2f007823

test: the coverage run finishes the suite — three more pids that were the tracer's

a73x 2026-08-26 17:54

Commit message

test: the coverage run finishes the suite — three more pids that were the tracer's

82 of 82 under kcov, for the first time. Each stop was a leg asking the OS
about the daemon through a pid that, under `make coverage`, is the wrapper:

- the upgrade candidate is `cp $MUXD`, and $MUXD is the shim script — an
  exec into it can never show the candidate on /proc/pid/exe. coverage.sh
  names the binary behind the shim (MUXD_ELF; natively it is $MUXD).
- the resumed-argv check read the wrapper's cmdline. `real_pid` is the
  hardkill rule for /proc reads: the tracee, when the pid held is kcov.
- the stop-gone leg's instant `kill -0` must name the pid `stop` itself
  waited on, captured before the stop — kcov's first child at that
  instant can be a reparented zombie, which still answers kill -0.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

test/coverage.sh

Old		New
@@ -140,6 +140,9 @@ export E2E_OK_LOG="$OK_LOG"
140	# told to wait longer — not to demand less. See e2e.sh's TIME_SCALE note.	140	# told to wait longer — not to demand less. See e2e.sh's TIME_SCALE note.
141	export E2E_TIME_SCALE="${COV_TIME_SCALE:-4}"	141	export E2E_TIME_SCALE="${COV_TIME_SCALE:-4}"
142		142
		143	# The upgrade legs copy a candidate binary and watch it exec; a shim cannot be
		144	# exec'd into. The real path goes beside the shim.
		145	MUXD_ELF="$MUXD"; export MUXD_ELF
143	set +e	146	set +e
144	# Each argument is the shim if that binary is wrapped, the real thing if not.	147	# Each argument is the shim if that binary is wrapped, the real thing if not.
145	pick() { if [ -x "$SHIM/$1" ]; then echo "$SHIM/$1"; else echo "$2"; fi; }	148	pick() { if [ -x "$SHIM/$1" ]; then echo "$SHIM/$1"; else echo "$2"; fi; }

test/e2e.sh

Old		New
@@ -3,6 +3,10 @@
3	# output flows pty -> engine -> snapshot -> client replica -> client stdout.	3	# output flows pty -> engine -> snapshot -> client replica -> client stdout.
4	set -eu	4	set -eu
5	MUXD="$1"	5	MUXD="$1"
		6	# The upgrade candidate must be an ELF: the daemon execs it and the leg reads
		7	# the exec'd process's /proc exe link. Under the coverage tracer $MUXD is a
		8	# shim script, so coverage.sh names the binary behind it here.
		9	MUXD_ELF="${MUXD_ELF:-$MUXD}"
6	MUX="$2"	10	MUX="$2"
7	# M9 prediction helpers: a deterministic stand-in for an editor, and a pipe	11	# M9 prediction helpers: a deterministic stand-in for an editor, and a pipe
8	# that makes a slow round trip without netem or root.	12	# that makes a slow round trip without netem or root.
@@ -689,6 +693,17 @@ hardkill() {
689	# its own — measured on the hub leg, where `kill $W3PID` left both alive and	693	# its own — measured on the hub leg, where `kill $W3PID` left both alive and
690	# wait_pid_gone timed out. TERM the tracee; the wrapper exits with it and	694	# wait_pid_gone timed out. TERM the tracee; the wrapper exits with it and
691	# writes its database. Off the tracer this is a plain kill, same exit status.	695	# writes its database. Off the tracer this is a plain kill, same exit status.
		696	# A /proc claim about the daemon is read off the tracee: under `make coverage`
		697	# the pid the suite holds is the tracer's (hardkill's reason), and kcov's own
		698	# cmdline is what a resumed-argv check would otherwise read.
		699	real_pid() {
		700	if [ "$(ps -o comm= -p "$1" 2>/dev/null)" = kcov ]; then
		701	ps -o pid= --ppid "$1" 2>/dev/null \| head -1 \| tr -d ' '
		702	else
		703	echo "$1"
		704	fi
		705	}
		706
692	softkill() {	707	softkill() {
693	if [ "$(ps -o comm= -p "$1" 2>/dev/null)" = kcov ]; then	708	if [ "$(ps -o comm= -p "$1" 2>/dev/null)" = kcov ]; then
694	_rc=1	709	_rc=1
@@ -9234,7 +9249,7 @@ grep -qF "refused: version: $UPVER is not newer than $UPVER" "$OUT.upref1" \|\| {
9234	# `readlink /proc/PID/exe` is the witness that the copy has already	9249	# `readlink /proc/PID/exe` is the witness that the copy has already
9235	# exec'd, so the chmod cannot beat it and turn the leg into an exec	9250	# exec'd, so the chmod cannot beat it and turn the leg into an exec
9236	# failure that asserts nothing.	9251	# failure that asserts nothing.
9237	cp "$MUXD" "$UPBIN"	9252	cp "$MUXD_ELF" "$UPBIN"
9238	chmod 755 "$UPBIN"	9253	chmod 755 "$UPBIN"
9239	kill -STOP "$D69PID"	9254	kill -STOP "$D69PID"
9240	"$UPBIN" upgrade --sock "$SOCK69" --allow-same-version > "$OUT.upref2" 2>&1 &	9255	"$UPBIN" upgrade --sock "$SOCK69" --allow-same-version > "$OUT.upref2" 2>&1 &
@@ -9289,7 +9304,7 @@ grep -qF "muxd: upgraded to $UPVER" "$OUT.upok" \|\| {
9289	kill -0 "$D69PID" 2>/dev/null \|\| {	9304	kill -0 "$D69PID" 2>/dev/null \|\| {
9290	echo "e2e FAIL: upgrade: the daemon pid $D69PID is gone — that is a restart, not an upgrade"	9305	echo "e2e FAIL: upgrade: the daemon pid $D69PID is gone — that is a restart, not an upgrade"
9291	cat "$OUT.up.d"; exit 1; }	9306	cat "$OUT.up.d"; exit 1; }
9292	tr '\0' ' ' < "/proc/$D69PID/cmdline" > "$OUT.upst2" 2>/dev/null \|\| true	9307	tr '\0' ' ' < "/proc/$(real_pid "$D69PID")/cmdline" > "$OUT.upst2" 2>/dev/null \|\| true
9293	grep -qF -- "--resume-fd" "$OUT.upst2" \|\| {	9308	grep -qF -- "--resume-fd" "$OUT.upst2" \|\| {
9294	echo "e2e FAIL: upgrade: pid $D69PID is not running the resumed argv; it holds:"	9309	echo "e2e FAIL: upgrade: pid $D69PID is not running the resumed argv; it holds:"
9295	cat "$OUT.upst2"; exit 1; }	9310	cat "$OUT.upst2"; exit 1; }
@@ -9639,10 +9654,11 @@ sleep 0.5
9639	pipe_detach "stop-gone: the second session's client"	9654	pipe_detach "stop-gone: the second session's client"
9640	[ "$("$MUXD" stats --sock "$SOCK73" \| sed -n 's/.sessions=$[0-9]$.*/\1/p')" = "2" ] \|\| {	9655	[ "$("$MUXD" stats --sock "$SOCK73" \| sed -n 's/.sessions=$[0-9]$.*/\1/p')" = "2" ] \|\| {
9641	echo "e2e FAIL: stop-gone: wanted two sessions of stubborn shells"; exit 1; }	9656	echo "e2e FAIL: stop-gone: wanted two sessions of stubborn shells"; exit 1; }
		9657	D73REAL=$(real_pid "$D73PID")
9642	"$MUXD" stop --sock "$SOCK73" 2> "$OUT.sg.stop" \|\| {	9658	"$MUXD" stop --sock "$SOCK73" 2> "$OUT.sg.stop" \|\| {
9643	echo "e2e FAIL: stop-gone: stop failed"; cat "$OUT.sg.stop"; exit 1; }	9659	echo "e2e FAIL: stop-gone: stop failed"; cat "$OUT.sg.stop"; exit 1; }
9644	# Asked of the OS the instant stop returns — no wait, no retry.	9660	# Asked of the OS the instant stop returns — no wait, no retry.
9645	if kill -0 "$D73PID" 2>/dev/null; then	9661	if kill -0 "$D73REAL" 2>/dev/null; then
9646	echo "e2e FAIL: stop said stopped, but pid $D73PID is still running"; cat "$OUT.sg.stop"; exit 1	9662	echo "e2e FAIL: stop said stopped, but pid $D73PID is still running"; cat "$OUT.sg.stop"; exit 1
9647	fi	9663	fi
9648	grep -q '^muxd: stopped' "$OUT.sg.stop" \|\| {	9664	grep -q '^muxd: stopped' "$OUT.sg.stop" \|\| {