e09da54f

fix: review conditions — soak refuses a dirty baseline; the scroll bug's true severity

a73x 2026-08-09 21:47

Commit message

fix: review conditions — soak refuses a dirty baseline; the scroll bug's true severity

The reviewer ran five sampled mutations and all held; two conditions
folded here. soak.sh charged pre-existing /tmp residue to run 1 of every
soak (green suite, FAIL verdict — the evidence a failing e2e deliberately
leaves is exactly what poisons the next soak), so it now refuses to start
over residue rather than baseline-subtracting it. And the scroll-
suppression bug was overstated everywhere it was described: Shift+PageDown's
scroll_pages==0 arm clears the mode unconditionally, so the state is
recoverable by a keystroke no user would guess, not terminal — comment and
both docs now say so. Also: assert_converged's styled-only failure printed
an empty diff and no explanation (the [ -s ] case is the styled case);
and a drifting line-number cite became a test-name cite.

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

docs/decisions.md

Old		New
@@ -1620,8 +1620,11 @@ mode — it drops pending and retired predictions and nothing else. A
1620	reconnect taken while scrolled therefore leaves the client painting live	1620	reconnect taken while scrolled therefore leaves the client painting live
1621	rows with the overlay still suppressing, and the "any other key leaves	1621	rows with the overlay still suppressing, and the "any other key leaves
1622	scroll mode" branch can never rescue it because that branch is guarded by	1622	scroll mode" branch can never rescue it because that branch is guarded by
1623	`scroll_pages > 0`, which the reconnect has just made false. Local echo	1623	`scroll_pages > 0`, which the reconnect has just made false. Shift+PageDown
1624	prediction is off for the rest of the session, silently. Fixed	1624	does still clear it — its `scroll_pages == 0` arm calls
		1625	`setScrollMode(false)` unconditionally — so the state is recoverable, by a
		1626	keystroke no user has any reason to try. Local echo prediction is silently
		1627	off until they do. Fixed
1625	client-side with one call next to the existing `scroll_pages = 0` —	1628	client-side with one call next to the existing `scroll_pages = 0` —
1626	deliberately not by making `flush()` clear the mode, because every	1629	deliberately not by making `flush()` clear the mode, because every
1627	snapshot flushes and a snapshot is not a reason to leave history; the	1630	snapshot flushes and a snapshot is not a reason to leave history; the
@@ -1671,8 +1674,9 @@ forward for another milestone.
1671	pen it could protect against belongs to the HOST terminal — which no	1674	pen it could protect against belongs to the HOST terminal — which no
1672	rendered-grid comparison can ever hold, since the render helper replays	1675	rendered-grid comparison can ever hold, since the render helper replays
1673	into a fresh engine at the default pen by construction. The unit test	1676	into a fresh engine at the default pen by construction. The unit test
1674	at `src/client.zig:1837` is the real guard and fails under the mutation	1677	`paintDeltaClipped skips rows beyond the tty and clamps the cursor`
1675	today; the bank names the dependency guarantee it rests on.	1678	(client.zig) is the real guard and fails under the mutation today; the
		1679	bank names the dependency guarantee it rests on.
1676	- Catch parity is not diagnosis parity (rows 11 and 13). Both suites	1680	- Catch parity is not diagnosis parity (rows 11 and 13). Both suites
1677	catch both mutations, so the disposition column reads "both caught" and	1681	catch both mutations, so the disposition column reads "both caught" and
1678	says nothing interesting. What the columns hide: the old suite fails	1682	says nothing interesting. What the columns hide: the old suite fails

docs/roadmap.md

Old		New
@@ -65,8 +65,8 @@ markers; new suite 13 of 18; the six-row delta is all paint, overlay
65	and clip — the byte-blind class, a wrong picture drawn from right bytes	65	and clip — the byte-blind class, a wrong picture drawn from right bytes
66	(decisions.md, M11, which carries the table in full). Zero rows ended	66	(decisions.md, M11, which carries the table in full). Zero rows ended
67	undecided. En route the campaign found a real bug in unmutated code — a	67	undecided. En route the campaign found a real bug in unmutated code — a
68	reconnect taken while scrolled left prediction silently off for the rest	68	reconnect taken while scrolled left prediction silently off until the
69	of the session — and a computed timing knife-edge in the burst scenario	69	user happened to press Shift+PageDown — and a computed timing knife-edge in the burst scenario
70	that had been sitting on the expiry bound exactly, whose formula also	70	that had been sitting on the expiry bound exactly, whose formula also
71	explains the M9-era `DELAY_MS=400` collapse retroactively. The rule	71	explains the M9-era `DELAY_MS=400` collapse retroactively. The rule
72	extracted: **a timing margin is a number you compute, not a sentence you	72	extracted: **a timing margin is a number you compute, not a sentence you

src/client.zig

Old		New
@@ -610,10 +610,14 @@ fn session(
610	// The overlay has to be told, and only the client can tell it:	610	// The overlay has to be told, and only the client can tell it:
611	// `flush()` drops predictions but deliberately leaves the mode	611	// `flush()` drops predictions but deliberately leaves the mode
612	// bit alone, so a reconnect taken while scrolled would leave the	612	// bit alone, so a reconnect taken while scrolled would leave the
613	// overlay suppressing forever. The exit-scroll branch below	613	// overlay suppressing with no page to suppress for. The
614	// cannot rescue it either — that one is guarded by	614	// "any other key" exit at the bottom of the loop cannot rescue
615	// `scroll_pages > 0`, which the line above has just made false.	615	// it — that branch is guarded by `scroll_pages > 0`, which the
616	// Prediction would be silently off for the rest of the session.	616	// line above has just made false. Shift+PageDown still can
		617	// (its `scroll_pages == 0` arm clears the mode unconditionally),
		618	// so this is recoverable rather than terminal — but only by a
		619	// keystroke the user has no reason to guess, so prediction is
		620	// silently off until they do.
617	overlay.setScrollMode(false);	621	overlay.setScrollMode(false);
618	if (!reconnect(	622	if (!reconnect(
619	alloc,	623	alloc,

test/e2e.sh

Old		New
@@ -167,8 +167,16 @@ assert_converged() {
167	CONV_COUNT=$((CONV_COUNT + 1))	167	CONV_COUNT=$((CONV_COUNT + 1))
168	converged_quiet "$1" "$2" \|\| {	168	converged_quiet "$1" "$2" \|\| {
169	echo "e2e FAIL: $3: client render diverges from daemon grid (-daemon +client):"	169	echo "e2e FAIL: $3: client render diverges from daemon grid (-daemon +client):"
170	head -40 "$1.diff" 2>/dev/null \|\| \	170	# An empty diff file is the styled-only case, not a passing one —
171	echo "(no plain diff: divergence is styled-only, or the helper failed)"	171	# the glyphs agree and the pens do not, which is precisely what the
		172	# byte-exact leg exists to catch, so say so instead of printing
		173	# forty lines of nothing.
		174	if [ -s "$1.diff" ]; then
		175	head -40 "$1.diff"
		176	else
		177	echo "(no plain diff: the grids agree on glyphs and differ on STYLE —"
		178	echo " compare $1.dvt against $1.rvt)"
		179	fi
172	echo "grids left in $1.render / $1.dump / $1.rvt / $1.dvt"	180	echo "grids left in $1.render / $1.dump / $1.rvt / $1.dvt"
173	exit 1	181	exit 1
174	}	182	}

test/soak.sh

Old		New
@@ -14,6 +14,19 @@ FAILDIR="$TMP/mux-soak-$$-failures"
14	SUMMARY="$TMP/mux-soak-$$.summary"	14	SUMMARY="$TMP/mux-soak-$$.summary"
15	LOG="$TMP/mux-soak-$$.log"	15	LOG="$TMP/mux-soak-$$.log"
16	: > "$SUMMARY"	16	: > "$SUMMARY"
		17	# Residue predating the soak is not this soak's leak. Refuse to start rather
		18	# than baseline-subtract: the files are probably evidence a previous failing
		19	# e2e deliberately left behind, and the per-run hygiene check below cannot
		20	# tell them apart from a leak — forcing a look is better than silently
		21	# discarding them.
		22	BASE_STRAYS=$(find "$TMP" -maxdepth 1 $ -name 'muxd-e2e-' -o -name 'mux-e2e-' $ 2>/dev/null \| wc -l)
		23	[ "$BASE_STRAYS" -eq 0 ] \|\| {
		24	echo "soak FAIL: $BASE_STRAYS mux e2e temp files already in $TMP before run 1."
		25	echo " They are probably evidence from an earlier failing e2e."
		26	echo " Inspect them, then clear them — the per-run hygiene check"
		27	echo " cannot tell them apart from a leak."
		28	exit 1
		29	}
17	FAILED=0	30	FAILED=0
18	i=1	31	i=1
19	while [ "$i" -le "$N" ]; do	32	while [ "$i" -le "$N" ]; do