Workspace Bootstrapping With Boots
I tend to try out a bunch of different languages and platforms quite often, and I usually don’t want to go through the trouble of installing these for small experiments. Ideally I’d like to have a simple way to provision a ready to use dev-environment pre-configured to handle the code I need to look at.
Remote development’s the fad these days
I’ve looked at services like Coder and GitPod that do this, but they’re either overkill, require a hosted repository, or didn’t give me the editing experience I hoped for. I have tried VS Code’s devcontainers in the past and they’re pretty good, but I have an OSS distribution now and can’t use their Remote Extensions anymore. So I decided to build something myself.
Need cleanly separated toolchain installations? Well that’s something Docker does really well. If I want to try Rust, I’d simply pull an image, start a container with my code mounted and play with it.
cd my-code
docker run -v $(pwd):/my-code --rm -it rust:latest
Simple and effective. But what about my editor? Giving up VS Code made me fall back in love with Neovim. I’ve sunk too many hours into getting my init.vim just right, and I don’t intend to
part ways with it anytime soon. I need to be able to use it inside the container.
I wasn’t able to get a client/server thing running for Neovim, so I settled for a simpler solution, just chuck it into
the image as well. I created a bunch of Dockerfiles extending standard images, copied over my Neovim config (along with all LSP configuration I might need) and started
hosting them over on GitHub Packages.
How do I connect a new terminal? With the container running I can do this to get a new shell into it.
docker container exec -it <container-id> /bin/sh
Now I get an editor for my workspace and a way to attach any number of terminals to it. Moreover, I can start keeping all my other dotfiles in these images as well, so ZSH, prompt etc. are all exactly how I like them.
Wrapping it all up
Okay so far so good, it’s everything I needed. But it’s ugly. I don’t want to write commands a billion characters long. I’m also not a fan of having to specify the image every time. So I wrapped it all up in a good ol’ bash script, boots.
All CLI arguments to boots (like the image name) can also be statically defined in a configuration file called .bootsrc in the directory where it’s called
cd my-code
boots start # if I have a .bootsrc
boots start rust:latest # if I need to specify a container
To connect a new terminal to the container I’ll do
boots connect <container-id>
There, much better. I can stop configuring and start experimenting again :D