I must've tried to set up stow five or six times over the years, in between various hand rolled custom setups. I can't put my finger on why but I set up chez moi & it's been my setup since, much longer than any previous solution.
Chez moi is definitely not without its rough edges but it seems to have gotten the subtle essentials right enough for adhd me to not have abandoned it yet.
I always like to mention rcm. It's not a popular one but I tried probably 10 managers before it and much prefer it to the competition: https://github.com/thoughtbot/rcm
It's great to manage your dotfiles, but I took it a step farther. I rebuilt the minimal Linux desktop environment of my dreams (startx, xinit, i3, i3status etc) with Ansible. It begins from a vanilla Ubuntu server 24.04.4 install. I bootstrapped it using a KVM + spice setup (using a spare physical SSD rather than a virtual one) and iterating over and over again until I finally got everything mostly working. I then booted off that physical disk, and kept iterating until everything was perfect. I've since adapted the setup to work on virtual aarch64 on macOS. I just recently tuned it to work on a crappy old Haswell Dell laptop, now properly detecting and configuring hardware vaapi capabilities, backlight, battery, trackpad, trackpoint, etc.
Pretty snazzy, watching YouTube in Firefox on a 13 year old laptop with hardware h264 decode and everything tuned exactly to my liking.
I’ve always managed this problem in a different way. I don’t know if my way is better, but it works really well for me.
I treat my powerful desktop computer as my main machine. Then I have a bunch of laptops.
Then I just rsync my entire home directory out to all the laptops.
From there. The rule is quite simple. Any file created on a laptop are considered ephemeral. If I create data that I have to keep. It gets rsynced back the other direction to the main machine.
This process has served me well for at least 15 years now and is supported by a small handful of shell scripts to automate this process
I like to use environment modules [1] (or lmod [2]) for that purpose. You can make each manually built software package available or not on a per-shell-session basis, just by running
module load myapp
or
module unload myapp
in the shell where you want it (or don't). The small downside is that it only works with software that actually uses the standard environment variables like PATH, CPATH, etc. for their intended purposes rather than hardcoding filesystem paths, but in my experience it's rare to find something that doesn't. Also, you have to write a modulefile for each package, but that's not a big deal.
I hate . dirs. In fact, I hate them so much that I don't use them.
My configuration lives primarily in .yml files. These are kept
super-simple. When need be and another format is required, ruby
autogenerates these for me. For instance, all my bash aliases
are kept in .yml files which then get turned into bash rc files
or any other target format for other shells. Same for most of my
other configuration too - not always .yml but usually some text
file. I never understood the neet for .foobar directories or files.
They just hide a system that is intrinsically ugly and needlessly
complicated.
Chezmoi strikes a nice balance between the overkill of home-manager while still being more powerful than simpler solutions.
Yadm is another alternative, the main thing I don't like about it though is that I'm not a fan of cross OS dotfiles. Having niri files on my work Mac and aerospace dotfiles on Linux annoys me quite a bit.
As powerful as the templating in chezmoi is, I think it should be considered a last resort and only used for simple files. They break your editor features like highlighting.
Once you hit the Chezmoi stage, you're only about 6 months from Nix and Home Manager. I mean, why climb _almost_ to the top of a mountain and then just sit down?
Switched to Chezmoi from random assortment of manually authored scripts. The workflow takes some getting used to, because I constantly edit the actual files without calling `chezmoi edit` first, and have to merge.
I like that when combined with `mise` (https://mise.jdx.dev) I can roll out a new computer in 2-3 commands and have my entire environment configured the way I like, with neovim and all the plugins and language servers.
I never did get used to the Chezmoi workflow. `chezmoi edit` never became muscle memory, and I was constant finding myself resolving diffs and what not.
It’s quite new but I’ve been cooking up some new bootstrapping features with mise which people may find relevant here: https://mise.jdx.dev/bootstrap.html
It’s for things like dotfiles, apt/brew packages, and LaunchAgents/systemd.
EDIT: I feel a little bad having hijacked this, as someone that hears a lot of opinions about devtools I can definitely say chezmoi is a darling of the community and I highly recommend checking it out.
I saw it the other day in the release notes and I'm definitely setting this up since I already use mise for many of my global tools and projects.
Some months ago I ported part of my git bare repo to chezmoi but never really picked it up since, even if I see it as a great and very complete tool for managing complex dotfiles, I don't have many machines and can live by with a bunch of if uname -s, and the fact that I still need to wrangle brewfiles and scripts for packages. Having the sytem package managers glue along the configs and symlinks is exactly what I wanted.
I learned about Stow after I found out about Chezmoi, and felt like Chezmoi was the better fit for me. I make heavy use of templating to keep work / personal aliases and functions separate, and could not be happier with the outcome.
I use syncthing to automatically sync my dotfiles git directory across PC/laptops, and stow to manually update symlinks when I add a new dotfile (the content of existing dotfiles is synced by syncthing already)
That way I don't have to remember to commit+push+pull changes to existing dotfiles (like bashrc or vimrc which I edit often) to sync them to other machines, it happens automatically in almost real time as soon as the file is saved on one of my machines (syncthing uses inotify to detect changes on monitored directories)
This is what I did too. My dotfiles used a custom install process that didn't really handle ~/.config very well, so I switched to stow, and then added syncthing because the push + pull dance got too annoying. Really happy with it.
I've never really understood why people get so fancy with their setups when you can just plop a git repo into your home directory. I suspect it has to do with people being unfamiliar with git.
Ok, I came into this thread intending to say “I’ve been using stow for years and am perfectly happy with it”, started RTFA and the comments, realised that I was actually not happy with it, started considering chezmoi then remembered that I had had a pretty great experience building a Nix VM recently.
Now I want to use Nix* to manage my multi-machine MacOS and Linux setup (with lots of dotfile config overlap, of course).
That’s the HN experience for you.
Kind souls: what is currently the blessed way to manage MacOS dots with Nix? I recall there is more than one paradigm - what’s the approach that simplest, most robust and can be adopted incrementally?
Edit: Just to say that I think Atuin now also plays in this space. Haven’t checked it out, though.
> The trouble is that symlinks cut both ways. Every edit on every machine writes straight through the link into that machine’s clone of the repo
I have stow too in my micro DE, use it across a few machines and it's holding up really well. I designed my dotfiles so that changes would happens exclusively to files not tracked by stow. I have .zshrc tracked, but environment goes to .zshenv and general local customisation goes to .local/lib/zsh/overrides.zsh (https://gitlab.com/gabriel.chamon/archie/-/blob/main/deploym...). Hyprland config is tracked but device specific configs live in .config/hype/config/device.conf (https://gitlab.com/gabriel.chamon/archie/-/blob/main/deploym...). Deployment instructions/automation is to just copy dist folders in place, which aren't tracked by stow.
It depends on everyone's style. For me personally changes I don't remember having made to files I track in the repo are symptoms that I might need to review how I organize my files. Maybe those changes were something that broke and I forgot to commit after fixing. In any case for me this bidirectional nature of symlinks are a feature, not a bug in this management system.
I've checked out stow, chezmoi, yadm, and others over the years, but I originally started off by rolling my own dotfiles setup with a Git repo about 6 years ago: https://github.com/jaminthorns/environment
I don't really recommend it to others, since there's all these great tools that have the features you need (per-machine config, secrets, templating), but I get a deep satisfaction from the fact that I understand every part of this setup from top to bottom. It only has the functionality I need, and I know it doesn't depend on anything that might become unmaintained since it's just POSIX shell scripts.
Even still, I might eventually make the jump to something like chezmoi or nix if I'm not able to implement something I need easily, but that hasn't happened yet.
> The trouble is that symlinks cut both ways. Every edit on every machine writes straight through the link into that machine’s clone of the repo.
I find this is a key feature. If a file is edited, git shows it as dirty, and I get to decide if I discard of commit the change. No extra steps required.
> By the time Homebrew and a couple of tools have run on a new Mac, files like ~/.zprofile and ~/.gitconfig already exist.
I don't get why you'd manually provision those files instead of just putting them in dotfiles.
---
Personally, I found that most tools in this space tried to do too much or were too complex. I previously wrote a minimal one in Rust, but eventually re-wrote it into less than 200 lines of shell, which works pretty much anywhere, without having to install anything at all, and is part of the dotfiles repository itself:
I evaluated most of the popular dotfile management tools and found them far too complex for the problems they try to solve. I then used stow for about a year just to see whether symlinks would cause any issues. They did.
Eventually, I wrote my own bidirectional sync tool that synchronizes a git checkout (for example, ~/.dotfiles) with the actual home directory. I run it as a systemd service, so I can simply edit or delete files in ~/.dotfiles, and everything else happens automatically.
I can add or remove files in the repository, and they will automatically be created or removed in their synchronized locations. At the same time, the target directories can contain other unrelated files, including an entire $HOME, and this does not interfere with synchronization in any way.
The bidirectional sync is particularly useful when a program, such as vs code, decides to modify its own configuration files, or when I'm feeling lazy and just want to edit ~/.bashrc directly and commit the changes later.
Achieving this requires some careful logic that, hopefully, I've managed to get right. At least at the time I wrote it, I hadn't seen these implemented in other systems.
Oh, and I use the same sync for some files I need to /etc, only for that I run sudo scripts and dont' have a always-on service for obvious reasons.
46 comments
[ 2.5 ms ] story [ 62.3 ms ] threadChez moi is definitely not without its rough edges but it seems to have gotten the subtle essentials right enough for adhd me to not have abandoned it yet.
I think Chezmoi's templates and file naming conventions don't click for me, but it's nice to see a good variety in this problem space.
Pretty snazzy, watching YouTube in Firefox on a 13 year old laptop with hardware h264 decode and everything tuned exactly to my liking.
I treat my powerful desktop computer as my main machine. Then I have a bunch of laptops.
Then I just rsync my entire home directory out to all the laptops.
From there. The rule is quite simple. Any file created on a laptop are considered ephemeral. If I create data that I have to keep. It gets rsynced back the other direction to the main machine.
This process has served me well for at least 15 years now and is supported by a small handful of shell scripts to automate this process
stow is an indispensable tool for me to manage /usr/local for manually installed software. my workflow goes:
now, myapp and all its supporting files are in the right place in /usr/local. if i want to "uninstall", i just run[1]: https://modules.readthedocs.io/en/latest/ [2]: https://lmod.readthedocs.io/en/latest/
My configuration lives primarily in .yml files. These are kept super-simple. When need be and another format is required, ruby autogenerates these for me. For instance, all my bash aliases are kept in .yml files which then get turned into bash rc files or any other target format for other shells. Same for most of my other configuration too - not always .yml but usually some text file. I never understood the neet for .foobar directories or files. They just hide a system that is intrinsically ugly and needlessly complicated.
Yadm is another alternative, the main thing I don't like about it though is that I'm not a fan of cross OS dotfiles. Having niri files on my work Mac and aerospace dotfiles on Linux annoys me quite a bit.
As powerful as the templating in chezmoi is, I think it should be considered a last resort and only used for simple files. They break your editor features like highlighting.
I like that when combined with `mise` (https://mise.jdx.dev) I can roll out a new computer in 2-3 commands and have my entire environment configured the way I like, with neovim and all the plugins and language servers.
It’s for things like dotfiles, apt/brew packages, and LaunchAgents/systemd.
EDIT: I feel a little bad having hijacked this, as someone that hears a lot of opinions about devtools I can definitely say chezmoi is a darling of the community and I highly recommend checking it out.
Some months ago I ported part of my git bare repo to chezmoi but never really picked it up since, even if I see it as a great and very complete tool for managing complex dotfiles, I don't have many machines and can live by with a bunch of if uname -s, and the fact that I still need to wrangle brewfiles and scripts for packages. Having the sytem package managers glue along the configs and symlinks is exactly what I wanted.
Perfect timing, I'm already 80% done with the transition :D
That way I don't have to remember to commit+push+pull changes to existing dotfiles (like bashrc or vimrc which I edit often) to sync them to other machines, it happens automatically in almost real time as soon as the file is saved on one of my machines (syncthing uses inotify to detect changes on monitored directories)
Now I want to use Nix* to manage my multi-machine MacOS and Linux setup (with lots of dotfile config overlap, of course).
That’s the HN experience for you.
Kind souls: what is currently the blessed way to manage MacOS dots with Nix? I recall there is more than one paradigm - what’s the approach that simplest, most robust and can be adopted incrementally?
Edit: Just to say that I think Atuin now also plays in this space. Haven’t checked it out, though.
Now it should be "every app adds features until it manages dotfiles" :D
Dunno why I'd use atuin for dotfile management, but some people seem to like it. It's very good at syncing and managing shell history though.
https://github.com/logannc/janus
I have stow too in my micro DE, use it across a few machines and it's holding up really well. I designed my dotfiles so that changes would happens exclusively to files not tracked by stow. I have .zshrc tracked, but environment goes to .zshenv and general local customisation goes to .local/lib/zsh/overrides.zsh (https://gitlab.com/gabriel.chamon/archie/-/blob/main/deploym...). Hyprland config is tracked but device specific configs live in .config/hype/config/device.conf (https://gitlab.com/gabriel.chamon/archie/-/blob/main/deploym...). Deployment instructions/automation is to just copy dist folders in place, which aren't tracked by stow.
It depends on everyone's style. For me personally changes I don't remember having made to files I track in the repo are symptoms that I might need to review how I organize my files. Maybe those changes were something that broke and I forgot to commit after fixing. In any case for me this bidirectional nature of symlinks are a feature, not a bug in this management system.
I don't really recommend it to others, since there's all these great tools that have the features you need (per-machine config, secrets, templating), but I get a deep satisfaction from the fact that I understand every part of this setup from top to bottom. It only has the functionality I need, and I know it doesn't depend on anything that might become unmaintained since it's just POSIX shell scripts.
Even still, I might eventually make the jump to something like chezmoi or nix if I'm not able to implement something I need easily, but that hasn't happened yet.
I find this is a key feature. If a file is edited, git shows it as dirty, and I get to decide if I discard of commit the change. No extra steps required.
> By the time Homebrew and a couple of tools have run on a new Mac, files like ~/.zprofile and ~/.gitconfig already exist.
I don't get why you'd manually provision those files instead of just putting them in dotfiles.
---
Personally, I found that most tools in this space tried to do too much or were too complex. I previously wrote a minimal one in Rust, but eventually re-wrote it into less than 200 lines of shell, which works pretty much anywhere, without having to install anything at all, and is part of the dotfiles repository itself:
https://git.sr.ht/~whynothugo/dotfiles/tree/ac97cb196f02cafa...
This one is good, though, and very thorough—even though it's Mac-centric.
Eventually, I wrote my own bidirectional sync tool that synchronizes a git checkout (for example, ~/.dotfiles) with the actual home directory. I run it as a systemd service, so I can simply edit or delete files in ~/.dotfiles, and everything else happens automatically.
I can add or remove files in the repository, and they will automatically be created or removed in their synchronized locations. At the same time, the target directories can contain other unrelated files, including an entire $HOME, and this does not interfere with synchronization in any way.
The bidirectional sync is particularly useful when a program, such as vs code, decides to modify its own configuration files, or when I'm feeling lazy and just want to edit ~/.bashrc directly and commit the changes later.
Achieving this requires some careful logic that, hopefully, I've managed to get right. At least at the time I wrote it, I hadn't seen these implemented in other systems.
Oh, and I use the same sync for some files I need to /etc, only for that I run sudo scripts and dont' have a always-on service for obvious reasons.
I'd be happy if someone else found it useful too: https://github.com/senotrusov/etcdotica