Bridge warden-ui's debug channel out of the qemu VM

The rig could drive the UI and detect one outcome: the process died. Nothing
could ask the UI what page it was on or what a tap would land on, because
that channel is a FIFO inside the guest and the initramfs is busybox-only
with no sshd. Scenarios therefore asserted nothing and screenshots went
unread.

run.sh --ctl exposes a second pci-serial port as a unix socket, the same
device the RS485 bridge already rides, listed first so it is always ttyS0.
It also puts warden.ctl on the kernel command line, and init bridges only
when that marker is present: a VM launched with --rs485 alone has a ttyS0
too, and that one is the Modbus wire. The bridge relays one command line in
and the FIFO's reply out, then a sentinel so the reader needs no timeout.

qmp.py gains the channel verbs (nav, page, stats, hit, assert_page,
assert_hit), records every step to results.jsonl as ok/fail/fatal, continues
past an assertion mismatch so one run reports every broken expectation, and
checks the console after EVERY step for the stage-2 init's EXITED line so a
crash is pinned to the step that caused it. assert_hit matches the widget's
bounding box: an icon has no usable caption and two list rows share a class,
but the geometry the UI itself resolved is exact.

The vocabulary is what tools/warden-ctl already speaks over SSH to a real
panel, so a script that runs here runs there. Verified end to end on the rig
(11/11 verbs round-tripped) and against the bench panel, where the same
commands returned byte-identical results.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013aHKWzT5EF86RFKRMtAv9n
This commit is contained in:
Noah
2026-09-07 21:54:52 -06:00
co-authored by Claude Fable 5.1
parent 3ea6c837e3
commit 8a57057053
4 changed files with 278 additions and 13 deletions
+183 -7
View File
@@ -20,10 +20,33 @@ coordinates are PANEL PIXELS (0..size-1, default 720) rather than the
sleep SECONDS let animations settle or timers fire
echo TEXT progress marker in the driver's own output
With `--ctl SOCK` (the unix socket run.sh --ctl creates; ui-drive.sh passes
it) the script can also ask the UI itself, through the debug channel in
warden_debug.c. The same channel is what tools/warden-ctl reaches over SSH on
a real panel, so these verbs mean the same thing on the rig and on hardware:
nav MENU[/TAB] jump to a page; fails if the page is unknown
page | stats | hit X Y print the reply, judge nothing
ctl WORDS... raw passthrough for anything the channel grows
assert_page MENU/TAB the active page is exactly this
assert_hit X Y CLASS [TEXT...]
a real tap at X,Y lands on an object of CLASS whose
caption contains TEXT: the pre-tap self-check that
reports a MOVED button as such, not as broken
behaviour
Every step is recorded to <outdir>/results.jsonl as ok / fail / fatal. An
assertion mismatch is a `fail` and the script continues, so one run reports
every broken expectation. With `--console LOG` the console is checked after
EVERY step for the stage-2 init's "warden-ui EXITED" line, so a crash is a
`fatal` pinned to the step that caused it rather than a stale picture found at
the end. The exit status is non-zero if anything failed.
A step that names an unknown command is a FATAL error rather than a skip: a
silently-ignored line in a scenario is a test that proves nothing.
"""
import json
import re
import socket
import sys
import time
@@ -98,12 +121,106 @@ def do_swipe(s, f, x1, y1, x2, y2, size, ms=400, steps=None):
send_events(s, f, [btn_ev(False)])
def drive(s, f, script_path, outdir, size):
class Ctl:
"""The UI's debug channel, reached through run.sh --ctl.
Inside the VM that channel is a FIFO polled by warden-ui (warden_debug.c);
rootfs/sbin/init bridges it to a pci-serial port that run.sh exposes as a
unix socket. One command line in, the reply out, then a sentinel line. On
real hardware flare-edge tools/warden-ctl reaches the identical FIFO over
SSH, so the vocabulary here is the vocabulary there: a flow script that
runs against this rig runs against a panel.
A missing or silent channel is an infrastructure fault and raises; it is
never reported as an assertion failure, because a rig that cannot ask is
not evidence about the UI either way.
"""
SENTINEL = "<<END>>"
def __init__(self, path, timeout=15.0):
self.sock = socket.socket(socket.AF_UNIX)
self.sock.settimeout(timeout)
self.sock.connect(path)
self.buf = b""
def send(self, cmd):
self.sock.sendall((cmd + "\n").encode())
lines = []
while True:
nl = self.buf.find(b"\n")
if nl < 0:
chunk = self.sock.recv(4096)
if not chunk:
raise RuntimeError("control channel closed")
self.buf += chunk
continue
line = self.buf[:nl].decode("utf-8", "replace").rstrip("\r")
self.buf = self.buf[nl + 1:]
if line == self.SENTINEL:
break
# The tty is in cooked mode, so the command comes back echoed.
if line.strip() == cmd.strip():
continue
lines.append(line)
return "\n".join(lines).strip()
HIT_RE = re.compile(r'^hit \d+,\d+: (\S+)(?: text="(.*)" box=(-?\d+),(-?\d+),(\d+)x(\d+))?$')
def parse_hit(reply):
"""-> (cls, text, box) with box as 'x1,y1,WxH', or None for 'nothing';
raises on an unparseable reply."""
m = HIT_RE.match(reply)
if not m:
raise RuntimeError(f"unparseable hit reply: {reply!r}")
if m.group(1) == "nothing":
return None
box = f"{m.group(3)},{m.group(4)},{m.group(5)}x{m.group(6)}" if m.group(3) else ""
return m.group(1), m.group(2) or "", box
def ui_exited(console_path):
"""The stage-2 init announces a dead UI on the console; the framebuffer
does not, it just keeps the last frame. Checked after EVERY step so a crash
is pinned to the step that caused it, not discovered at the end."""
if not console_path:
return None
try:
with open(console_path, "rb") as fh:
data = fh.read()
except OSError:
return None
i = data.find(b"warden-ui EXITED")
if i < 0:
return None
return data[i:i + 80].split(b"\n", 1)[0].decode("utf-8", "replace")
def drive(s, f, script_path, outdir, size, ctl_path=None, console_path=None):
import os
os.makedirs(outdir, exist_ok=True)
with open(script_path) as fh:
lines = fh.readlines()
ctl = Ctl(ctl_path) if ctl_path else None
results = open(os.path.join(outdir, "results.jsonl"), "w")
counts = {"ok": 0, "fail": 0, "fatal": 0}
def record(lineno, cmd, status, detail=""):
counts[status] += 1
results.write(json.dumps({"line": lineno, "cmd": cmd, "status": status,
"detail": detail}) + "\n")
results.flush()
tag = {"ok": "", "fail": "FAIL ", "fatal": "FATAL "}[status]
print(f" [{lineno}] {tag}{cmd}{(' -- ' + detail) if detail else ''}", flush=True)
def need_ctl(lineno, cmd):
if ctl is None:
sys.exit(f"FATAL: {script_path}:{lineno}: '{cmd}' needs the control "
f"channel; run with --ctl (ui-drive.sh passes it)")
for lineno, raw in enumerate(lines, 1):
line = raw.split("#", 1)[0].strip()
if not line:
@@ -115,27 +232,84 @@ def drive(s, f, script_path, outdir, size):
name = args[0]
path = os.path.join(os.path.abspath(outdir), f"{name}.ppm")
rpc(s, f, {"execute": "screendump", "arguments": {"filename": path}})
print(f" [{lineno}] shot {name}", flush=True)
record(lineno, line, "ok")
elif cmd == "tap":
x, y = int(args[0]), int(args[1])
do_tap(s, f, to_axis(x, size), to_axis(y, size))
print(f" [{lineno}] tap {x},{y}", flush=True)
record(lineno, line, "ok")
elif cmd == "swipe":
ms = int(args[4]) if len(args) > 4 else 400
do_swipe(s, f, int(args[0]), int(args[1]), int(args[2]), int(args[3]), size, ms)
print(f" [{lineno}] swipe {args[0]},{args[1]} -> {args[2]},{args[3]}", flush=True)
record(lineno, line, "ok")
elif cmd == "fling":
do_swipe(s, f, int(args[0]), int(args[1]), int(args[2]), int(args[3]),
size, ms=120)
print(f" [{lineno}] fling {args[0]},{args[1]} -> {args[2]},{args[3]}", flush=True)
record(lineno, line, "ok")
elif cmd == "sleep":
time.sleep(float(args[0]))
print(f" [{lineno}] sleep {args[0]}", flush=True)
record(lineno, line, "ok")
elif cmd == "echo":
print(f" [{lineno}] {' '.join(args)}", flush=True)
elif cmd == "nav":
need_ctl(lineno, cmd)
reply = ctl.send("nav " + " ".join(args))
record(lineno, line, "ok" if reply.endswith(": ok") else "fail", reply)
elif cmd in ("page", "hit", "stats", "ctl"):
# Query verbs: print the reply, never judge it. `ctl` is a raw
# passthrough for anything the channel grows later.
need_ctl(lineno, cmd)
reply = ctl.send(line if cmd != "ctl" else " ".join(args))
record(lineno, line, "ok", reply)
elif cmd == "assert_page":
need_ctl(lineno, cmd)
want = " ".join(args)
got = ctl.send("page")
record(lineno, line, "ok" if got == want else "fail",
f"page is {got!r}" if got != want else "")
elif cmd == "assert_hit":
# assert_hit X Y CLASS [box=X1,Y1,WxH] [TEXT...]: a real tap at X,Y
# would land on an object of CLASS, with that exact bounding box if
# one is given, whose caption contains TEXT. This is the pre-tap
# self-check: "the button moved" fails HERE, by name, so it is
# never mistaken for the behaviour behind the button.
#
# The box is the strong identity. Two rows of a list share a class,
# and an icon's caption is a glyph nobody wants in a spec; the
# geometry the UI itself reports is what tells one instance from
# another.
need_ctl(lineno, cmd)
x, y, cls = args[0], args[1], args[2]
rest = args[3:]
want_box = None
if rest and rest[0].startswith("box="):
want_box = rest[0][4:]
rest = rest[1:]
want_text = " ".join(rest)
got = parse_hit(ctl.send(f"hit {x} {y}"))
if got is None:
record(lineno, line, "fail", "nothing clickable there")
elif got[0] != cls:
record(lineno, line, "fail", f"{got[0]} text={got[1]!r} box={got[2]}")
elif want_box and got[2] != want_box:
record(lineno, line, "fail", f"{cls} moved: box={got[2]}")
elif want_text and want_text not in got[1]:
record(lineno, line, "fail", f"{cls} text={got[1]!r}")
else:
record(lineno, line, "ok")
else:
sys.exit(f"FATAL: {script_path}:{lineno}: unknown command '{cmd}'")
crashed = ui_exited(console_path)
if crashed:
record(lineno, line, "fatal", crashed)
break
results.close()
print(f"== {counts['ok']} ok, {counts['fail']} failed, {counts['fatal']} fatal "
f"-> {os.path.join(outdir, 'results.jsonl')}", flush=True)
if counts["fail"] or counts["fatal"]:
sys.exit(1)
def main():
if len(sys.argv) < 3:
@@ -150,6 +324,8 @@ def main():
size = 720
if "--size" in sys.argv:
size = int(sys.argv[sys.argv.index("--size") + 1])
ctl_path = sys.argv[sys.argv.index("--ctl") + 1] if "--ctl" in sys.argv else None
console_path = sys.argv[sys.argv.index("--console") + 1] if "--console" in sys.argv else None
s = socket.socket(socket.AF_UNIX)
s.connect(path)
@@ -162,7 +338,7 @@ def main():
elif cmd == "tap":
do_tap(s, f, int(sys.argv[3]), int(sys.argv[4]))
elif cmd == "drive":
drive(s, f, sys.argv[3], sys.argv[4], size)
drive(s, f, sys.argv[3], sys.argv[4], size, ctl_path, console_path)
elif cmd == "quit":
s.sendall(b'{"execute":"quit"}\n')
+34 -5
View File
@@ -64,6 +64,7 @@ for _attempt in 1 2 3; do
PORT=$((21000 + RANDOM % 20000))
: > "$WORK/console.log"
bash "$QDIR/run.sh" --kernel "$ZIMAGE" --display headless --qmp "$WORK/qmp.sock" \
--ctl "$WORK/ctl.sock" \
--ssh-port "$PORT" --http-port $((PORT + 1)) --api-port $((PORT + 2)) \
> "$WORK/console.log" 2>&1 &
QEMU_PID=$!
@@ -118,15 +119,43 @@ done
exit 1
}
echo "== driving $SCRIPT"
python3 "$HERE/qmp.py" "$WORK/qmp.sock" drive "$SCRIPT" "$OUTDIR"
# The control bridge (init -> warden-ui's debug FIFO, see run.sh --ctl) comes
# up with the UI; a script's first `page`/`hit` must not race it. Bounded, and
# fail-closed: a scenario that asserts on UI state needs the channel, and a
# silently absent one would turn every assertion into an infrastructure error
# dressed as a test result.
echo "== waiting for the control bridge"
deadline=$((SECONDS + 60))
until grep -aq 'init: control bridge on' "$WORK/console.log"; do
[ $SECONDS -lt $deadline ] || {
echo "FATAL: the control bridge never announced itself (run.sh --ctl / init marker)" >&2
tail -25 "$WORK/console.log" >&2
exit 1
}
sleep 1
done
# The UI must still be alive: see the header. Checked AFTER the script so a
# crash caused by the interaction is caught, which is the usual case.
echo "== driving $SCRIPT"
# Every step lands in results.jsonl (ok / fail / fatal); an assertion mismatch
# is a `fail` and the run continues, so one run reports every broken
# expectation. The driver's exit status is the verdict; capture it rather than
# let `set -e` skip the backstop and the summary below.
drive_rc=0
python3 "$HERE/qmp.py" "$WORK/qmp.sock" drive "$SCRIPT" "$OUTDIR" \
--ctl "$WORK/ctl.sock" --console "$WORK/console.log" || drive_rc=$?
# The UI must still be alive: see the header. The driver checks this after
# every step and pins a crash to the step that caused it; this is the
# backstop for a death after the last step, or a driver that itself fell over.
if grep -aq 'warden-ui EXITED' "$WORK/console.log"; then
echo "FATAL: warden-ui DIED during the run:" >&2
grep -a -A22 'warden-ui EXITED' "$WORK/console.log" >&2
exit 1
fi
if [ "$drive_rc" -ne 0 ]; then
echo "UI-DRIVE-FAIL: see $OUTDIR/results.jsonl" >&2
grep -E '"status": "(fail|fatal)"' "$OUTDIR/results.jsonl" >&2 || true
exit 1
fi
echo "UI-DRIVE-PASS (screenshots in $OUTDIR)"
echo "UI-DRIVE-PASS (results in $OUTDIR/results.jsonl, screenshots in $OUTDIR)"