Skip to content

SpoddyCoder/wsl-builds

BranchesTags

Repository files navigation

WSL Builds

Lint Tests

This project contains clean, simple WSL2 builds that use Windows host native implementations of core components for maximum performance, convenience and flexibility.

  • Combine components from different builds to tailor WSL environments for different purposes.
  • Simple installations, often featuring quality of life configurations and helpers.
  • Caching of large downloads on the Windows host, so you don't have to re-download install packages for rebuilds.
  • Other build specific cache directories on the Windows host for convenience (eg: AI models and .pkl cache).
  • Project motivation: WSL instances are disposable - streamlines a clean rebuild whenever its needed.

Install

After provisioning a basic WSL instance, clone the project repo on it...

git clone https://github.com/SpoddyCoder/wsl-builds.git

Configure

Run the configure wizard to create a wsl-builds.conf from the template...

cd wsl-builds
./configure.sh

Building

./wsl-builder.sh <build-dir> <component>[,<component>...] [additionalargs]... [--force]
  • build-dir any valid build directory name, see Build List. Each build has it's own README.md.
  • component[,<component>...] comma separated list of build components (packages to install etc.), varies per build.
  • additionalargs... additional arguments required for some builds.
  • --force by default components that are already installed will be skipped with warning messages. Use this flag to force reinstallation of already-installed components.
  • The builder should be run as your current user, but some components will run commands that require escalated privileges using sudo.
  • It is not designed to be used non-interactively, many installs require user input / confirmation.

Assembling and combining builds

./wsl-builder.sh            # show all builds
./wsl-builder.sh dev-js     # show components for build dev-js

# build a gen dev env
./wsl-builder.sh system update,qol,symlinks
./wsl-builder.sh dev essentials,vscode
./wsl-builder.sh dev-js node,yarn,nvm,essentials
change-hostname my-dev-box

# build a python environment for AI coding
./wsl-builder.sh system update,qol,symlinks
./wsl-builder.sh dev essentials,cursor
./wsl-builder.sh dev-python conda
./wsl-builder.sh ai cuda132
change-hostname python-ai
  • Each build is very simple, containing only a few related components intended to deliver a single purpose.
  • Pick and choose! Use multiple runs of the build tool to combine components, packages, and features from different builds as you need.
  • Build history is kept in ~/.wsl-build.info
  • Maintainers: optional advisory component reviews (no installs) are documented in review/README.md.

Stacks

Saved build sequences live under stacks/<namespace>/ as .wslb files (one build-dir and comma-separated components per line; blank lines and # comments are ignored). ./wsl-stacker.sh runs the builder once per line:

./wsl-stacker.sh <namespace> <stack-name> [--force]
  • With no arguments, or with a namespace only, it prints usage and available namespaces or stacks (same idea as listing builds or components with the builder).
  • The first argument may be a stacks directory path instead of a repo stacks/<namespace> shorthand.
  • Shipped examples: stacks/spoddycoder/README.md.
  • Each line is a separate the builder run; apt index refresh is throttled across those runs in the same stack session (shared stamp and package-list freshness).
./wsl-stacker.sh spoddycoder dev

Build List

Build Packages, Frameworks, Tools & Extras Additional Conf
ai cuda124: CUDA 12.4
cuda132: CUDA 13.2
ollama
huggingface-cli: Hub hf CLI (pip)
llama-cpp: build from source, optional CUDA
cuda-wsl-lib-symlinks: fix 'not symlinks' issue		 ollama: OLLAMA_MODELS
huggingface-cli: HF_HOME,
HF_HUB_CACHE
ai-agents setup-env: conda agents env,
python-dotenv, httpx,
jupyter, pydantic
langchain: LangChain core
langchain-ollama: Ollama
integration
langchain-llama-cpp:
langchain-community +
llama-cpp-python
langgraph: LangGraph
openai-agents: OpenAI
Agents SDK
langsmith: LangSmith Python SDK
(LangSmith Cloud)
langfuse: Langfuse Python SDK
(Langfuse Cloud)
mcp: Model Context
Protocol Python SDK
mcp-inspector: MCP
Inspector (npm global)
ai-resources sg3: stylegan3, pkl cache, pytorch cache
lsd: lucid-sonic-dreams
spleeter
rudalle
bfcl-eval: Gorilla BFCL benchmark		 sg3, lsd, spleeter,
rudalle, bfcl-eval:
AI_RESOURCES_PROJECT_DIR
sg3: STYLEGAN3_PKL_CACHE,
STYLEGAN3_PYTORCH_CACHE
bfcl-eval: BFCL_PROJECT_ROOT
db mysql-client
mysql-server
postgres-client
postgres-server
dev essentials: curl, wget, git, vim, nano, jq, yq
vscode
cursor: tree, code alias		 essentials: GIT_*,
GIT_CREDENTIALS_HELPER
dev-bash shellcheck
bats
dev-js node: Node.js, npm
nvm
yarn
pnpm
react: create-vite, react-devtools
nextjs
angular
vue: create-vue
express
essentials: TypeScript, ESLint, Prettier, PM2, nodemon, serve
dev-python python3
pipx
uv
conda: Anaconda
dev-ssg hugo
jekyll: Bundler, Ruby deps
media ffmpeg: CLI encoder,
transcoder
ffmpeg-dev: ffmpeg CLI +
libav dev libs, libx264-dev
libprojectm: libprojectM
from source, SDL UI		 libprojectm:
LIBPROJECTM_SRC_DIR,
LIBPROJECTM_INSTALL_PREFIX
devops docker
docker-desktop
terraform
packer
kubectl
k9s
devops-aws awscli
qol: aws-profile alias
security harden-sandbox: DrvFs automount off,
interop off, optional ~/ → /mnt symlink removal
system update: apt update + upgrade
essentials: htop, rsync
x11: Windows native GUI
smb: smbclient, cifs-utils
nfs: nfs-common
systemd
wslu: wslview, wslsys
qol: safety aliases, change-hostname, default user, WSL2 Distro Manager login → $HOME
symlinks: host path → ~/… links
apt-mirror-switch: Canonical vs Uni of Kent apt mirror helper
fstab: WSL mount config		 symlinks: SYMLINK_HOST_*

Advanced Configuration

  • The builder loads WSL_BUILDS_CONF (full path) if set, else ~/.wsl-builds.conf.
  • On the Windows host, a shared file under %USERPROFILE%\.wsl-builds\wsl-builds.conf is useful if you want many WSL2 instances configured in the same way.
    • Choosing it in the configuration wizard saves export WSL_BUILDS_CONF=… in your shell rc.
  • Otherwise the configuration wizard creates ~/.wsl-builds.conf on the WSL instance
  • The wsl-builds.conf file contains optional paths and settings
    • See the wsl-builds.conf.example for more info on what's available.
    • Each build directory README will call out configurable values for that component.
    • Tip: Caching things on the host can be uesful.
  • ./configure.sh --noninteractive if the default Windows host wsl-builds.conf already exists it is adopted, otherwise the wizard uses the shipped default.

External Builds Root

  • EXTERNAL_BUILDS_ROOT in wsl-builds.conf can point at another builds/ directory outside the repo.
  • Each inner build directory must have conf.sh, install.sh, and README.md at its root; component install scripts live at builds/<build-dir>/<slug>/install.sh (CSV hyphens in the token map to underscores in slug, same as the builder dispatch).
  • See CONTRIBUTING.md for more info.

External Stacks Root

  • EXTERNAL_STACKS_ROOT in wsl-builds.conf can point at another stacks/ directory outside the repo.

Enabling + Configuring WSL2 on the Windows Host

  • Open PowerShell as an Administrator
  • wsl --install

Git Integration

  • Install git on the Windows machine, ensure you select the option to use unix line endings.
    • We are coding in linux, not Windows.
  • To use the credentials from your Windows host on the WSL instance, see the wsl-builds.conf section.

Cursor Integration

  • Install Cursor on the Windows machine, open it and launch it.
  • Press CTRL + SHIFT + P to bring up the command palette
  • Search for and run: Shell Command: Install 'cursor' command
  • You can now type cursor on the WSL instance to launch the current directory in a Cursor editor, running in Windows, with a remote connection to your WSL instance — sweet.

VSCode Integration

CUDA Integration

  • Install an NVIDIA Windows driver with WSL CUDA support
  • Use ./wsl-builder.sh ai cuda132 to install the CUDA toolkit (or cuda124) on the WSL instance. Do not install Linux GPU drivers in WSL.
  • For more info: CUDA on WSL User Guide
  • If you see libcuda.so.1 is not a symbolic link, run the ai cuda-wsl-lib-symlinks component to fix it.

Windows Host Resource Allocation

[wsl2]
memory=24GB
swap=16GB
networkingMode=mirrored       # Bridge Windows and WSL2 networking (use this carefully)

[experimental]
autoMemoryReclaim=gradual
autoMemoryReclaim=dropcache   # better for ai workloads
sparseVhd=true                # reduces disk space usage with neglibible i/o overhead

WSL2 Distro Manager

Useful GUI for managing instances: https://github.com/bostrot/wsl2-distro-manager

WSL Command Reference

Useful WSL commands

wsl -l -v
wsl --list --online
wsl --install Ubuntu-22.04
wsl --export Ubuntu E:\WSL\builds\build-name
wsl --unregister Ubuntu-22.04
wsl --import my-project-name E:\WSL\instances\project-name D:\WSL\builds\build-name
wsl --shutdown
wsl --distribution my-project-name

Manual Snapshots

  • Shutdown the instance, export it to your build dir and kill it;
    • wsl --shutdown
    • wsl --export Ubuntu-22.04 E:\WSL\builds\system-base
    • wsl --unregister Ubuntu-22.04
  • Your buttery biscuit base is ready to create new instances from :)
    • wsl --import my-new-project E:\WSL\instances\my-new-project E:\WSL\builds\biscuit

About

Because native Linux development on Windows facilitates some very easy and comfortable workflows

Resources

Readme

License

MIT license

Contributing

Contributing
Activity

Stars

1 star

Watchers

1 watching

Forks

0 forks
Report repository

Contributors

Languages