drewdevault.com

dotfiles.md (3725B)

---
date: 2019-12-30
layout: post
title: Managing my dotfiles as a git repository
---
There are many tools for managing your dotfiles - user-specific configuration
files. GNU stow is an example. I've tried a few solutions over the years, but I
settled on a very simple system several years ago which has served me very well
in the time since: my $HOME is a git repository. [This
repository](https://git.sr.ht/~sircmpwn/dotfiles), in fact. This isn't an
original idea, but I'm not sure where I first heard it from either, and I've
extended upon it somewhat since.
The key to making this work well is my one-byte `.gitignore` file:
```
*
```
With this line, and git will ignore all of the files in my $HOME directory, so I
needn't worry about leaving personal files, music, videos, other git
repositories, and so on, in my public dotfiles repo. But, in order to track
anything at all, we need to override the gitignore file on a case-by-case basis
with `git add -f`, or `--force`. To add my vimrc, I used the following command:
```
git add -f .vimrc
```
Then I can commit and push normally, and .vimrc is tracked by git. The gitignore
file does not apply to any files which are already being tracked by git, so any
future changes to my vimrc show up in git status, git diff, etc, and can be
easilly committed with `git commit -a`, or added to the staging area normally
with `git add` — using `-f` is no longer necessary. Setting up a new
machine is quite easy. After the installation, I run the following commands:
```
cd ~
git init
git remote add origin git@git.sr.ht:~sircmpwn/dotfiles
git fetch
git checkout -f master
```
A quick log-out and back in and I feel right at $HOME. Additionally, I have
configured $HOME as a prefix, so that ~/bin is full of binaries, ~/lib has
libraries, and so on; though I continue to use ~/.config rather than ~/etc. I
put $HOME/bin ahead of anything else in my path, which allows me to shadow
system programs with wrapper scripts as necessary. For example, ~/bin/xdg-open
is as follows:
```sh
#!/bin/sh
case "${1%%:*}" in
	http|https|*.pdf)
		exec qutebrowser "$1"
		;;
	mailto)
		exec aerc "$1"
		;;
	*)
		exec /usr/bin/xdg-open "$@"
		;;
esac
```
Replacing the needlessly annoying-to-customize xdg-open with one that just
does what I want, falling back to /usr/bin/xdg-open if necessary. Many other
non-shadowed scripts and programs are found in ~/bin as well.
However, not all of my computers are configured equally. Some run different
Linux (or non-Linux) distributions, or have different concerns being desktops,
servers, laptops, phones, etc. It's often useful for this reason to be able to
customize my configuration for each host. For example, before $HOME/bin in my
$PATH, I have $HOME/bin/$(hostname). I also run several machines on
different architectures, so I include $HOME/bin/$(uname -m)[^1] as well. To
customize my sway configuration to consider the different device configurations
of each host, I use the following directive in ~/.config/sway/config:
```
include ~/.config/sway/`hostname`
```
Then I have a host-specific configuration there, also tracked by git so I can
conveniently update one machine from another. I take a similar approach to
per-host configuration for many other pieces of software I use.
Rather than using (and learning) any specialized tools, I find my needs quite
adequately satisfied by a simple composition of several Unix primitives with a
tool I'm already very familiar with: git. Version controlling your configuration
files is a desirable trait even with other systems, so why not ditch the
middleman?
[^1]: `uname -m` prints out the system architecture. Try it for yourself, I bet it'll read "x86_64" or maybe "aarch64".