0c377dce

tools: harden valgrind-quic.sh failure paths

a73x 2026-08-14 09:36

Commit message

tools: harden valgrind-quic.sh failure paths

Four fixes from review, all on paths that only matter when something
has already gone wrong:

- stop is now `|| kill $VGPID || true` — a crashed-mid-run daemon no
  longer lets errexit delete vg.log unprinted, and a stop-ignoring live
  daemon gets killed instead of hanging the wait.
- INT/TERM traps now `exit`, not just clean up — Ctrl-C mid-bind-loop
  used to kill valgrind and rm $TMP while the loop kept polling the
  now-deleted socket path for up to a minute.
- the bind loop checks `kill -0 "$VGPID"` each tick and cats the log on
  timeout, so a daemon that dies at startup under valgrind fails fast
  instead of waiting out the full 60s.
- --error-exitcode=99 is no longer swallowed by `wait || true`; the
  valgrind exit code is captured and printed ahead of the summary.

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

tools/valgrind-quic.sh

Old		New
@@ -18,8 +18,10 @@ command -v valgrind >/dev/null \|\| { echo "valgrind is not installed"; exit 1; }
18	[ -x "$MUXD" ] && [ -x "$MUXA" ] \|\| { echo "binaries missing — run zig build"; exit 1; }	18	[ -x "$MUXD" ] && [ -x "$MUXA" ] \|\| { echo "binaries missing — run zig build"; exit 1; }
19	TMP="${TMPDIR:-/tmp}/mux-valgrind-$$"	19	TMP="${TMPDIR:-/tmp}/mux-valgrind-$$"
20	mkdir -p "$TMP"	20	mkdir -p "$TMP"
21	trap 'kill "$VGPID" 2>/dev/null; rm -rf "$TMP"' EXIT INT TERM
22	VGPID=""	21	VGPID=""
		22	trap '[ -n "$VGPID" ] && kill "$VGPID" 2>/dev/null \|\| true; rm -rf "$TMP"' EXIT
		23	trap 'exit 130' INT
		24	trap 'exit 143' TERM
23	# Hermetic key: keygen into a private XDG home, --key it explicitly.	25	# Hermetic key: keygen into a private XDG home, --key it explicitly.
24	XDG_CONFIG_HOME="$TMP/cfg"; export XDG_CONFIG_HOME	26	XDG_CONFIG_HOME="$TMP/cfg"; export XDG_CONFIG_HOME
25	"$MUXD" keygen > /dev/null	27	"$MUXD" keygen > /dev/null
@@ -30,17 +32,22 @@ valgrind --leak-check=full --error-exitcode=99 --log-file="$TMP/vg.log" \
30	"$MUXD" run --sock "$SOCK" --shell /bin/sh \	32	"$MUXD" run --sock "$SOCK" --shell /bin/sh \
31	--quic "127.0.0.1:$PORT" --key "$KEY" &	33	--quic "127.0.0.1:$PORT" --key "$KEY" &
32	VGPID=$!	34	VGPID=$!
33	# valgrind start is SLOW; give the bind a full minute.	35	# valgrind start is SLOW; give the bind a full minute. Bail early if the
		36	# daemon dies under valgrind instead of waiting out the whole timeout.
34	i=0	37	i=0
35	while [ ! -S "$SOCK" ] && [ "$i" -lt 600 ]; do sleep 0.1; i=$((i + 1)); done	38	while [ ! -S "$SOCK" ] && [ "$i" -lt 600 ]; do
36	[ -S "$SOCK" ] \|\| { echo "daemon never bound under valgrind"; exit 1; }	39	kill -0 "$VGPID" 2>/dev/null \|\| break
		40	sleep 0.1; i=$((i + 1))
		41	done
		42	[ -S "$SOCK" ] \|\| { echo "daemon never bound under valgrind:"; cat "$TMP/vg.log" 2>/dev/null; exit 1; }
37	# One real command over QUIC — handshake, frames, teardown — with a	43	# One real command over QUIC — handshake, frames, teardown — with a
38	# timeout sized for valgrind's clock, then a clean stop so every exit	44	# timeout sized for valgrind's clock, then a clean stop so every exit
39	# path (and the allocator teardown) runs.	45	# path (and the allocator teardown) runs.
40	"$MUXA" run --quic "127.0.0.1:$PORT" --key "$KEY" --timeout 60000 "echo vg-probe" \	46	"$MUXA" run --quic "127.0.0.1:$PORT" --key "$KEY" --timeout 60000 "echo vg-probe" \
41	\|\| echo "muxa run failed under valgrind (timing?) — the leak summary below still stands"	47	\|\| echo "muxa run failed under valgrind (timing?) — the leak summary below still stands"
42	"$MUXD" stop --sock "$SOCK"	48	"$MUXD" stop --sock "$SOCK" \|\| kill "$VGPID" 2>/dev/null \|\| true
43	wait "$VGPID" \|\| true	49	wait "$VGPID" && VGRC=0 \|\| VGRC=$?
44	VGPID=""	50	VGPID=""
		51	echo "valgrind exit: $VGRC (99 = valgrind-detected errors)"
45	echo "---- valgrind summary ----"	52	echo "---- valgrind summary ----"
46	grep -E "definitely lost\|indirectly lost\|ERROR SUMMARY" "$TMP/vg.log" \|\| cat "$TMP/vg.log"	53	grep -E "definitely lost\|indirectly lost\|ERROR SUMMARY" "$TMP/vg.log" \|\| cat "$TMP/vg.log"