fix: start:wayland defaults to wrong compositor socket - #4268
flightlesstux wants to merge 1 commit into
Conversation
WAYLAND_DISPLAY fell back to wayland-1 when unset, but that's not the real socket on every setup (wayland-0 on mine). With a saved pm2 env that never set WAYLAND_DISPLAY, electron connected to the wrong socket and SIGSEGV'd on every launch, and pm2's auto-restart turned that into a crash loop. Now it checks XDG_RUNTIME_DIR for the actual wayland-* socket first and only falls back to wayland-0 if it can't find one.
| "start": "node --run start:wayland", | ||
| "start:dev": "node --run start:wayland -- dev", | ||
| "start:wayland": "WAYLAND_DISPLAY=\"${WAYLAND_DISPLAY:=wayland-1}\" ./node_modules/.bin/electron js/electron.js --ozone-platform=wayland", | ||
| "start:wayland": "WAYLAND_DISPLAY=\"${WAYLAND_DISPLAY:-$(ls \"${XDG_RUNTIME_DIR:-/run/user/$(id -u)}\" 2>/dev/null | grep -m1 '^wayland-[0-9]*$')}\"; WAYLAND_DISPLAY=\"${WAYLAND_DISPLAY:-wayland-0}\" ./node_modules/.bin/electron js/electron.js --ozone-platform=wayland", |
There was a problem hiding this comment.
Could this be simplified to the following?
| "start:wayland": "WAYLAND_DISPLAY=\"${WAYLAND_DISPLAY:-$(ls \"${XDG_RUNTIME_DIR:-/run/user/$(id -u)}\" 2>/dev/null | grep -m1 '^wayland-[0-9]*$')}\"; WAYLAND_DISPLAY=\"${WAYLAND_DISPLAY:-wayland-0}\" ./node_modules/.bin/electron js/electron.js --ozone-platform=wayland", | |
| "start:wayland": "./node_modules/.bin/electron js/electron.js --ozone-platform=wayland", |
The current script explicitly defaults WAYLAND_DISPLAY to wayland-1, while the standard Wayland default is wayland-0. On my Debian system, removing the explicit assignment works correctly.
I haven't tested this on a Raspberry Pi, though. Could someone confirm that Electron also uses wayland-0 there when WAYLAND_DISPLAY is unset?
There was a problem hiding this comment.
works with unset WAYLAND_DISPLAY on a pi 4:
#### System Information ####
- MM: version: v2.38.0-develop; git: e1cc0696; branch: develop
- SYSTEM: manufacturer: Raspberry Pi Foundation; model: Raspberry Pi 4 Model B Rev 1.5; virtual: false
- OS: platform: linux; distro: Debian GNU/Linux; release: 13; arch: arm64; kernel: 6.18.50+rpt-rpi-v8
- VERSIONS: electron: 44.1.1; used node: 24.19.0; installed node: 22.22.2; npm: 10.9.7; pm2:
- ENV: XDG_SESSION_TYPE: tty; MM_CONFIG_FILE: undefined
WAYLAND_DISPLAY: undefined; DISPLAY: undefined; ELECTRON_ENABLE_GPU: undefined
- RAM: total: 3795.74 MB; free: 3183.61 MB; used: 612.13 MB
- OTHERS: uptime: 18 minutes; timeZone: Europe/BerlinThere was a problem hiding this comment.
@sdetweil What do you think about the simplified version "./node_modules/.bin/electron js/electron.js --ozone-platform=wayland" ?
There was a problem hiding this comment.
is that a temporary parameter? or do you think it will last?
There was a problem hiding this comment.
Good question! I just tested this with Electron 44.1.1 on Debian, and MagicMirror started successfully without --ozone-platform=wayland.
Could we simplify the script even further to this?
"start:wayland": "./node_modules/.bin/electron js/electron.js"
There was a problem hiding this comment.
no, this fails on my pi 4 ...
There was a problem hiding this comment.
mm@pi4-argon:~/MagicMirror $ env | grep -i xdg_
XDG_SESSION_TYPE=tty
XDG_SESSION_CLASS=user
XDG_SESSION_ID=4
XDG_RUNTIME_DIR=/run/user/1000Running XDG_SESSION_TYPE=wayland ./node_modules/.bin/electron js/electron.js works. Electron uses XDG_SESSION_TYPE if --ozone-platform=auto (the default).
So we should not remove --ozone-platform=wayland.
There was a problem hiding this comment.
Okay, thanks for testing! So "./node_modules/.bin/electron js/electron.js --ozone-platform=wayland" stays the candidate.
Out of curiosity: Why do you have XDG_SESSION_TYPE=tty and not XDG_SESSION_TYPE: wayland?
There was a problem hiding this comment.
this is the default in raspberry pi os
There was a problem hiding this comment.
@flightlesstux Please confirm if "start:wayland": "./node_modules/.bin/electron js/electron.js --ozone-platform=wayland" also works for you.
4 participants
WAYLAND_DISPLAY falls back to wayland-1 when it's not set. On my Pi the real socket is wayland-0, so electron connected to nothing and crashed on every launch. Since pm2 auto-restarts the process, this turned into a crash loop that filled my disk with core dumps and broke git, npm and pm2 along with it.
This checks XDG_RUNTIME_DIR for the actual wayland-* socket and only falls back to wayland-0 if nothing is found.
Tested manually with a socket present, no socket, and WAYLAND_DISPLAY already set. Ran node --run lint:fix, no issues.