Skip to content

fix(install): make the failure notifier reachable from the first failure (#12) #13

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
phmatray merged 3 commits into from
Sep 4, 2026
+165 −5
Merged

fix(install): make the failure notifier reachable from the first failure (#12) #13

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
03431e7
chore(#12): scaffold draft PR for the notifier ordering fix
phmatray Sep 4, 2026
01d9422
fix(install): make the failure notifier reachable from the first failure
phmatray Sep 4, 2026
4b1f7b7
fix: address code-review findings
phmatray Sep 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
54 changes: 51 additions & 3 deletions doctor.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,9 +78,57 @@ else
skip "auto-appearance sun (no omarchy on this machine)"
fi

# A muted check is worse than none, so the doctor checks that it can speak.
[[ -e $HOME/.config/systemd/user/macarchy-failed@.service ]] \
&& ok "failure notifier" || bad "failure notifier (macarchy-install/install.sh)"
# A muted check is worse than none, so the doctor checks that it can speak. Testing
# that the template FILE exists is not that test: it passed on every first install
# while the units naming it were told "Unit macarchy-failed@….service not found",
# because the file arrived after they did (#12). Ask the units instead.
#
# Three things this deliberately does NOT do, each of which it did in review:
# * scan every unit on the machine. doctor.sh:2 scopes this file to "the pieces
# install.sh puts in place"; jarvis, voxtype and aikit-sync also live in that
# directory, and blaming macarchy-install for a template THEY fail to ship is
# a false MISS that reds CI and points the user at the wrong installer.
# * look only under ~/.config. systemd resolves user units from /etc/systemd/user
# and /usr/lib/systemd/user too -- which is exactly where a PACKAGED notifier
# would land, so a package would otherwise make this check cry wolf at login.
# * read only the first target on the line. `OnFailure=a@%n.service b@%n.service`
# is legal, and stopping at the first is the silent pass this check exists to end.
_unit_dirs=("$HOME/.config/systemd/user" /etc/systemd/user /usr/lib/systemd/user)
_resolves() { # _resolves <unit-or-template-name>
local d; for d in "${_unit_dirs[@]}"; do [[ -e $d/$1 ]] && return 0; done; return 1
}
declare -A _want=()
for _d in "${_unit_dirs[@]}"; do
[[ -d $_d ]] || continue
# macarchy's own units only -- every unit type, and drop-ins, not just *.service.
for _u in "$_d"/macarchy-*.{service,timer,path,socket} "$_d"/macos-dynamic-wallpaper*.{service,timer} \
"$_d"/macarchy-*.d/*.conf; do
[[ -f $_u ]] || continue # no glob match
while read -r _line; do
for _tgt in $_line; do # unquoted: split the list
# a@%n.service -> a@.service; a plain name stays itself
[[ $_tgt == *@* ]] && _tgt="${_tgt%@*}@.service"
_want["$_tgt"]="${_want["$_tgt"]:+${_want["$_tgt"]} }$(basename "$_u")"
done
done < <(sed -nE 's/^[[:space:]]*OnFailure=[[:space:]]*(.*)/\1/p' "$_u")
done
done
# The floor: install.sh installs the template unconditionally, so its absence is a
# MISS even when nothing happens to name it -- otherwise a run whose clones failed
# leaves no declaring units and the check congratulates itself.
_miss=()
_resolves "macarchy-failed@.service" || _miss+=("macarchy-failed@.service is not installed")
for _tmpl in "${!_want[@]}"; do
_resolves "$_tmpl" || _miss+=("$_tmpl needed by ${_want[$_tmpl]}")
done
if (( ${#_miss[@]} )); then
bad "failure notifier: ${_miss[*]} (macarchy-install/install.sh)"
elif (( ${#_want[@]} == 0 )); then
ok "failure notifier installed (no macarchy unit declares OnFailure= yet)"
else
_n=0; for _tmpl in "${!_want[@]}"; do read -ra _u <<<"${_want[$_tmpl]}"; _n=$((_n + ${#_u[@]})); done
ok "failure notifier ($_n macarchy unit$( ((_n>1)) && printf s) declare$( ((_n==1)) && printf s) it, all reachable)"
fi
systemctl --user is-enabled -q macarchy-doctor.service 2>/dev/null \
&& ok "login self-check enabled" || bad "login self-check (macarchy-install/install.sh)"
b=/sys/class/power_supply/macsmc-battery/charge_control_end_threshold
Expand Down
24 changes: 22 additions & 2 deletions install.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -231,6 +231,27 @@ sudo pacman -S --needed --noconfirm \

# ----------------------------------------------------------------- repos

# Before every component, because three of them name it and none of them ships it:
# macarchy-touchbar.service, macarchy-auto-appearance.service and
# macarchy-bar-contrast.service all declare `OnFailure=macarchy-failed@%n.service`.
# It used to be installed with macarchy-doctor.service in the health section near
# the end -- a tidy grouping and the wrong constraint -- so on a FIRST install
# every failure before that point was dropped with "Failed to enqueue OnFailure=
# job, ignoring: Unit macarchy-failed@….service not found", and the one thing the
# notifier exists to catch (a daemon that dies silently) died silently. #12.
#
# A static file in this repo: no clone, no omarchy, no session. Nothing here has
# to wait for anything.
say "Installing the failure notifier the daemons name"
mkdir -p "$HOME/.config/systemd/user"
install -m644 systemd/macarchy-failed@.service "$HOME/.config/systemd/user/" \
&& note "macarchy-failed@.service in place before anything can name it" \
|| warn "could not install the failure notifier template"
# Not left to systemd's unit-directory mtime rescan: it is reliable in practice,
# but the whole point of this block is an ordering GUARANTEE, and one reload makes
# it hold unconditionally. The doctor step further down does the same after its own.
systemctl --user daemon-reload 2>/dev/null || true

say "Fetching the macarchy repos into $MACARCHY_DIR"
mkdir -p "$MACARCHY_DIR"
for r in "${REPOS[@]}"; do
Expand Down Expand Up @@ -422,8 +443,7 @@ say "Wiring the machine's own health report"
# it is worth having on PATH by hand anyway.
mkdir -p "$HOME/.local/bin" "$HOME/.config/systemd/user"
install -m755 doctor.sh "$HOME/.local/bin/macarchy-doctor"
install -m644 systemd/macarchy-doctor.service systemd/macarchy-failed@.service \
"$HOME/.config/systemd/user/"
install -m644 systemd/macarchy-doctor.service "$HOME/.config/systemd/user/"
systemctl --user daemon-reload
# enable, not --now: firing it here would grade the half-installed state it is
# standing in, and the tail of this script already runs ./doctor.sh once.
Expand Down
92 changes: 92 additions & 0 deletions tests/test_notifier_order.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,92 @@
#!/bin/bash
# tests/test_notifier_order.sh — the failure notifier has to exist before anything
# can name it, and the doctor has to be able to tell reachable from merely present.
#
# Three units declare `OnFailure=macarchy-failed@%n.service`: macarchy-touchbar,
# macarchy-auto-appearance and macarchy-bar-contrast. All three are installed by
# their own repos, from the component steps in the middle of install.sh. The
# template that answers them used to be installed in the second-to-last section,
# so on a FIRST install every failure before that point was dropped with
# Failed to enqueue OnFailure= job, ignoring: Unit macarchy-failed@….service not found
# while doctor.sh reported `ok failure notifier` throughout -- it only tested that
# the file existed. #12.
set -uo pipefail
cd "$(dirname "$0")/.."

fails=0
check() { local name=$1; shift; if "$@"; then echo "ok $name"; else echo "FAIL $name"; fails=$((fails+1)); fi; }
lineno() { grep -n "$1" install.sh | head -1 | cut -d: -f1; }

# --- ordering, read off install.sh itself ------------------------------------
notifier=$(lineno 'systemd/macarchy-failed@\.service')
sanity=$(lineno '^say "Checking this machine"')
repos=$(lineno '^say "Fetching the macarchy repos')
core=$(lineno '^say "Installing the macarchy-core suite')

check "the notifier install is found" [ -n "$notifier" ]
check "it comes after the sanity checks" [ "$notifier" -gt "$sanity" ]
check "it comes BEFORE the repo clone" [ "$notifier" -lt "$repos" ]
check "and before any component install" [ "$notifier" -lt "$core" ]

# --- the doctor's check, exercised hermetically ------------------------------
TMP=$(mktemp -d); trap 'rm -rf "$TMP"' EXIT
export HOME="$TMP/home" XDG_STATE_HOME="$TMP/state"
UNITS="$HOME/.config/systemd/user"; mkdir -p "$UNITS" "$TMP/bin"
export PATH="$TMP/bin:$PATH"
for stub in pgrep omarchy-aquarium-toggle systemctl journalctl omarchy; do
printf '#!/bin/bash\nexit 1\n' > "$TMP/bin/$stub"
done
chmod +x "$TMP/bin"/*

T="$UNITS/macarchy-failed@.service"

# A macarchy unit that names the notifier, with no template beside it.
printf '[Unit]\nOnFailure=macarchy-failed@%%n.service\n' > "$UNITS/macarchy-probe.service"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "a named-but-absent notifier is a MISS" grep -q 'MISS.*failure notifier' <<<"$out"
check "and it says which unit needs it" grep -q 'macarchy-probe' <<<"$out"

# Same unit, template present.
printf '[Unit]\nDescription=notifier\n' > "$T"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "reachable notifier is ok" grep -q 'ok.*failure notifier' <<<"$out"
check "and it says what it verified" grep -qE 'failure notifier \(1 macarchy unit' <<<"$out"

# A SPACE-SEPARATED list: systemd allows it, and stopping at the first target is
# the silent pass this check exists to end.
printf '[Unit]\nOnFailure=macarchy-failed@%%n.service other-notify@%%n.service\n' > "$UNITS/macarchy-probe.service"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "the second target is checked too" grep -q 'other-notify@.service needed by' <<<"$out"

# A THIRD-PARTY unit is none of macarchy's business: jarvis, voxtype and aikit-sync
# all live in this directory, and blaming macarchy-install for a template THEY do
# not ship is a false MISS that reds CI and names the wrong installer.
printf '[Unit]\nOnFailure=macarchy-failed@%%n.service\n' > "$UNITS/macarchy-probe.service"
printf '[Unit]\nOnFailure=somebody-elses@%%n.service\n' > "$UNITS/voxtype.service"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "a third-party unit is not our MISS" bash -c '! grep -q "somebody-elses" <<<"$1"' _ "$out"
check "and ours still reports ok" grep -q 'ok.*failure notifier' <<<"$out"
rm -f "$UNITS/voxtype.service"

# A TIMER declaring it counts: this suite ships macarchy-*.timer units.
printf '[Unit]\nOnFailure=macarchy-failed@%%n.service\n' > "$UNITS/macarchy-probe.timer"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "a .timer is scanned too" grep -qE 'failure notifier \(2 macarchy units' <<<"$out"
rm -f "$UNITS/macarchy-probe.timer"

# Leading whitespace is legal in a unit file.
printf '[Unit]\n OnFailure=macarchy-failed@%%n.service\n' > "$UNITS/macarchy-probe.service"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "indented OnFailure= is parsed" grep -qE 'failure notifier \(1 macarchy unit' <<<"$out"

# Nothing declares it -- but the template's own absence is STILL a MISS. Otherwise a
# run whose repo clones failed leaves no units and the check congratulates itself.
rm -f "$UNITS/macarchy-probe.service"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "no declarations, template present: ok" grep -q 'no macarchy unit declares' <<<"$out"
rm -f "$T"
out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
check "no declarations, no template: MISS" grep -q 'MISS.*is not installed' <<<"$out"

(( fails == 0 )) && echo "all ok" || echo "$fails failed"
exit $(( fails > 0 ))
Loading