d5e04be1

test: a refused birth paints [refused] and the wall stays up

a73x 2026-08-28 19:53

Commit message

test: a refused birth paints [refused] and the wall stays up

`.refused` lost its only end-to-end witness with the `mux TARGET#session`
spelling that used to drive it, and rested on a labelText unit test that
pins the string rather than a pump reaching it. A daemon holding
max_clients sessions refuses the picker's ninth attach: the bar wears the
state word, and the eight tiles already up outlive it.

test/e2e.sh

Old		New
@@ -177,8 +177,8 @@ done
177	# one of those and adds a convergence point would be pinning a fact every	177	# one of those and adds a convergence point would be pinning a fact every
178	# leg above already establishes.	178	# leg above already establishes.
179		179
180	[ "$OK_COUNT" = "88" ] \|\| {	180	[ "$OK_COUNT" = "89" ] \|\| {
181	echo "e2e FAIL: $OK_COUNT scenario checkpoints ran, the pin says 88 —"	181	echo "e2e FAIL: $OK_COUNT scenario checkpoints ran, the pin says 89 —"
182	echo " a scenario was added (update the pin) or silently lost"	182	echo " a scenario was added (update the pin) or silently lost"
183	exit 1	183	exit 1
184	}	184	}

test/e2e_13_birth.sh

@@ @@ -672,3 +672,94 @@ grep -q "lfd-two" "$OUT.lfdsess1" && { @@
+assert_stopped "$SOCK74" "$D74PID" "lowest-free-digit" "$OUT.lfdstop"
+D74PID=""
+ok "a new tile takes the lowest free digit, and the daemon takes the name back"
+# ---- a birth the DAEMON refuses narrates, and the wall stays up ---------
+#
+# `.refused` is a live tile state and the only one with no CLI spelling
+# left to reach it: `mux TARGET#session` is gone, so what used to drive
+# this leg went with it and the state word rested on a unit test of
+# `labelText` — which pins the STRING, not that a refused pump ever
+# reaches it.
+#
+# The refusal is the client TABLE, not the session table, and that is what
+# makes this cheap: every tile is its own attach and so its own client
+# slot, so a daemon holding eight sessions on a terminal with room for ten
+# stripes has eight tiles and no slot left. The picker's Enter dials a
+# ninth, the daemon refuses it before any replay frame, and the pump paints
+# `[refused]` on that tile's bar on its way out.
+#
+# Two claims, and the second is the newer one: the wall STAYS. A refused
+# picker birth used to take mux down with it (rc 1 on a terminal); a tile
+# that keeps the wall (`Tile.keeps_wall`) leaves the eight standing, which
+# is what the marker typed after the refusal proves — it reaches a session
+# because there is still a wall to type into.
+SOCK75="${TMPDIR:-/tmp}/muxd-e2e-reftile-$$.sock"
+defer_sock "$SOCK75"
+RTSTATE="${TMPDIR:-/tmp}/mux-e2e-reftile-state-$$"
+defer_rm "$RTSTATE"
+start_daemon "$SOCK75" "$OUT.rt.d" "refused-tile daemon never bound" --shell /bin/sh
+D75PID=$DPID
+# Seven more, so the daemon holds eight sessions with its own `0`: exactly
+# max_clients, so the wall fills the table and the ninth attach is the one
+# under test. The fills detach as they go and leave the slots free.
+fill_sessions "$SOCK75" "$OUT.rtfill" f 1 7
+wait_sessions "$SOCK75" 8 "refused-tile: the daemon should hold eight sessions"
+mkdir -p "$RTSTATE/mux"
+printf -- '--sock %s\n' "$SOCK75" > "$RTSTATE/mux/hosts"
+no_saved_tree "$RTSTATE"
+set +e
+# Tall and not wide, the hosts group's shape and for its reason: 80 columns
+# is under the aspect rule's `cols >= 2 * rows`, so the cut is stacked and
+# every bar gets the full width — beside, eight stripes are nine columns
+# each and `labelText` spends them all on the state word. Forty-four rows
+# over eight stripes leaves room for a ninth (three rows each, two of
+# content under a bar): the WALL is not what refuses here, and a terminal
+# that ran out of stripes first would say `[no room on the wall for another
+# tile]` and never dial at all.
+XDG_STATE_HOME="$RTSTATE" timeout 90 "$PTYCLIENT" --cols 80 --rows 44 \
+    --out "$OUT.rtcap" --err "$OUT.rtcap.err" -- "$MUX" > "$OUT.rtpc" 2>&1 <<'EOF'
+expect [up] 25000
+settle 2000 25000
+send \x1cs
+expect Enter/c new session 15000
+send \r
+expect [refused] 25000
+settle 1500 20000
+send printf 'rt-%s\n' alive\n
+expect rt-alive 20000
+settle 500 15000
+send \x1cd
+waitexit 15000
+EOF
+RC=$?
+set -e
+# The exit code IS the second claim: a wall that ended on the refusal would
+# be gone before the marker, and this is where that shows.
+[ "$RC" -eq 0 ] || {
+    echo "e2e FAIL: refused-tile: the wall exited $RC — a refused birth must"
+    echo "          leave the tiles that were already up standing:"
+    cat "$OUT.rtpc"; echo "--- stderr ---"; cat "$OUT.rtcap.err"; exit 1; }
+# The narration, off the STREAM: the refused tile vanishes on the pass
+# after it paints, so the final grid is not where the word lives.
+grep -qF -- "[refused]" "$OUT.rtcap" || {
+    echo "e2e FAIL: refused-tile: no tile ever wore the refused state word:"
+    grep -ao '\[[a-z ]*\]' "$OUT.rtcap" | sort -u | head; exit 1; }
+# ...and the refusal created nothing. A daemon that refused the attach but
+# kept the session would leave a session nobody can see — the worse of the
+# two failures, and invisible to the screen.
+wait_sessions "$SOCK75" 8 "refused-tile: a refused birth must leave no session behind"
+# The marker landed in a session on the wall, asked of the DAEMON: the
+# eight tiles outlived the ninth. Which one holds it is the focus's
+# business — the refusal puts the focus back on the tile it was born from
+# — so every session is asked and one hit is the claim.
+RTWHERE=""
+for _rt in 0 f1 f2 f3 f4 f5 f6 f7; do
+    if dump_session "$SOCK75" "$_rt" 2>/dev/null | grep -q "rt-alive"; then RTWHERE="$_rt"; fi
+done
+[ -n "$RTWHERE" ] || {
+    echo "e2e FAIL: refused-tile: the marker typed after the refusal reached no"
+    echo "          session — the wall did not survive its own refused birth:"
+    cat "$OUT.rtpc"; exit 1; }
+assert_stopped "$SOCK75" "$D75PID" "refused-tile" "$OUT.rtstop"
+D75PID=""
+ok "a birth the daemon refuses paints [refused] and leaves the wall standing"