- Nushell 61.8%
- Lua 18.3%
- Python 16.9%
- Shell 1.7%
- JavaScript 0.6%
- Other 0.7%
| Filename | Latest commit message | Latest commit date |
|---|---|---|
Several unrelated issues that surfaced while chasing "slow internet":
alsa-card-order.conf, install-alsa-card-order.nu (new)
The DP KVM / USB switcher returns devices on two hubs ~65s apart, so
ALSA card indices rotate on every switch (BRIO 6->4, Razer 4->5,
Yeti 5->6). PipeWire creates nodes at the new paths but leaves the old
ones behind retrying a path that now belongs to another device:
spa.alsa: 'front:4': playback open failed: No such file or directory
Duplicates share a node name, and both the default sink and
module-stream-restore resolve by name, so Firefox reattached to a dead
node, stalled ~60s on cubeb stream init, then played video with no
audio. Pin indices by VID:PID so enumeration order stops mattering.
Helper rather than post-install: the VID:PIDs are antimond-specific.
wayland.conf
Firefox's RDD process opened /dev/nvidia0 and renderD130 on a card with
no monitors attached. Not DRM device selection -- libglvnd loads EGL
vendors from /usr/share/glvnd/egl_vendor.d in filename order, so
10_nvidia.json won over 50_mesa.json for every EGL client. Restrict the
vendor list to mesa. Verified with eglinfo: 5 EGL devices -> 4, and the
NVIDIA vendor string disappears. Does not affect ollama (environment.d
is the user manager; both ollama units are system services).
ollama-cuda.service
ExecStartPost blocked "started" for ~7.3s of a 10.2s userspace boot, and
WantedBy=multi-user.target propagated that to graphical.target.
Background the warm; it stays in the cgroup, so the model is still
resident ~7s in. Userspace boot 10.181s -> 3.239s.
liquidcfg-user.service
Raced the logind uaccess ACL on /dev/hidraw* and failed at every boot,
leaving fans on firmware curves and skipping LED setup. Retry up to 30s.
reflector.conf
Root cause of the slow updates: --sort age picked freshly-synced but
slow mirrors, one serving ~6 KB/s at position 1. Rank by rate over a
20-mirror pool.
post-install/ollama.nu
Wrote its own shorter copy of the AMD drop-in to the same path
setup-ollama-gpus.nu uses, dropping CUDA_VISIBLE_DEVICES=-1 and letting
the AMD instance claim the NVIDIA card. Read the one source instead.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DYuU48VPGdVmsqzrGc2Qy5
|
||
| .themes/marshians/gnome-shell | ||
| .vscode | ||
| agents/skills | ||
| archive | ||
| dotfiles | ||
| helpers | ||
| nushell | ||
| nvim | ||
| packages | ||
| vms | ||
| windows | ||
| .gitignore | ||
| devdocs.json | ||
| etc-greetd-config.toml | ||
| etc-pam-d-greetd | ||
| install.nu | ||
| LICENSE | ||
| network_issue_summary.txt | ||
| README.md | ||
| scdaemon.conf | ||
| vscode-keybindings.json | ||
| vscode-settings-vim.json | ||
| webpackDevServer.config.js.patch | ||
| xterm-24bit.terminfo | ||
dotfiles
My dotfiles
Arch Linux
For faster pacman downloads during the initial bootstrap, uncomment the
ParallelDownloads line in /etc/pacman.conf (with =):
sed -i 's/#ParallelDownloads = 5/ParallelDownloads = 5/g' /etc/pacman.conf
(update-system does this automatically afterwards.)
Do a basic install — make sure you include:
base base-devel linux linux-firmware nushell git iptables-nft sudo [amd-ucode|intel-ucode] pipewire neovim efivar efibootmgr networkmanager
Dotfiles Installation
mkdir -p ~/dev
git clone https://github.com/icub3d/dotfiles ~/dev/dotfiles
cd ~/dev/dotfiles
nu install.nu
install.nu symlinks the nushell config and then calls update-system, which
installs paru, the packages listed in .selected_packages, and runs the
matching packages/post-install/*.nu scripts.
Boot & Login Manager (systemd-boot & greetd)
We configure systemd-boot loader entries with specific PCIe performance profiles, standard system loglevels, and beautiful Catppuccin console (TTY) colors. In addition, we configure greetd with tuigreet as the login manager using a matching console theme that launches a niri session.
Configuration Files
- systemd-boot entry:
/boot/loader/entries/arch.conf(sets kernel parameters and console palette) - greetd greeter:
/etc/greetd/config.toml(usestuigreetconfigured to launchniri-session)
Helper Script
A dedicated helper script is provided at helpers/boot-login.nu to audit and apply these configurations safely.
🔍 Run status check / audit:
To check if your local configurations are active and in-sync with the repository:
nu helpers/boot-login.nu check
# Run with sudo to view full detailed kernel options and partition UUID audit:
sudo nu helpers/boot-login.nu check
⚙️ Apply configurations:
You can deploy the greetd configuration and the systemd-boot loader entry together or separately:
# Apply both configurations and check status (recommended)
sudo nu -c "use helpers/boot-login.nu; boot-login apply-all"
# Or apply them individually:
sudo nu -c "use helpers/boot-login.nu; boot-login apply-boot"
sudo nu -c "use helpers/boot-login.nu; boot-login apply-greetd"
Note
apply-bootdynamically auto-detects the root partition's UUID usingfindmnt. You can optionally override this by passing the--uuidparameter:sudo nu -c "use helpers/boot-login.nu; boot-login apply-boot --uuid 'your-custom-uuid'"
Setup Git
Add this to ~/.gitconfig.local:
[user]
name = Joshua Marsh (icub3d)
email = joshua.marshian@gmail.com
signingkey = [KEY]
If you want GitHub auth (gh is installed via the package manifests):
gh auth login
gh auth setup-git
Syncthing
- Enable syncthing
systemctl enable --now --user syncthing
-
http://localhost:8384 > Actions > Show ID
-
http://otherhost:8384 > Add Remote Device
-
Back on http://localhost:8384 you'll eventually see a request that you can approve and then after a few seconds you'll also see the new shares that you can map.
3rd Monitor Not Working
I kept cycling the power (unplug from monitor) until it finally was recognized?!?!?
Firefox / Bitwarden Lag (Multi-GPU)
If you experience "bursty" typing lag in Firefox or Bitwarden popups on a multi-GPU setup (e.g., AMD + NVIDIA), it is likely due to the compositor trying to sync frames across GPUs.
The Fix:
- Prioritize the AMD card for wlroots-based compositors in
~/.config/environment.d/wayland.conf:
WLR_DRM_DEVICES=/dev/dri/card1:/dev/dri/card0
card1: Primary AMD GPU (7900 XTX)card0: Integrated AMD GPU- Excluded
card2: NVIDIA GPU (reserved for AI/CUDA)
- For Niri (which uses the Smithay backend and ignores
WLR_DRM_DEVICES), ignore the Nvidia GPU in~/.config/niri/config.kdl:
debug {
ignore-drm-device "/dev/dri/renderD130" // Nvidia GPU
}
GPU Helper Script:
Use nu helpers/gpu.nu to check clocks and performance levels. If UI stuttering persists during video playback, you can force the AMD GPU into high-performance mode:
nu -c 'use helpers/gpu.nu; gpu set-perf high'
Docker Slow Stopping
sudo systemctl edit docker and add:
[Unit]
After=containerd.service
Wants=containerd.service