nosignal/deckshift
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

fix: serialise portal recovery to avoid pipewire/portal race

Browse source 
The previous portal-recovery helper restarted xdg-desktop-portal-hyprland,
xdg-desktop-portal, pipewire, pipewire-pulse, and wireplumber in a single
systemctl --user restart call. That introduced a race where the portals
could come back up before wireplumber had finished rebuilding the node
graph, so the screencast portal would bind to nothing and "Share desktop"
would silently fail intermittently.

Reported by a user on the issue tracker: the cure is to (1) push the
live session env into D-Bus + systemd --user so D-Bus-activated portals
target the new Wayland socket, (2) stop portals first, (3) kill any
gamescope-era zombies (SIGTERM then SIGKILL for stragglers — SIGKILL
alone leaves stale D-Bus name registrations), (4) restart the pipewire
stack and wait for wireplumber, (5) bring portals back up last.

The exec-once marker guard and the initial 2s sleep are preserved so
this stays a no-op on normal logins and so the env-import sees a
populated WAYLAND_DISPLAY.

Bumps DECKSHIFT_VERSION from the stale 0.1.1 string in the file (last
shipped tag is 0.1.3) to 0.1.4.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
28allday 2026-05-12 19:55:16 +01:00
parent 49046c4444
commit 8f0599f221
1 changed files with 33 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

42
deckshift.sh
View file

@ -34,7 +34,7 @@ set -Euo pipefail
# -u: Treat unset variables as errors (catches typos in variable names)
# -o pipefail: A pipeline fails if ANY command in it fails, not just the last one
DECKSHIFT_VERSION="0.1.1"
DECKSHIFT_VERSION="0.1.4"
# Resolve the directory this script lives in so we can find sibling files like
# bin/deckshift-settings and applications/deckshift-settings.desktop when
@ -2424,6 +2424,14 @@ KEYBIND_MONITOR
# checks for that marker on every Hyprland start, and if present, bounces
# the portal + pipewire user services so they reattach to the live HIS.
# No marker = no-op (cheap; runs in milliseconds).
#
# The recovery sequence is ordered to avoid a race that earlier versions
# hit when restarting all five services simultaneously: portals could come
# up before wireplumber had finished building the node graph, so the
# screencast portal would bind to nothing. We now (a) push live session
# env into D-Bus + systemd --user so portals activate against the new
# Wayland socket, (b) stop portals first, (c) restart pipewire and wait
# for the graph, (d) start portals last.
info "Creating portal recovery helper..."
local portal_recovery="/usr/local/bin/deckshift-portal-recovery"
@ -2432,16 +2440,32 @@ KEYBIND_MONITOR
[[ -f /tmp/.deckshift-just-returned ]] || exit 0
rm -f /tmp/.deckshift-just-returned
# Give the new Hyprland a moment to fully come up and the user manager to
# import HYPRLAND_INSTANCE_SIGNATURE / WAYLAND_DISPLAY into its environment.
# Give the new Hyprland a moment to fully come up and export its env to the
# user manager (HYPRLAND_INSTANCE_SIGNATURE / WAYLAND_DISPLAY).
sleep 2
systemctl --user restart \
xdg-desktop-portal-hyprland.service \
xdg-desktop-portal.service \
pipewire.service \
pipewire-pulse.service \
wireplumber.service 2>/dev/null || true
# Pull the live session env into systemd --user and the D-Bus activation env,
# so D-Bus-activated portals bind to the new Wayland socket, not the dead one
# left over from the gamescope session.
systemctl --user import-environment WAYLAND_DISPLAY XDG_CURRENT_DESKTOP XDG_SESSION_TYPE 2>/dev/null || true
dbus-update-activation-environment --systemd WAYLAND_DISPLAY XDG_CURRENT_DESKTOP XDG_SESSION_TYPE 2>/dev/null || true
# Stop portals first so they don't try to talk to a half-restarted pipewire.
systemctl --user stop xdg-desktop-portal-hyprland.service xdg-desktop-portal.service 2>/dev/null || true
# Kill any zombies left over from gamescope's session (SIGTERM, then SIGKILL
# only for stragglers — SIGKILL alone leaves stale D-Bus name registrations).
pkill -TERM -x xdg-desktop-portal-hyprland xdg-desktop-portal xdg-desktop-portal-wlr 2>/dev/null
sleep 0.5
pkill -KILL -x xdg-desktop-portal-hyprland xdg-desktop-portal xdg-desktop-portal-wlr 2>/dev/null
systemctl --user reset-failed xdg-desktop-portal-hyprland.service xdg-desktop-portal.service 2>/dev/null || true
# Restart pipewire stack and wait for wireplumber to rebuild the node graph
# before portals come back up (otherwise screencast portal binds to nothing).
systemctl --user restart pipewire.service pipewire-pulse.service wireplumber.service 2>/dev/null || true
sleep 2
# Now bring portals up cleanly.
systemctl --user start xdg-desktop-portal-hyprland.service xdg-desktop-portal.service 2>/dev/null || true
PORTAL_RECOVERY
sudo chmod +x "$portal_recovery"