docker

Docker configurations and conventions that support my personal workflows.

Overview

I generally like to use the native macOS host environment for my development work, and to enjoy the speed, interpretability, and comfort of a native setup. However, virtualization is sometimes a must, and the right tool.

I'm finding this especially true for agentic tools like Claude Code. I also sometimes need it for tools that just aren't as well supported on macOS.

This docker directory is for me to build up my own conventions, Docker images, and ideas around how Docker fits into my dev workflow.

my-dev Docker Image

The my-dev Docker image is designed as a base image for my development work that happens in containers.

Miscellaneous Notes

• I do not want to just use Docker for the sake of using it.
• I like the ethos of package managers. I want to get leverage out of them, and sprinkle in custom build scripts to fill in the gaps between what they offer and the snowflakes of what I need and the snowflakes of dependencies I use that don't always exist in package managers
• Right now, I am not intending to build Docker images for a "runtime" workload. So, not sure I'll use build stages much. My workloads are all just more dev environments to build even more software
• I prefer the Docker exec form of RUN. I really don't like the shell form of RUN.
• I don't like the ergonomics of Dockerfile files. I don't like how the ecosystem has chosen to do heavy inheritance-based relationships of "using the Node Docker image for Node apps", or "using the Python Docker image for Python apps". What if you need both?
• I'd like to inherit from a base Linux distribution image, then layer in the tools I need via COPY'ed in scripts

Instructions

Follow these instructions to build the my-dev Docker image and interact with it.

1. Activate the do.nu script

   • do activate

2. Build the my-dev image

   • do build

3. Run the container and start a Nushell session

   • do run

4. Say "hello world" from the container using Node.js

   • node -e 'console.log("Hello, World!")'

   • When you're done, exit the shell session with exit. Then stop the container with the following command.

   • do stop

Wish List

General clean-ups, TODOs and things I wish to implement for this project:

• DONE Layer in Node.js into the base development image
• DONE Layer in Claude Code.
• I haven't figured out if the build being so slow (over 2 min) is going to get annoying and what I can/should do about caching... doesn't matter yet. Also I might need to cache bust just in general (I guess that's why I added --no-cache optional flag in the nushell script)
• Consider BuildKit LLB, and/or Dagger. I'd like to see what's possible. I'd like to fuse the downloading/untarring that can happen outside the container, and the instructions for copying it into the container, in the same program/project for cohesion. I'd just like to see what's possible. I'd really like to see if I can write some code that cache-busts based on the hash of the scripts we copy in. One thing I think is really neat about Docker/containers is the permanence of setting env vars, instead of having to bootstrap that from the shell. Doing it earlier than shell is nice.
• DONE (This is causing problems for the Claude Code install; so need to address it) I need to re-think some of root/user stuff. Install Rustup/Cargo as root is kind of obnoxious because now my Cargo bin is in /usr/local/bin so when 'me' user tries to build a project it won't have write access to put the binary there.
• Consider the idea: can I get away with saving containers in a stateful way? Instead of saving images? Is there a compressed workflow hidden somewhere in here? Now I'm thinking about putting DevContainer's style instructions/Dockerfile in the .my directory and using that by convention.
• Define an agentic runbook (instructions in a prompt + scripts) to find latest versions of deps and update them in our code
• Consider sandboxing/LSM stuff (e.g. Landlock) for Claude Code similar to what I did with seatbelt. I'd like the exec tool (the shell it uses is configurable I think) to be some wrapper executable that self-sandboxes bash
• Layer in my own Nu scripts. A pre-req to this will be splitting out my macOS-specific stuff. Shouldn't be too crazy.
• Get my rules MCP server in the image, and also my rules files.
• SSH server. It's been three years that I've been off-and-on exploring the "dev containers" idea (linux-playground, dev-containers-playground, intellij-playground, vscode-playground) and SSH is central to how clients interface with the dev container. I'd like to run an SSH server in the container. DevPod does something super clever, which is install an agent in the container and then makes special SSH config entry using ProxyCommand to basically get SSH but via stdout/stdin? I won't do that I don't think, but clever.