fix(ci): make the clean-machine failures diagnosable, then green (#9)

doctor.sh

@@ -62,6 +62,22 @@ fi
 f=$(systemctl --user --failed --no-legend --plain 2>/dev/null | awk '{print $1}' \
 	| sed 's/^macarchy-failed@\(.*\)\.service$/\1/' | sort -u | paste -sd' ' -)
 [[ -z $f ]] && ok "no failed user units" || bad "failed user units: $f"
+# macarchy-auto-appearance now exits 0 when it cannot compute a sun: a deliberate
+# no-op must not park its unit in --failed for ever (macarchy-core#12). That was
+# also the only thing a MISCONFIGURED laptop had to tell it, so the doctor asks
+# directly instead. Gated on omarchy, because without it the theme switch could
+# not happen at all -- false on a CI runner, true on any real machine.
+if command -v omarchy >/dev/null 2>&1 && command -v macarchy-auto-appearance >/dev/null 2>&1; then
+	aa=$(macarchy-auto-appearance status 2>/dev/null)
+	if [[ $aa == *"enabled=yes"*"error=sun"* || $aa == *"error=sun"*"enabled=yes"* ]]; then
+		bad "auto-appearance is on but cannot compute the sun (set coordinates; see macarchy-sun)"
+	else
+		ok "auto-appearance can decide a theme"
+	fi
+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)"
@@ -134,6 +150,16 @@ if [[ -n ${WAYLAND_DISPLAY:-} ]]; then
 	broken=${broken% }
 	[[ -z $broken ]] && ok "Touch Bar modules loaded" \
 		|| bad "Touch Bar module failed to load: $broken (re-run install.sh in the repo that ships it)"
+	# Same as auto-appearance above: bar-contrast now exits 0 when it has
+	# nothing to sample, so a session that is missing grim/magick/hyprctl/jq
+	# would otherwise say nothing at all. Only meaningful with a session, which
+	# is exactly what this block already has.
+	if command -v macarchy-bar-contrast >/dev/null 2>&1; then
+		bc=$(macarchy-bar-contrast status 2>&1)
+		[[ $bc == *"nothing to sample"* ]] \
+			&& bad "bar-contrast has nothing to sample: ${bc#*: }" \
+			|| ok "bar-contrast can sample the screen"
+	fi
 	pgrep -f "macarchy-als daemon" >/dev/null && ok "macarchy-als"   || bad "macarchy-als not running"
 	pgrep -f "macarchy-pinch"      >/dev/null && ok "macarchy-pinch" || bad "macarchy-pinch not running"
 	# The tank remembers off across reboots (omarchy-aquarium-toggle:30, an

install.sh

@@ -275,6 +275,24 @@ fi
 
 # Ordered before the themes on purpose: its installer only seeds a config when
 # there is none, and the wallpapers it names are the ones apple-glass ships.
+# Moved ahead of macos-dynamic-wallpaper: that step STARTS its unit, and the unit
+# reads the theme's wallpapers out of ~/.config/omarchy/themes/<t>/backgrounds/
+# (macos-dynamic-wallpaper#5). A oneshot that fails once stays failed, so the
+# files have to be on disk BEFORE it first runs. This half needs no omarchy --
+# it is an rsync; the `omarchy theme set` half stays where it was, further down.
+say "Installing the apple-glass themes"
+for t in apple-glass apple-glass-light; do
+	src="$MACARCHY_DIR/$t"
+	dst="$HOME/.config/omarchy/themes/$t"
+	if [[ -d $src ]]; then
+		mkdir -p "$dst"
+		rsync -a --delete --exclude .git "$src/" "$dst/"
+		note "$t synced to ~/.config/omarchy/themes"
+	else
+		warn "$t repo missing"
+	fi
+done
+
 say "Installing macos-dynamic-wallpaper (time-of-day backgrounds)"
 if [[ -x $MACARCHY_DIR/macos-dynamic-wallpaper/install.sh ]]; then
 	(cd "$MACARCHY_DIR/macos-dynamic-wallpaper" && ./install.sh) \
@@ -374,19 +392,6 @@ LUA
 
 # ---------------------------------------------------------------- themes
 
-say "Installing the apple-glass themes"
-for t in apple-glass apple-glass-light; do
-	src="$MACARCHY_DIR/$t"
-	dst="$HOME/.config/omarchy/themes/$t"
-	if [[ -d $src ]]; then
-		mkdir -p "$dst"
-		rsync -a --delete --exclude .git "$src/" "$dst/"
-		note "$t synced to ~/.config/omarchy/themes"
-	else
-		warn "$t repo missing"
-	fi
-done
-
 # The applied theme is a copy; re-apply so edits actually take. On a box
 # that never chose a theme from this pair, apply the dark one.
 if command -v omarchy-theme-set >/dev/null || command -v omarchy >/dev/null; then

tests/clean-machine.sh

@@ -60,6 +60,12 @@ systemctl --user show -p Version --value >/dev/null 2>&1 || {
 	exit 1
 }
 
+# On the way out, whatever the exit status, say WHY every failed user unit failed.
+# It has to be a trap: install.sh runs doctor.sh ITSELF, so a MISS reds the run from
+# inside the `|| exit 1` below and there is no "after install, before doctor" moment
+# to hook. `|| true` and the saved status keep the diagnostics from deciding the run.
+trap 'rc=$?; bash tests/dump-failed-units.sh || true; exit $rc' EXIT
+
 ./install.sh || exit 1   # it runs doctor.sh itself at the end, so a MISS reds here
 sha256sum "$HOME/.config/hypr/autostart.lua" "$HOME/.config/hypr/bindings.lua" >"$MACARCHY_DIR/lua.sha"
 

tests/dump-failed-units.sh

@@ -0,0 +1,47 @@
+#!/bin/bash
+# Say WHY every failed user unit failed. Read-only; prints nothing when nothing failed.
+#
+# clean-machine.sh could only ever report THAT units failed: install.sh runs
+# doctor.sh itself, so a MISS reds the run from inside install.sh and the journals
+# that would explain it are never read. Three units sat unexplained on #9 for a day
+# because of it.
+#
+# Its own script, not a function in clean-machine.sh: that script sudo-writes into
+# /etc and /usr/local/bin, so nothing can exercise it in a hermetic suite, while
+# this is pure stdout over two stubbable commands (tests/test_clean_machine_journal.sh).
+set -uo pipefail
+
+# macarchy-failed@<unit>.service is the NOTIFIER for <unit> -- its own journal says
+# only that it fired. Fold it onto its target, the same sed doctor.sh:63 uses, or
+# the dump explains the messenger instead of the message.
+dump() {
+	echo "=== journal: $1 ==="
+	# `|| true`: a unit with no journal at all must not red the run -- the dump
+	# is diagnostics, never an assertion.
+	journalctl --user -u "$1" --no-pager -n 40 2>&1 || true
+}
+
+systemctl --user --failed --no-legend --plain 2>/dev/null | awk '{print $1}' | sort -u \
+	| while read -r raw; do
+		[[ -n $raw ]] || continue
+		# The target first -- that is the message. But when the NOTIFIER instance
+		# is itself what failed, its own journal is the only one that says why,
+		# and folding it away left a block that explained nothing.
+		# Strip ONLY when the prefix is really there: `${raw%.service}` applied
+		# unconditionally turns a plain `a.service` into `a` and dumps it twice.
+		if [[ $raw == macarchy-failed@* ]]; then
+			target=${raw#macarchy-failed@}; target=${target%.service}
+		else
+			target=$raw
+		fi
+		dump "$target"
+		# An `if`, not `[[ … ]] && dump`: as the LAST statement of the loop body
+		# that compound returns 1 whenever the condition is false, which becomes
+		# the pipeline's status and then the script's. This is diagnostics; it
+		# must never be the reason a run goes red.
+		if [[ $raw != "$target" ]]; then
+			dump "$raw"
+		fi
+	done
+
+exit 0

tests/test_clean_machine_journal.sh

@@ -0,0 +1,72 @@
+#!/bin/bash
+# tests/test_clean_machine_journal.sh — the failed-unit journal dump.
+#
+# clean-machine.sh reports THAT user units failed and never why: it is install.sh
+# that runs doctor.sh, so its MISS reds the run from inside, and the journals that
+# would explain it are never read. Three units sat unexplained on #9 for exactly
+# that reason.
+#
+# The seam is tests/dump-failed-units.sh's stdout, NOT clean-machine.sh's: that
+# script sudo-writes into /etc and /usr/local/bin and enables linger, so it can
+# never run in a hermetic suite. The dump lives in its own script so it CAN.
+set -uo pipefail
+cd "$(dirname "$0")/.."
+
+TMP=$(mktemp -d); trap 'rm -rf "$TMP"' EXIT
+mkdir -p "$TMP/bin"; export PATH="$TMP/bin:$PATH"
+
+fails=0
+check() { local name=$1; shift; if "$@"; then echo "ok   $name"; else echo "FAIL $name"; fails=$((fails+1)); fi; }
+
+# systemctl replays whatever the case wrote; journalctl echoes which unit it was asked for.
+cat > "$TMP/bin/systemctl" <<'STUB'
+#!/bin/bash
+[[ $* == *--failed* ]] || exit 1
+cat "$TMP_FAILED"
+STUB
+cat > "$TMP/bin/journalctl" <<'STUB'
+#!/bin/bash
+for a in "$@"; do [[ $prev == -u ]] && echo "MARKER body of $a"; prev=$a; done
+STUB
+chmod +x "$TMP/bin"/*
+export TMP_FAILED="$TMP/failed"
+
+# --- (a) one failed unit: its name AND its journal reach stdout ---------------
+echo 'macarchy-auto-appearance.service loaded failed failed Some daemon' > "$TMP_FAILED"
+out=$(./tests/dump-failed-units.sh 2>&1)
+check "(a) names the unit"        grep -q 'macarchy-auto-appearance.service' <<<"$out"
+check "(a) carries its journal"   grep -q 'MARKER body of macarchy-auto-appearance.service' <<<"$out"
+
+# --- (b) nothing failed: no dump at all ---------------------------------------
+: > "$TMP_FAILED"
+out=$(./tests/dump-failed-units.sh 2>&1)
+check "(b) silence when nothing failed" [ -z "$out" ]
+
+# --- (c) the notifier instance folds onto its target, like doctor.sh:63 -------
+# macarchy-failed@macarchy-touchbar.service.service is the NOTIFIER for
+# macarchy-touchbar.service. Dumping the notifier's own journal says nothing;
+# doctor.sh already folds the pair into one name and so must this.
+echo 'macarchy-failed@macarchy-touchbar.service.service loaded failed failed Notifier' > "$TMP_FAILED"
+out=$(./tests/dump-failed-units.sh 2>&1)
+check "(c) folds onto the target"      grep -q 'MARKER body of macarchy-touchbar.service' <<<"$out"
+# ...and the notifier's OWN journal too: when the instance is what failed, that
+# is the only place saying why. The target still comes first -- it is the message.
+check "(c) dumps the notifier as well" grep -q 'MARKER body of macarchy-failed@macarchy-touchbar.service.service' <<<"$out"
+check "(c) target before notifier"     bash -c '[ "$(grep -n "MARKER body of macarchy-touchbar.service" <<<"$1" | cut -d: -f1)" -lt "$(grep -n "MARKER body of macarchy-failed@" <<<"$1" | cut -d: -f1)" ]' _ "$out"
+
+# --- (d) several failed units: one block each ---------------------------------
+printf '%s\n' 'a.service loaded failed failed A' 'b.service loaded failed failed B' > "$TMP_FAILED"
+out=$(./tests/dump-failed-units.sh 2>&1)
+check "(d) one block per unit" [ "$(grep -c 'MARKER body of' <<<"$out")" = 2 ]
+
+# --- (e) a unit with no journal must not kill the run -------------------------
+echo 'c.service loaded failed failed C' > "$TMP_FAILED"
+printf '#!/bin/bash\nexit 1\n' > "$TMP/bin/journalctl"   # journalctl fails outright
+./tests/dump-failed-units.sh >/dev/null 2>&1; rc=$?
+# `$?` must be captured FIRST: inside `check "..." [ "$?" = 0 ]` it expands while
+# the argument list is built, so the assertion would silently retarget to
+# whatever command happened to run just before it.
+check "(e) survives a journalctl failure" [ "$rc" = 0 ]
+
+(( fails == 0 )) && echo "all ok" || echo "$fails failed"
+exit $(( fails > 0 ))

tests/test_doctor_noop_signals.sh

@@ -0,0 +1,81 @@
+#!/bin/bash
+# tests/test_doctor_noop_signals.sh — the checks that replaced a failed unit.
+#
+# macarchy-core#12 made macarchy-auto-appearance and macarchy-bar-contrast exit 0
+# when they deliberately do nothing, so a oneshot no longer parks itself in
+# `systemctl --user --failed` for ever. That was also the only signal a
+# MISCONFIGURED laptop had, so the doctor has to ask the two tools directly --
+# and a check that can only ever say "ok" would be worse than the unit it
+# replaced. These cases exist to prove each one can still say MISS.
+set -uo pipefail
+cd "$(dirname "$0")/.."
+
+TMP=$(mktemp -d); trap 'rm -rf "$TMP"' EXIT
+export HOME="$TMP/home" XDG_STATE_HOME="$TMP/state"
+mkdir -p "$HOME" "$TMP/bin"; export PATH="$TMP/bin:$PATH"
+
+fails=0
+check() { local name=$1; shift; if "$@"; then echo "ok   $name"; else echo "FAIL $name"; fails=$((fails+1)); fi; }
+
+for stub in pgrep omarchy-aquarium-toggle systemctl journalctl; do
+	printf '#!/bin/bash\nexit 1\n' > "$TMP/bin/$stub"
+done
+printf '#!/bin/bash\nexit 0\n' > "$TMP/bin/omarchy"
+printf '#!/bin/bash\ncat "$AA_STATUS"\n'  > "$TMP/bin/macarchy-auto-appearance"
+printf '#!/bin/bash\ncat "$BC_STATUS"\n'  > "$TMP/bin/macarchy-bar-contrast"
+chmod +x "$TMP/bin"/*
+export AA_STATUS="$TMP/aa" BC_STATUS="$TMP/bc"
+echo 'sample=#003575 verdict=light' > "$BC_STATUS"
+
+# --- auto-appearance: on but sunless is a MISS ------------------------------
+echo 'mode=solar enabled=yes error=sun' > "$AA_STATUS"
+out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
+check "sunless-but-enabled is a MISS"  grep -q 'MISS.*cannot compute the sun' <<<"$out"
+check "and it says how to fix it"      grep -q 'set coordinates' <<<"$out"
+
+# --- auto-appearance: a normal answer is ok ---------------------------------
+echo 'mode=solar enabled=yes want=dark sunrise=06:57 sunset=20:15' > "$AA_STATUS"
+out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
+check "a working sun is ok"            grep -q 'ok.*auto-appearance can decide' <<<"$out"
+check "and not also a MISS"            bash -c '! grep -q "cannot compute the sun" <<<"$1"' _ "$out"
+
+# --- auto-appearance: disabled is not a MISS --------------------------------
+# error=sun with the timer OFF is nobody's problem: the user turned it off.
+echo 'mode=solar enabled=no error=sun' > "$AA_STATUS"
+out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
+check "sunless but switched off is not a MISS" bash -c '! grep -q "cannot compute the sun" <<<"$1"' _ "$out"
+
+# --- no omarchy: skipped with a reason, never a silent pass -----------------
+# Removing the stub is not enough: the developer running this suite HAS a real
+# omarchy on PATH, and `command -v` would find it whatever $TMP/bin holds. So
+# build a PATH with every directory containing an omarchy dropped -- on a CI
+# runner that is a no-op, on a laptop it is the whole point of the case.
+without_omarchy() {
+	local d out= dirs
+	IFS=: read -ra dirs <<<"$PATH"
+	for d in "${dirs[@]}"; do
+		[[ -x $d/omarchy ]] && continue
+		out="${out:+$out:}$d"
+	done
+	printf '%s' "$out"
+}
+echo 'mode=solar enabled=yes error=sun' > "$AA_STATUS"
+rm -f "$TMP/bin/omarchy"
+out=$(PATH="$(without_omarchy)" WAYLAND_DISPLAY= ./doctor.sh 2>&1)
+check "no omarchy skips, not passes"   grep -q 'skip.*auto-appearance sun' <<<"$out"
+check "and names why"                  grep -q 'no omarchy on this machine' <<<"$out"
+check "and does not MISS"              bash -c '! grep -q "cannot compute the sun" <<<"$1"' _ "$out"
+printf '#!/bin/bash\nexit 0\n' > "$TMP/bin/omarchy"; chmod +x "$TMP/bin/omarchy"
+
+# --- bar-contrast: nothing to sample, inside a session, is a MISS -----------
+echo 'macarchy-bar-contrast: grim missing; nothing to sample' > "$BC_STATUS"
+out=$(WAYLAND_DISPLAY=wayland-0 ./doctor.sh 2>&1)
+check "nothing to sample is a MISS"    grep -q 'MISS.*bar-contrast has nothing to sample' <<<"$out"
+check "and quotes the reason"          grep -q 'grim missing' <<<"$out"
+
+# --- bar-contrast: no session, no claim -------------------------------------
+out=$(WAYLAND_DISPLAY= ./doctor.sh 2>&1)
+check "no session: bar-contrast unasked" bash -c '! grep -q "bar-contrast" <<<"$1"' _ "$out"
+
+(( fails == 0 )) && echo "all ok" || echo "$fails failed"
+exit $(( fails > 0 ))