The previous fix listed both profiles' primaries, "output": ["eDP-1",
"DP-1"] -- but DP-1 is the laptop's second monitor too, so the laptop
would have got a bar on it as well as on eDP-1. No static list can express
"the main monitor" when the two machines share an output name.
Resolve it at launch instead. hypr/scripts/bar asks Hyprland for the
monitor at position 0x0 -- the main monitor in both profiles -- writes
{"output": "<name>"} to ~/.cache/waybar-output.jsonc, and execs waybar;
waybar/config.jsonc includes that fragment. autostart/services, the
bar_reload bind and the performance script all go through the wrapper, so
a reload re-resolves after docking. Signals still target the waybar
process, not the wrapper.
Verified all three paths on mainPC: the fragment pins the bar to DP-1
only, an empty fragment falls back to a bar on every output, and a
fragment naming a disconnected output maps nothing (the failure mode this
started from). Config parses on both profiles.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01StGBghPW9Dm7z7uj2VNWHr
The deployed config pinned "output": "eDP-1", the laptop's internal
display. On the PC waybar started fine but never mapped a layer surface,
so there was no bar and SIGUSR1 had nothing to toggle. Listed both
profiles' primaries instead -- waybar skips names that are not connected,
so the one file works on the laptop and the PC.
Two bugs in the same config that this uncovered:
- bars/main.jsonc included modules/temperature.cpu.jsonc, but the file is
temperature-cpu.jsonc, so the temperature#cpu module the bar referenced
was silently absent.
- get_wg_status.sh ran `sudo wg show` with no tty, logging "a terminal is
required to read the password" on every poll. Now `sudo -n`, which fails
instantly and falls into the existing "disconnected" branch, and starts
working as-is if a NOPASSWD rule is ever added.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01StGBghPW9Dm7z7uj2VNWHr
Startup errors:
- catppuccin: the bufferline integration moved to catppuccin.special.bufferline
- lspconfig-extended: drop the removed LazyVim keymaps API for
opts.servers["*"].keys, and move remove-workspace-folder to <leader>clR
so it stops colliding with LspRestart on <leader>clr
- options: lazyvim_cmp = "blink" is not a valid value, it is "blink.cmp"
Language support (python, c/cpp, rust, zephyr, and the common ones) via
lazyvim.json extras: dap.core, clangd, cmake, docker, git, json, rust,
toml, yaml.
- lang/zephyr.lua: gated entirely on ZEPHYR_BASE / ZEPHYR_SDK_INSTALL_DIR /
GNUARMEMB_TOOLCHAIN_PATH, returning {} when none are set, so it is inert
on machines without Zephyr. Nothing is hardcoded: the clangd
--query-driver globs are built from those vars, falling back to resolving
the cross-gcc through exepath()+fs_realpath(). Adds devicetree/kconfig
treesitter and filetypes (*.overlay, Kconfig.*, prj.conf, *_defconfig).
- clangd/config.yaml: strip the GCC-only flags Zephyr emits into
compile_commands.json (-fno-printf-return-value, -mfp16-format=*, ...),
which clang rejects as "Unknown argument". Remove only drops flags that
are present, so it is a no-op for native builds. Takes a built Zephyr app
from 15 diagnostics to 0, with CONFIG_* resolving into the generated
autoconf.h.
- npm-servers.lua: npm is not installed here, so yaml/docker servers failed
to install on every startup and to spawn on every matching file. Sets
both enabled = false and mason = false for servers whose binary is
missing from PATH and from mason/bin.
Project runner under <leader>R (<leader>r is refactoring), dispatching on
the detected project type -- zephyr, rust, python, cmake, make:
Rr run project Rb build Rt test RF build+flash Rf run file
Rk configure Rc clean Rs select target/board Ri info
Zephyr boards are picked with vim.ui.select over `west boards`, cached per
scan and persisted per project root under stdpath("state"), never written
into the project. A build with no board set prompts first, then proceeds.
Detection walks up from the last real file buffer, tracked by an autocmd,
because a terminal opened by one of these verbs would otherwise become the
buffer detection ran from.
Also <leader>b1..b9/b0 jump to buffer N (ordinal, no longer skipping the
current one) and <leader>b<Left>/<Right> cycle.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01StGBghPW9Dm7z7uj2VNWHr
With the Lua config, `hyprctl dispatch` evaluates Lua rather than the
hyprlang dispatcher strings, and `hyprctl keyword` is refused outright
("can't work with non-legacy parsers"). Three scripts still used the old
forms and silently failed.
- toggle_floating (SUPER+W): rewritten around hl.dsp.window.{float,resize,
center}. The Lua resize dispatcher takes absolute pixels only, so the
70% is resolved against the logical size of the window's monitor, and
fullscreen is left first since resize is a no-op while fullscreen.
Floats every window at 70%x70% now, not just 50%x55% for kitty.
- performance: runtime overrides via `hyprctl eval hl.config{}`; drop the
removed decoration:drop_shadow for decoration:shadow:enabled; read the
gamemode flag from `getoption -j` (.bool), which the awk on line 2 had
been picking the "set:" line out of.
- move_by_rules: movetoworkspace -> hl.dsp.window.move{workspace=N}.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01StGBghPW9Dm7z7uj2VNWHr
hyprlang has been deprecated since Hyprland 0.55 and is dropped after 1-2
releases; this box runs 0.56.2. Migrate Hyprland's own config to Lua and
split the machine-specific parts into profiles.
Config:
- hyprland.lua entry point; the .conf tree is left in place as a rollback
path (Hyprland prefers hyprland.lua and only decides at startup).
- Each require() is its own Lua scope, so hyprlang's global "$VAR" no longer
crosses files: vars.lua / apps.lua / theme/colors.lua return tables.
- hypridle, hyprlock, plugins/* and pyprland.toml stay hyprlang -- the other
hypr* tools keep it.
Profiles (configs/profiles/):
- pc -> DP-1 / HDMI-A-1 / DP-2, all 1920x1080
- laptop -> eDP-1 3072x1920@120 scale 2.0, DP-1, HDMI-A-1 mirroring eDP-1
- Each owns its monitors and workspace-to-monitor mapping; common.lua
generates the 30 workspace rules from the three outputs.
- Selected by $HYPR_PROFILE, else hostname (mainPC -> pc), else laptop.
Fixes:
- pyprland.toml called swww, which is not installed; the wallpaper backend
here is awww. The rotation timer failed silently and both outputs stayed
at color: 000000.
- XF86InputRaiseVolume / XF86InputLowerVolume are not real keysyms (no
XF86Input* family in xkbcommon). They were dead in .conf too; hyprlang
ignored them silently, hl.bind rejects them. Commented out.
- SUPER+SHIFT+R was bound twice (notification reload + bar reload). hyprlang
ran both; in Lua the second would shadow the first, so they are merged.
- The waybar layer rule set no effects and was a no-op; dropped.
- Polkit autostart now picks hyprpolkitagent or polkit-gnome by what is
installed, so one file works on both machines.
Verified with `Hyprland --verify-config` on both profiles (config ok).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01StGBghPW9Dm7z7uj2VNWHr
