in and not create massive PRs that live forever, life becomes so much easier.
I‘m using git machete since a few months to help with the stacking workflow. The nice thing is that it can use the GitHub CLI to manage the resulting PR stacks as well.
I fail to see how it makes it much easier to review a lot of interdependent branches where branch C needs new logic from branch B which needs commits from branch A.
Sounds like a nightmare to me.
This problem isn't specific to cases where you are rebasing "again and again". Just needing to do a single rebase (e.g. prior to opening a PR) for a stacked feature is enough to need a solution.
And, as others have pointed out, the modern solution is `--update-refs`, so there's no need for complicated workflows any more anyway.
If you mean rebasing as each PR in the stack is merged, most Git platforms have the ability to do that automatically.
No, I mean small chunks that are actually merged instead of having a stack of them floating around.
A rebase before a merge can always happen, of course. But there are not stacked commits then it is just a standard rebase of your small chunk and that's it. And this will also have a shorter life than a stack so rebases will be rarer and simpler.
This marker branch step feels like a workaround to a missing capability. It's something I can easily see one forgetting especially if they haven't been doing stacked diff workflows regularly.
I agree it seems error prone. I'm not sure if I'm misunderstanding something, but I use `git cherry-pick` when I know I need to move commits around that might have conflicts. The problem with rebase can be that the user doesn't fully understand all the options being applied and end up with a "bad" merge.
I don't usually want to rewrite history. I just want the target branch with all my commits on top (I usually squash the feature branch into one commit anyway). I have yet to run into a situation where this isn't good enough.
If the branch diverges so much and has so many commits that this simpler approach doesn't work, that might not be a git problem, but a project management one. It's still always nice to know git has tools to get me out of a jam.
I’ve been meaning to write this article for a long time @flexdinesh . Thanks for taking the time to share this technique for managing stacked diffs using vanilla git rebase!
Eh I just use `git rebase -i` and delete the commits I don't want. Much easier to think about.
But the real problem with this workflow is that neither Github nor Gitlab support it at all. Not even Forgejo does. Which blows my mind because is such an obvious way to work.
As far as I know only Tangled actually supports it, but unfortunately Tangled is tangled up in its own weird federation ideas. There's no way to privately host it at all.
Is there any reason besides merge commits ending up in history to not do this with merges instead? ie merge main into feature-1, then feature-1 into feature-2.
Sounds like using --update-refs would let you do all that in a single operation, but you still need to force-push and don't maintain an explicit merge/conflict resolution history, both of which could be considered sub-optimal for collaborative scenarios.
They meant merging the other way, i.e. merging the new changes from main into the stacked feature branches.
This is functionally the same as rebasing, except that the new changes show up at the tip of the commit chain rather than the base. And because it doesn't rewrite history you don't need to force-push.
Every time I see one of these nifty git tricks or workarounds I find myself wondering, “why not just use jj?”
You get a nicer, significantly simpler interface. You don’t need any tricks. You don’t have to google how to work yourself out of a bad state, ever. And you get near-perfect git compatibility (ie you can use jj on a shared git repo, doing all the same things, and your teammates won’t know the difference).
I’ve wondered if there is a psychological thing here: someone who spent time memorizing all the git nonsense may have some pride in that (which is earned, certainly), that introduces some mental friction in walking away???
I'm one of the git users you describe who are resistant to jj. jj sounds great, Steve Klabnik's endorsement is very convincing, and I would probably love it, but here's the issue: I've used git for 17 years and have internalized its idiosyncracies sufficiently to practically never run into problems, and help anyone on my team who does.
jj is harder to adopt for people with a thorough mental model of git, because it's harder to accept jj commands at face value. I know it's modifying my git tree, so I feel compelled to grok exactly what it's doing, but that's distracting and time consuming.
People like me should probably trial jj exclusively for two weeks, to shed that antijjotic resistance and form a more clear-headed opinion.
> jj is harder to adopt for people with a thorough mental model of git
No, it really isn’t. I have used git since shortly after it was first released and I’ve written a git implementation.
I switched to jj in one day. And the amount of git arcana I have to keep in working memory is now basically nil. My VCS now works in almost a 1:1 mapping with how my brain wants to interact with my repo rather than having to go through a translation layer.
If you understand what git commands are doing, what jj does is essentially trivial to add to your mental model.
I also get the benefit of being able to use workflows that I always want to use in git but which are an enormous pain in practice. And I get access to wildly powerful new workflows I didn’t even consider because they would be outlandish in git.
Glad you like that I like it!
What I will say is this: there is certainly an adjustment period, and I also totally hear you about how learning internals can be time consuming.
I think you can get a lot of the way there with at least the core concept with something like this, if you'll indulge me:
With git, you build up stuff on your filesystem. You then select which bits go into a diff in your index, and then when you're done, you stamp out a commit.
With jj, you instead start by creating a commit. In this case, it's empty. Then, every time you run a jj command, it does a snapshot, and this produces a new commit. However, because we want to have a stable identifier for a commit, we introduce a new one: the change id. This is basically an alias for the latest commit snapshot that was made. Your git history is just made up of each of these latest commits for each change.
... does that make any sense? Obviously there's more to all of the features than this, but that's sort of the core way that the histories map to each other.
I was in the same position. Then I tried jj. I knew within the day that I wouldn't switch back.
It's not a trick or workaround. It's a very straightforward use of a command flag on one of the most used git commands. It's even conceptually very simple, you're just rebasing a subset of commits of a branch onto a different parent. Anyone who has ever rebased already has a working mental model of this operation. Framing this issue where knowing a single flag on a command every git user uses every day to perform an operation they already broadly understand as some arcane knowledge and 'nonsense' is ridiculous.
> introduces some mental friction in walking away???
I don't think it's just mental friction. Suppose you've learned git well enough that everything you do in it is automatic and fast, and the things which aren't fast by default you've built aliases and tooling for over the years. Yes, starting from ground zero you might want something like jj, but at the current point in your life you're not starting from ground zero. Switching to jj means learning another tool to achieve similar outcomes on your workflows.
If jj took many weeks of relearning, I might be right there with you. But the overwhelming majority of people I’ve personally seen who try the switch convert within a day, are barely slowed down by day two, and are effectively fluent within three days to a week at most.
It really just depends. I was very comfortable with the git cli. It didn't take long to learn jj's, and I'm faster with it now than I ever was with git, simply because a lot of things are easier to do and take less commands.
But with jj there are better workflows that aren’t really doable with git.
13 replies →
I don't think it's pride, more that I don't see the problem that it solves for me. I don't find that git gets in my way at all, and I don't find it confusing. It's a pretty transparent tool that I use every day and hardly notice.
I don't mind other people using jj, but I simply don't feel a need to try it. There's nothing prideful about that, it's just pragmatism.
Pragmatic is using the best tool for the job. Certainly subjectively, and arguably objectively, jj is the better tool in many ways.
My advice is to try it. You should like it, if you're in the majority of folks to give it a sincere shot. If you don't like it, cool. But then my advice changes. jj is probably the future. Adapt or become one of those old-timers who froze their learning in time.
2 replies →
“why not just use jj?”
The most common reason is that the git user has no idea it exists. I am in this bucket alongside 99% of git users.
There are an assortment of other potential reasons:
- It is a new tool that hasn't been battle tested as much as git has, which can decrease confidence.
- Git has inertia. People have learned it, it takes less effort to add a new git skill to your repertoire than learn a new tool from scratch, even if the new tool is easy to pick up.
- Due to its novelty, the auxiliary tooling ecosystem for jj is smaller (does it have plugins for all the popular editors? Lots of people like those, git's are high quality)
- Git is good enough. It's not perfect, but its popularity means that its shortcomings have readily available fixes or tweaks from users. It simply isn't bad enough and there are bountiful resources on how to use it effectively.
I think a core reason (besides not knowing jj exists), is the framing that there is a choice that has to be made, or a switch that has to occur. It is, instead, additive. I have Sublime Merge (GUI git client) and jj both looking at my git repo all day. Zed's git stuff is watching it too.
jj is sort of a bag of git tricks for me that I use when needed. It's no different than some things being easier with the git CLI vs others being easier in Sublime. I'll be at a stage where my committing/branching/rearranging wants are something that jj nails perfectly, and I do those there. As far at the other tools are concerned, I just did a bunch of sophisticated git operations.
The "colocated with git" capability of jj is probably it's most amazing feature tbh, and is key to any sort of adoption.
> why not just use jj
1. It's very new; I haven't had time to learn it properly yet.
2. It's very new and tooling doesn't support it well, e.g. VSCode. There aren't many GUIs yet.
3. I tried it once co-locating with Git and you definitely can't use both at the same time, even if it can use a `.git` directory. It ended up in a huge mess.
I'm definitely in favour of better-than-Git alternatives but I don't think it's reasonable to expect everyone to switch to JJ right now. It isn't so much better that abandoning the de facto standard is obviously worth it yet. (In contrast to things like the iPhone, Rust, SSDs, etc.).
Also I really wish they would focus on some of the bigger pain points of Git. I can deal with rebasing and whatnot. Sometimes it's painful but it's usually not that bad.
What I can't deal with are submodules and LFS. Both are awful and fixing them properly requires fundamental changes to the VCS which aren't going to happen in Git. JJ has an opportunity to do that. Imagine if JJ could say "we have submodules but they aren't awful!" or "you can check in large files!". Those are the sort of huge advantages that would mean you can say "why not just use JJ".
Colocating is the default now, and you should be able to use both at the same time, though running git commands that mutate can confuse jj. Ideally you only use read-only git commands.
Both submodules and LFS are things that jj wants to address, but they take time.
`jj` isn't new. It's definitely not *very* new.
There is a vscode jj gui extension
For me, the answer is stgit. https://stacked-git.github.io/
Oh, there's another stgit user! ^5 Coming from darcs, I couldn't use git until stgit came along, and today, it's one of those few tools I can't imagine working without. Nothing else matches my way of code hacking. So often, I watch people making a big mess with git, and I always recommend stgit to them, so they can post proper and reviewable branches for merging. But in all these years, I could never convince anybody.
That's really neat.
It’s a bit like qwertz. Sure, it is not optimal, there are better alternatives available. But it is good enough, and it is universal. That trumps a 5% typing improvement on my own custom keyboard layout at the cost of not being able to use my coworkers keyboard.
Also, I dislike all of the alternate git frontends I tried, because they are opinionated in a way they clash with my workflow.
Moreover, I don’t think the git CLI is that bad. Once you learn some basic concepts, it makes a lot of sense and is pretty consistent.
Most problems people report stem from a refusal to learn the underlying structure and models. That is on them. And when using a different frontend, they don’t disappear either. They are just abstracted, to allow you to avoid learning them. But they are still there, and you will probably still need to know them at some point.
> Most problems people report stem from a refusal to learn the underlying structure and models.
It's very easy to fall into the trap of believing this: git's implementation fits together neatly enough that it feels like the best you could do. Like, yes it's complex, but surely that's just intrinsic complexity of the problem? (Also, I think we all sometimes feel like someone with a different view must just not know as much as us.)
But if you have used other version control systems (I'm thinking particularly Mercurial here) you realise that actually some of that complexity is just totally made up by git.
1 reply →
Jujutsu is not a Git frontend, it is an independent VCS that is compatible with Git. This is important because it has concepts and ideas of its own. It is used entirely independent of Git within Google.
jj does not support git submodules, this precludes even a casual use on my own personal repo.
This is one I'm okay with. Git submodules are absolutely awful.
I recognize that some folks are stuck with it, and yeah, in your case jj probably just doesn't make sense. Stay tuned though.
It just means that you use git commands to update your submodules, jj still works for the rest of the repo just fine.
2 replies →
I've never heard of jj before this thread...
For me the answer is lazygit. I rarely use the git cli. I don't want to learn the jj cli and the TUI wrappers for jj seem less polished.
jjui is great, give it a try!
i went from being a "jj cli power user" to relying on jjui for all of my complex rebase needs so quickly that i now have to read the man page to recall basic commands
1 reply →
This particular tip is basic. It just describes how to copy changes from one branch to another with one kind of rebase. But there's always someone who is learning git for the first time and may not even know how it's used.
Git is simple enough and has features and capabilities that jj does not have. Contrary to popular belief, git is not hard to use. I refuse to use any "simpler" system that is slower or less feature-rich than git. I don't even want to learn another commit graph model, because git's model is very good. About 95% of what people like yourself call "git nonsense" consists of useful features that many people would be annoyed to not have.
I believe that a large number of git or general VCS users have no idea about commit hygiene. They have not had to cherry-pick or edit commits, and have no idea what to do about conflicts. To people like that, git's features and methods will appear especially foreign.
I looked over jj specifically many moons ago and concluded it would annoy me and not function at my job. I forgot what the reasons were. One reason was most likely because I need submodules and worktrees to work. I just looked at its FAQ, and saw a bunch of nonsensical new terms as well. Nothing is more compatible with git than git itself, and I am very satisfied with how well git works for me.
submodules aren't native with jj, you use git commands to manage them, and it works fine. Eventually there'll be some sort of native support.
worktrees are called "workspaces" in jj, but are the same.
A lot of people find jj easier to have good commit hygiene with, and find it simpler and more powerful, not less. But that said, if you're happy with git, you should continue to use it.
20 replies →
> I don't even want to learn another commit graph model, because git's model is very good.
I agree. That's why jj uses practically the same model. That's how Git can quite easily be used as a backend.
> I just looked at its FAQ, and saw a bunch of nonsensical new terms as well.
Like what? Perhaps we can improve it.
14 replies →
There is also mental friction with learning an entirely new tool. `jj` is different enough from `git` that one can't transfer knowledge. Currently the documentation is not good enough to assuage that issue.
This really isn’t true. All of my git knowledge—except for the CLI flags—is directly useful for jj.
The jj CLI is very easy to grok, even for a seasoned git user. Maybe even especially so for a seasoned git user.
Frankly, this is a rather poor, uninformed excuse. Most people who try it grok the concepts in days. And they feel right at home in weeks. That's when you start hearing, "I'll never go back."
Wish i could remember my issues with jj. I tried it, i wanted to stick with it because i loved the fact that i could reorder commits while deferring the actual conflicts.. but something eventually prevented me from switching. Searching my slack history where i talked about this with a coworker who actually used jj:
1. I had quite a bit of trouble figuring out a workflow for branches. Since my companies unit of work is the branch, with specifically named branches, my `jj ls` was confusing as hell.
`jj st` might have helped a bit, but there were scenarios where creating an commit would abandon the branch... if i'm reading my post history correctly. My coworker who was more familiar explained my jj problems away with "definitely pre-release software", so at the time neither of us were aware of a workflow which considered branches more core.
Fwiw, I don't even remember when the jj workflow had branches come into play.. but i was not happy with the UX around branches.
2. iirc i didn't like how it auto stashed/committed things. I found random `dbg!` statements could slip in more easily and i had to be on guard about what is committed, since everything just auto pushed. My normal workflow has me purposefully stashing chunks when i'm satisfied with them, and i use that as the visual metric. That felt less solid with jj.
Please take this with a huge grain of salt, this is 10 month old memory i scavenged from slack history. Plus as my coworker was saying, jj was changing a lot.. so maybe my issues are less relevant now? Or just flat out wrong, but nonetheless i bounced off of jj despite wanting to stick with it.
"creating a commit would abandon the branch" is certainly something lost in translation. There are other reasons you may have not liked the UX, largely that if you create branches and then add a bunch of commits after it, the branch head doesn't automatically move by default. There is a config setting you can change if you prefer that, or the `jj tug` alias some people set up.
Auto-commit is still a thing, but you can regain the stuff you like with a workflow change, this is called the "squash workflow" and is very popular: https://steveklabnik.github.io/jujutsu-tutorial/real-world-w...
I more or less use the method described [here](https://steveklabnik.github.io/jujutsu-tutorial/advanced/sim...) for branches. One thing I do change is that I set the bookmark to an empty commit that serves as the head of each branch. When I am satisfied with a commit on head and want to move it to a branch I just `jj rebase -r @ -B branch`. When I want to create a new branch it's just `jj new -A main -B head` and `jj bookmark set branch_name -r @`
4 replies →
Can you explain how to do same workflow in jj? It would be nice to compare bads and goodies of git/jj
This part:
> You rebase feature-1 onto main:
feature2 moves as well. Basically, the behavior here is just the default.
Google is definitely your friend here. Plenty of tutorials online.
I have one and a half decades of muscle memory burned in with inoremap jj <Esc>`^
It's not something I can just shift away from.
Definitely let your keybinding keep you from trying out and using a fantastic tool.
1 reply →
`jj` doesn’t allow temporary files in the repository that aren’t listed in `.gitignore`.
This makes it unusable for me.
There's multiple ways of handling this, depending on what you prefer.
Keeping these things in a private commit is one option.
Configuring auto-tracking to be off is another option.
It does via snapshot.auto-track https://docs.jj-vcs.dev/latest/config/#paths-to-automaticall...
Jujutsu comes in handy here for the same usecase:
https://github.com/jj-vcs/jj
https://www.stavros.io/posts/switch-to-jujutsu-already-a-tut...
Also found https://github.com/gitbutlerapp/gitbutler
Half my team switched to JJ this year, and I do find stacking PRs to be much more pleasant now. We had previously tried out Graphite but it didn't really stick.
I wrote up a little way to use JJ's revsets to make it easy to push an entire stack of branches in one command:
https://crabmusket.net/2025/jj-bough-a-useful-alias-for-stac...
The main issue I kept having when trying to do this with just git is then managing all the branch names to be attached to the right moved commits, so that my stack could be reviewable on github's open PRs.
Does jj help with that at all?
I've experimented a bit with git-town.com (OSS) and now everyone at $DAYJOB uses graphite.com (SaaS) which does that part very well.
It’s one of the core features that rebases, including branch names (bookmarks in jj) work ‘correctly’. You can rebase whole dags, including merges, with multiple named heads with just one jj rebase -b.
8 replies →
`--update-refs` flag helps a lot in vanilla git. That and `--autosquash` should probably be default flags to `git rebase`. I also don't entirely trust rebase without `-i` (`--interactive`), personally. I hear there is talk about shaking up the out-of-the-box default flags in git 3, and I think rebase should especially get new defaults.
1 reply →
That particular case can be solved much easier by rebasing outer-most branch with `--update-refs` flag.
I came into the comments specifically to ask if this flag existed. I feel bad that the author developed this whole flow just because they didn't know about this, but that's pretty common with git.
I'm pretty sure the author was Claude, so don't feel too bad for it.
Thanks. This is going to be so useful, but it pains me to know I could have been using —update-refs for the last three years.
I used to dutifully read release notes for every git release, but stopped at some point. Apparently that point was more than three years ago.
discoverability is a big problem, especially for CLI tools which can't afford to show small hints or "what's new" popups. I myself learned it from someone else, not docs.
2 replies →
Exactly, I was reading the blog and wondering the whole time how it's better than --update-refs, which I have been using a lot recently.
Yep. I set this in .gitconfig
I'm guilty lol. I wrote a helper to do rebase chains like this
update-refs works only in a narrow case when every branch starts form the tip of a previous. Your helper might still be useful if it properly "replants" whole tree keeping its structure.
2 replies →
I think ‘git rebase —-update-refs’ is the better way to go for this scenario
Sweet, looks like this is pretty new (2022).
Running a git command on one branch and multiple branches being affected is really unusual for me! This really does look like it is designed for just this problem, though. Simple overview: https://blog.hot-coffee.dev/en/blog/git_update_refs/
It breaks if you amend the top commit instead of adding a new one.
Is there any good guide on how to solve the issue which OP solves?
I was reading this the other day when I came across this feature because I’m stacking PRs recently which I don’t usually do
https://andrewlock.net/working-with-stacked-branches-in-git-...
Another commenter posted this link which was a bit more succinct
https://blog.hot-coffee.dev/en/blog/git_update_refs/
There isn’t much to it though, you just go to the branch and run git rebase with the update refs flag.
You don’t really need docs as --update-refs does what the OP does automatically instead of manually like the OP does.
4 replies →
I've settled on a workflow that reverses the situation. I simply commit all my work to the main branch and cherry pick commits into temporary feature branches only when submitting PRs.
This way I only need to worry about maintaining a single consistent lineage of commits. I've been using this workflow for about a year now and find it to be much easier than juggling and rebasing feature branches.
In case anyone's interested, I made a tool that automates this workflow. The worfklow and tool are described here: https://github.com/bjvanderweij/dflock/
You might like Jujutsu – you commit without being on any branch and then later you can decide where and how you put things onto branches.
Interesting, I'll be sure to check it out. It sounds pretty similar to the tool I built which lets you edit a "plan" in a text editor to assign commits to feature branches - the plan is saved so it can be amended continuously.
How many people on the team?
I work in a small team, about five people.
GitButler handles all of this pretty automatically, if you don't want to deal with the Git gymnastics needed here.
https://blog.gitbutler.com/stacked-branches-with-gitbutler
I usually just `git rebase origin/main -i` after the base branch has been merged there, and this means I need to explicitly drop the merged commits, but I can inspect what's happening.
Yeah, I do this too: The `--onto` solution feels a bit too magical at times and an interactive rebase is pretty clear about what's happening.
Add `--update-refs` to your interactive rebase and it will give you an easy line to know how many commits to drop because it will add an `update-ref` line for the old branch. You can just easily delete everything up to and including that `update-ref` line and don't have to manually pull up a git log of the other branch to remember which commits already merged.
(Plus, of course, if you have multiple branches stacked, `--update-refs` makes it easier to update all of them if you start from the outermost branch.)
2 replies →
I'm a heavy user of git-spice: https://abhinav.github.io/git-spice (created by a former coworker) and can't really go back to a time without it. While still not nearly as good as Facebook's Phabricator, it's probably the best workflow for small, focused stacked PRs you can achieve in a Github / Gitlab based repository.
Even if `--update-refs` didn't exist, my experience is that git can identify duplicate commits produced by rebase, and knows to skip them when rebasing the same commits to the same place again. Am I imagining that?
That works until you had to fix conflicts during the rebase and the commits are no longer identical.
It definitely fast-forwards unchanged commits.
What I would really like is a Git equivalent to Mercurial's "fold" operation. I usually make a bunch of commits as I work, just as checkpoints, which I then want to turn into a single final commit when it's done, which could be quite some time later, e.g. "started on thing", "broke it", "broke it more", "Add thing to improve foo of bar".
Mercurial's "histedit" command offers two operations to combine the commits: "fold" and "roll", where "fold" brings the earlier commit into the later commit and "roll" does the reverse. The "fold" operation does exactly what I want in this case: It brings all the changes into the final commit from when I actually finished it, using the commit date of that commit.
Git's "rebase -i" command offers "squash" and "fixup" and "fixup -c" and "fixup -C", but they all do the same thing as "roll", i.e. they keep the earliest commit's date, the date when I started working on the thing and not the date when I finished working on it. (So I then cut and paste the correct date and update the commit afterwards. This may not be the best way to do it.)
That is an interesting desire to use a later commit date rather than an earlier one. So many prefer that commit date of when an effort started.
One way to accomplish that is to reorder the commmits in `rebase -i`: "pick" the last commit first and squash/fixup the rest after. That can produce some very weird merge conflicts, but it works well more often than you might think, too.
At the very least, you can do your hand editing of the commits during the rebase instead of after by switching the earliest commit from "pick" to "edit" to have the full power to amend the commit before it moves on (with `git rebase --continue` when you are satisfied). (Versus "reword" if you just want to change the commit message only.)
Also, instead of naming commits things like "broke it" and "broke it more" an option is to use `git commit --fixup={previous commit hash}`. That auto-generates a "fixup!" name based on the previous commit and auto-marks it as a fixup if you use the `--autosquash` flag to rebase.
I do use fixup commits as well, for fixing small issues discovered after I make the proper commit, and in that case it makes sense to use the date and message of the earlier commit.
But in the case I'm describing, the earlier commits are essentially just temporary snapshots on the way to making a proper commit. Just a more explicit undo history, basically. I usually don't even bother naming them anything as meaningful as "broke it", actually. Maybe most people wouldn't even bother making these commits – or maybe they would if Git supported "fold".
4 replies →
If you ever try out jj, both fold and roll exist as `jj squash`. You'd choose as the destination the change of the time you'd want to keep.
You could use "git reset --soft oldest-commit" and then "git commit -C newest-commit".
Even for 'normal' rebases of a multi-commit series (without named feature branches), I habitually use the --onto form. It is simply easier to conceptualise what is happening if one is explicit about the 3 references.
As such, using it for the situation described in the piece then becomes a trivial matter, especially if one also habitually runs a graph viewer - I generally have one or more instances of gitk running all the time.
This made good sense, though by the time I got to bits saying "that last step is critical. Without it, your next sync will break" and "Don't forget!" I was laughing out loud. I love git :-)
Stacked commits is awesome, but it sucks that with git you need all these workarounds, and that the servers (GitHub etc) are not designed with that workflow in mind.
I left Google a few months back to work on another project. I missed the internal version control so much that I dropped that other project and am now focused http://twigg.vc It's very similar to what's used at Meta and Google. Atomic commits, linear history, auto rebase, code review compatible with stacked commits.
I feel that stacked PR in git should be common knowledge, however many of the documentations are scattered. I found stacked.dev but feels that its not exploring pure-git workflow that well. thats why I tried to collect all of those docs into a single website that's dedicated into that singular workflow. would love to get any feedback:
https://stacked-pr.github.io
(disclaimer: partly supported by AI, but most of the writing structure were made by hand)
If you adopt a good continuois integration process you don't need stacked commits at all, which are more a symptom of a process issue, IMHO.
It's fashionable at the moment so there is a bit of herd mentality going on. In 2 years they'll be a reinvention of the wheel again against it.
Fun stuff, but I'll stick to trunk based dev, small PRs... thanks!
Stacking commits lets you do that without having to wait for each change to be reviewed/merged to the main branch before you iterate on top of those changes.
True. I find I rarely need it: standard rebase or merge do the trick. If they don't the review cycle or PR size may be too high. Super rare I need onto. So rare I look up how to do it when I do.
It's such a complicated way to work though, you start another set of changes then you go back addressing comments then you go back updating the stacked branch and you might need to do that few times... Teams should focus on getting stuff merged in and not create massive PRs that live forever, life becomes so much easier.
I‘m using git machete since a few months to help with the stacking workflow. The nice thing is that it can use the GitHub CLI to manage the resulting PR stacks as well.
https://github.com/VirtusLab/git-machete
I fail to see how it makes it much easier to review a lot of interdependent branches where branch C needs new logic from branch B which needs commits from branch A. Sounds like a nightmare to me.
People should really try to work on trunk.
Instead of "stacked diffs", isn't the more "continuous integration" solution to split a big feature into small chunks that actually get merged?
Having to rebase again and again is a symptom that a dev branch is living for too long.
This problem isn't specific to cases where you are rebasing "again and again". Just needing to do a single rebase (e.g. prior to opening a PR) for a stacked feature is enough to need a solution.
And, as others have pointed out, the modern solution is `--update-refs`, so there's no need for complicated workflows any more anyway.
If you mean rebasing as each PR in the stack is merged, most Git platforms have the ability to do that automatically.
No, I mean small chunks that are actually merged instead of having a stack of them floating around.
A rebase before a merge can always happen, of course. But there are not stacked commits then it is just a standard rebase of your small chunk and that's it. And this will also have a shorter life than a stack so rebases will be rarer and simpler.
2 replies →
I’m amazed that this comment is so low down
Stacked diffs seems like a solution to managing high WIP - but the best solution to high WIP is always to lower WIP
Absolutely everything gets easier when you lower your work in progress.
This seems idealistic. It's very normal to be working on a feature that depends on a not-yet-merged feature.
3 replies →
I believe the author would love stg: https://stacked-git.github.io/guides/tutorial/#patches
For the example given, would merging branch 2 into branch 1 then branch 1 into main achieve the same effect?
Perhaps not an option if you need to release the work in branch 1 before the work in branch 2 is ready/reviewed/etc.
The point of this technique is to keep them separate. See the other comments about `--update-refs`.
This marker branch step feels like a workaround to a missing capability. It's something I can easily see one forgetting especially if they haven't been doing stacked diff workflows regularly.
The capability is there.
Just use git rebase --update-refs ...
Wow you aren't wrong, the first blog post on Google talking about this is exactly what this complicated method does just built-in.
1 reply →
I agree it seems error prone. I'm not sure if I'm misunderstanding something, but I use `git cherry-pick` when I know I need to move commits around that might have conflicts. The problem with rebase can be that the user doesn't fully understand all the options being applied and end up with a "bad" merge.
I don't usually want to rewrite history. I just want the target branch with all my commits on top (I usually squash the feature branch into one commit anyway). I have yet to run into a situation where this isn't good enough.
If the branch diverges so much and has so many commits that this simpler approach doesn't work, that might not be a git problem, but a project management one. It's still always nice to know git has tools to get me out of a jam.
Rebase is just automated cherry-pick, so it ends being the same. The pick command in rebase is exactly that.
> and end up with a "bad" merge.
They end up with exactly the same merge when using cherry-pick directly?
> I don't usually want to rewrite history. I just want the target branch with all my commits on top
That's ... what rewriting history is?
I’ve been meaning to write this article for a long time @flexdinesh . Thanks for taking the time to share this technique for managing stacked diffs using vanilla git rebase!
Hard part is keeping the feature branches orthogonal and maintaing all the -base branches. Every fixup commit would invalidate them.
I consider myself a shmedium experienced dev who likes to learn their tools and read source.
This seems like a house of cards whose juice isn’t worth the squeeze. But I would love to split up my PRs into smaller pieces.
I just would hate to waste time with an incomprehensibly goofed git history because I forgot a command.
We use graphite at work to automate this workflow. The whole point is avoid this toil.
Ah, I've been doing this for ages but apparently this practice has a name
Eh I just use `git rebase -i` and delete the commits I don't want. Much easier to think about.
But the real problem with this workflow is that neither Github nor Gitlab support it at all. Not even Forgejo does. Which blows my mind because is such an obvious way to work.
As far as I know only Tangled actually supports it, but unfortunately Tangled is tangled up in its own weird federation ideas. There's no way to privately host it at all.
I’m not sure what you mean, I stack PRs all the time with GitHub - select the earlier branch as the merge target instead of the main branch.
That doesn't work for cross-repo PRs which is by far the most common way of creating PRs on GitHub (at least for open source stuff).
Is there any reason besides merge commits ending up in history to not do this with merges instead? ie merge main into feature-1, then feature-1 into feature-2.
Sounds like using --update-refs would let you do all that in a single operation, but you still need to force-push and don't maintain an explicit merge/conflict resolution history, both of which could be considered sub-optimal for collaborative scenarios.
The use case is that they are not ready to merge yet.
They meant merging the other way, i.e. merging the new changes from main into the stacked feature branches.
This is functionally the same as rebasing, except that the new changes show up at the tip of the commit chain rather than the base. And because it doesn't rewrite history you don't need to force-push.