a73x

f1d908b5

test: the hidden-tile leg pins silence, not a repaint schedule

a73x   2026-08-20 10:58

Commit message
test: the hidden-tile leg pins silence, not a repaint schedule

`bhidden-pin` was asserted to appear exactly once in the capture, and that
count failed about three runs in five with "2 time(s)". Both paints were
legitimate: after the unzoom a demoted tile has two paths to the stripe — a
frame arriving, and `repaint_gen` — so the second occurrence is a schedule
this leg has no business pinning.

The invariant the scenario's own comment states is narrower and is the one
worth a gate: a tile the zoom is not on is SILENT. That is a statement about
the window before the unzoom, so the capture is split there. The zoomed
session prints `zhzoom-ok` immediately before the `\x1cw` chord, which makes
it the split point; zero occurrences before it, at least one after.

The split is a byte offset, not `sed '1,/zhzoom-ok/p'`. The line-addressed
version was written first and passed nothing: the capture of a wall is CUP
sequences with no newlines in them at all — measured, 0 newlines in 3529
bytes — so a line address hands back the whole file and the "before" half
silently becomes the entire run. The same absent newlines are already why
the count is `grep -o` rather than `grep -c`, two lines above.

The "at least one after" half is not decoration: without it a hidden tile
that never painted would satisfy the silence.

make e2e rc=0, pin unmoved at 55 scenarios / 35 convergence points.
test/e2e.sh
Old New
@@ -6104,14 +6104,21 @@ ok "an application in a zoomed tile gets the wheel, and the tile does not"
6104 # shell — typed at, through the zoom — that creates the flag. So the 6104 # shell — typed at, through the zoom — that creates the flag. So the
6105 # hidden tile's output happens strictly INSIDE the zoom, with no sleep 6105 # hidden tile's output happens strictly INSIDE the zoom, with no sleep
6106 # to race and no ordering to hope for. 6106 # to race and no ordering to hope for.
6107 # * `bhidden-pin` must therefore appear exactly ONCE in this terminal's 6107 # * the capture is then SPLIT at `zhzoom-ok`, which the zoomed session
6108 # whole life: the stripe repaint after the unzoom, from the replica that 6108 # prints immediately before the unzoom chord. Before that split
6109 # was hot the entire time. A second occurrence is the hidden tile having 6109 # `bhidden-pin` must appear ZERO times — that is "silent while hidden",
6110 # painted over the session the user was typing at. 6110 # stated exactly. After it, at least once: the stripe repaint from the
6111 # 6111 # replica that was hot the entire time, which is also what proves the
6112 # The zoomed session gets its own marker in between, so a leg that lost the 6112 # hidden session really did have something to say.
6113 # zoom (and would then pass the count for the wrong reason) fails loudly on 6113 #
6114 # the expect instead. 6114 # What is deliberately NOT pinned is the count after the unzoom. A demoted
6115 # tile has two legal paint paths — a frame arriving, and `repaint_gen` — so a
6116 # second stripe paint is a repaint schedule rather than a broken invariant,
6117 # and pinning it to one failed this leg about three runs in five.
6118 #
6119 # The zoomed session's own marker is therefore load-bearing twice: it is the
6120 # split point, and a leg that lost the zoom fails loudly on its expect
6121 # instead of passing the count for the wrong reason.
6115 cat > "$ZHASH" <<'EOF' 6122 cat > "$ZHASH" <<'EOF'
6116 #!/bin/sh 6123 #!/bin/sh
6117 printf 'zha-%s\n' pin 6124 printf 'zha-%s\n' pin
@@ -6177,12 +6184,28 @@ set -e
6177 cat "$OUT.zhpc"; exit 1; } 6184 cat "$OUT.zhpc"; exit 1; }
6178 # The money assertion. `grep -o`, not `grep -c`: a stripe paint is escape 6185 # The money assertion. `grep -o`, not `grep -c`: a stripe paint is escape
6179 # sequences with no newlines in them, so a line count reads 1 either way. 6186 # sequences with no newlines in them, so a line count reads 1 either way.
6180 _zh_paints=$(grep -oa "bhidden-pin" "$OUT.zhcap" | wc -l) 6187 #
6181 [ "$_zh_paints" -eq 1 ] || { 6188 # Which is also why the split is a BYTE offset and not `sed '1,/marker/p'`:
6189 # the same absent newlines make this whole capture one enormous line, so a
6190 # line-addressed split hands back the entire file and the "before" half
6191 # quietly becomes the whole run.
6192 _zh_off=$(grep -oab "zhzoom-ok" "$OUT.zhcap" | head -1 | cut -d: -f1)
6193 [ -n "$_zh_off" ] || {
6194 echo "e2e FAIL: hidden tile: the zoomed session's marker is not in the capture,"
6195 echo " so there is no before/after to split on:"
6196 cat "$OUT.zhpc"; exit 1; }
6197 _zh_pre=$(head -c "$_zh_off" "$OUT.zhcap" | grep -oa "bhidden-pin" | wc -l)
6198 _zh_post=$(tail -c "+$((_zh_off + 1))" "$OUT.zhcap" | grep -oa "bhidden-pin" | wc -l)
6199 [ "$_zh_pre" -eq 0 ] || {
6182 echo "e2e FAIL: hidden tile: the hidden session's marker reached this terminal" 6200 echo "e2e FAIL: hidden tile: the hidden session's marker reached this terminal"
6183 echo " $_zh_paints time(s); exactly one — the stripe repaint after the" 6201 echo " $_zh_pre time(s) WHILE THE ZOOM WAS UP; a tile the zoom is not"
6184 echo " unzoom — is the whole budget, and the rest is a tile painting" 6202 echo " on is silent, and that is a tile painting over the session the"
6185 echo " over the session the user was typing at:" 6203 echo " user was typing at:"
6204 cat "$OUT.zhpc"; exit 1; }
6205 [ "$_zh_post" -ge 1 ] || {
6206 echo "e2e FAIL: hidden tile: the marker never reached this terminal after the"
6207 echo " unzoom, so the stripe repaint from the hot replica never"
6208 echo " happened and the silence above proves nothing:"
6186 cat "$OUT.zhpc"; exit 1; } 6209 cat "$OUT.zhpc"; exit 1; }
6187 assert_stopped "$SOCK43" "$D40PID" "hidden tile A" "$OUT.zhstopa" 6210 assert_stopped "$SOCK43" "$D40PID" "hidden tile A" "$OUT.zhstopa"
6188 D40PID="" 6211 D40PID=""