max / alloy

swayosd: enable the backend the caps-lock bind needs, drop the preset line that named nothing Two defects in the same package's wiring, found by checking whether `exec swayosd-server` in the sway config was redundant the way `exec xdg-user-dirs-update` turned out to be. It is not: Fedora's SwayOSD 0.3.2 ships three binaries, one system unit and a D-Bus service file, and no user unit at all. The sway `exec` is the mechanism, not a workaround. So the user preset's `enable swayosd.service` named a unit that does not exist. It has been a no-op reading as configuration since it was written. Removed, with the reason recorded next to the alloy-hinged note above it, which was removed on 2026-07-19 for exactly the same reason. The other two lines in that file name units that do exist; checked rather than assumed. The half that does ship a unit was never enabled by anything, so the sway config's `bindsym --release Caps_Lock` had no daemon behind it. Enabling it turned out to need more than a preset line: SwayOSD installs swayosd-libinput-backend.service into /usr/lib64/systemd/system, and systemd's search path is /etc/systemd/system, /usr/local/lib/systemd/system and /usr/lib/systemd/system. /usr/lib64 is a real directory here, not a symlink to /usr/lib, so the unit is invisible: `systemctl enable` answers "Unit swayosd-libinput-backend.service does not exist" and `preset-all` matches nothing and says nothing, which is the same silent failure the alloy-debug-shell@ comment in this file already warns about. The Containerfile now links the unit into the path systemd reads, before preset-all runs. Guarded both ways: if a later SwayOSD ships the unit at the canonical path the shim stands aside rather than replacing a real unit file with a link to a file that moved, and if the unit is at neither path the build fails instead of quietly shipping the dead bind again. Verified in the rebuilt image: the unit resolves, and preset-all creates graphical.target.wants/swayosd-libinput-backend.service. Not yet verified at runtime, which is a different claim: that the daemon starts in a session and the bind draws an overlay needs a boot.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: Max Johnson <me@maxj.phd> · 2026-07-22 03:18 UTC

Signed with PGP, not checked

Commit: c1cd25ef5eeda30fc88183bb1e6b4f6e4bef7fa0

Parent: 7b977b2

3 files changed, +54 insertions, -3 deletions

M Containerfile +29

			@@ -382,6 +382,35 @@
382	382		# preset-all` reads every preset file and applies it — this must
383	383		# run after the config tree is in place, so it lives here.
384	384		# =====================================================================
	385	+
	386	+	# SwayOSD's system unit is not where systemd looks for units. Fedora's
	387	+	# SwayOSD 0.3.2 installs swayosd-libinput-backend.service into
	388	+	# /usr/lib64/systemd/system, and systemd's search path is /etc/systemd/system,
	389	+	# /usr/local/lib/systemd/system and /usr/lib/systemd/system. On this base
	390	+	# /usr/lib64 is a real directory rather than a symlink to /usr/lib, so the
	391	+	# unit is invisible: `systemctl enable` answers "Unit
	392	+	# swayosd-libinput-backend.service does not exist", and the preset line for it
	393	+	# matches nothing and says nothing. That is a packaging bug upstream, not a
	394	+	# choice Alloy gets to make, so link the unit into the path systemd reads.
	395	+	#
	396	+	# Without this the caps-lock and num-lock overlays have no daemon, and the
	397	+	# sway config's `bindsym --release Caps_Lock` does nothing at all.
	398	+	#
	399	+	# Written to survive the fix rather than assume the bug is permanent: if a
	400	+	# later SwayOSD ships the unit at the canonical path, this leaves it alone
	401	+	# instead of replacing a real unit file with a link to a file that moved. If
	402	+	# the unit is at neither path the build stops, because the alternative is
	403	+	# shipping the dead bind again and not knowing.
	404	+	RUN canon=/usr/lib/systemd/system/swayosd-libinput-backend.service; \
	405	+	real=/usr/lib64/systemd/system/swayosd-libinput-backend.service; \
	406	+	if [ -e "$canon" ]; then \
	407	+	echo "swayosd unit is already in the search path; the shim can go"; \
	408	+	elif [ -e "$real" ]; then \
	409	+	ln -s "$real" "$canon"; \
	410	+	else \
	411	+	echo "swayosd-libinput-backend.service is at neither path" >&2; exit 1; \
	412	+	fi
	413	+
385	414		RUN systemctl preset-all
386	415
387	416		# Template instances have to be enabled by name. `preset-all` iterates over

M etc/systemd/system-preset/50-alloy.preset +15

			@@ -4,6 +4,7 @@
4	4		# units are enabled or disabled by default. Rationale for each entry:
5	5		#
6	6		# - greetd: boot lands on the tuigreet TUI on VT1.
	7	+	# - swayosd-libinput-backend: the caps-lock and num-lock overlays.
7	8		# - tailscaled: opt-in only. Enrollment happens via `alloy tail`
8	9		# (see docs/CONTINUITY.md — Alloy does not phone home
9	10		# before the user says yes).
			@@ -26,4 +27,18 @@
26	27		# without a word, which reads exactly like the feature being broken. Its
27	28		# instances are enabled by name in the Containerfile instead.
28	29
	30	+	# Volume and brightness overlays need no daemon beyond `swayosd-server`,
	31	+	# which the sway config exec's, but the lock-key ones do: reading caps-lock
	32	+	# and num-lock state means watching the input devices, which a user session
	33	+	# cannot do. That is why this half of SwayOSD ships as a system unit, and
	34	+	# it is the only part of the package that has a unit at all. Without it the
	35	+	# sway config's `bindsym --release Caps_Lock` has nothing behind it, which
	36	+	# was the state until 2026-07-21.
	37	+	#
	38	+	# The privilege is real and worth naming: the daemon opens the libinput
	39	+	# devices to watch modifier state. Accepted because the alternative is a
	40	+	# bind that silently does nothing, and because it reads state rather than
	41	+	# injecting it.
	42	+	enable swayosd-libinput-backend.service
	43	+
29	44		disable tailscaled.service

M etc/systemd/user-preset/50-alloy.preset +10 -3

			@@ -3,18 +3,25 @@
3	3		# Applied on first user login for services that run under the user
4	4		# session (not system-wide). Rationale for each entry:
5	5		#
6		-	# - swayosd: Fn keys need the daemon live to render OSD overlays.
7	6		# - syncthing: opt-in only. Enrollment via `alloy sync`
8	7		# (see docs/CONTINUITY.md).
9	8		# - gammastep: opt-in only. User needs to set their location
10	9		# before enabling makes sense.
11	10		#
	11	+	# Both units above exist in the image; checked, because two lines here
	12	+	# have named units that do not.
	13	+	#
12	14		# alloy-hinged was enabled here until 2026-07-19. The fold-to-notes
13	15		# daemon was shelved in the 2026-07-17 pivot (see
14	16		# docs/HARDWARE-FW12.md#fold-to-notes-shelved) and no unit was ever
15	17		# shipped, so the line only produced a failed enable at first login.
16		-
17		-	enable swayosd.service
	18	+	#
	19	+	# swayosd.service was enabled here until 2026-07-21 and was the same
	20	+	# mistake. Fedora's SwayOSD 0.3.2 ships three binaries, one SYSTEM unit
	21	+	# (swayosd-libinput-backend.service, now in the system preset) and a
	22	+	# D-Bus service file. There is no user unit, so this line named nothing.
	23	+	# The server is started by `exec swayosd-server` in the sway config,
	24	+	# which is the mechanism rather than a workaround for this line failing.
18	25
19	26		disable syncthing.service
20	27		disable gammastep.service