lane-handoff reads the lane lifecycle whole and parses it afterwards, so an awk that exits early can no longer turn lanes-edit.sh's SIGPIPE into a lifecycle that could not be read (fixes #280)

lane-handoff

@@ -1486,24 +1486,36 @@ lc_on=""; lc_gen=""; lc_op=""; lc_state=""; lc_inventory_missed=""
 
 lifecycle_begin() {
 	lc_on=""; lc_gen=""; lc_op=""; lc_state=""
-	lcb_now=""; lcb_rc=0
-	lcb_now="$(LANES_NO_FETCH=1 "$LANES_EDIT" lane-state "$lane" 2>/dev/null \
-		| awk -F"\t" '$1 == "state" { print $2; exit }')" || lcb_rc=$?
+	lcb_now=""; lcb_raw=""; lcb_rc=0
+	# READ WHOLE, THEN PARSE (opensoft/openRepoTools#280). This read was
+	# `lane-state | awk '… { print $2; exit }'` under `pipefail`, and that `awk`
+	# exits at the line it wants: a `lane-state` that wrote once more after it
+	# — its builtin `printf` hands a multi-line snapshot to the pipe in pieces —
+	# died of SIGPIPE, the substitution's status was 141, and the `case` below
+	# took that for a read that FAILED: "could not be read", no fence, no
+	# inventory and no RUNNING -> SWAPPING -> SWAPPED, on timing alone (CI run
+	# 39322387035). The reader had stopped listening; the read had not failed.
+	# So the status below is `lane-state`'s own, from a command nothing can end
+	# early, and the state is parsed out of what it printed afterwards.
+	lcb_raw="$(LANES_NO_FETCH=1 "$LANES_EDIT" lane-state "$lane" 2>/dev/null)" || lcb_rc=$?
 	case "$lcb_rc" in
 	0 | 8) : ;;
 	*)
 		note "the lane lifecycle could not be read (\`$LANES_EDIT lane-state $lane\` exited $lcb_rc), so this handoff records no RUNNING -> SWAPPING -> SWAPPED transition. The swap itself goes on (\`R-A11-11\`) and the PAUSED record below is unaffected."
 		return 0 ;;
 	esac
+	lcb_now="$(printf '%s\n' "$lcb_raw" | awk -F"\t" '$1 == "state" { print $2; exit }')"
 	[ -n "$lcb_now" ] || lcb_now=none
 	# AN INTERRUPTED SWAP IS TAKEN OVER, NOT COMPETED WITH. A lane still
 	# reading `SWAPPING` when a new handoff starts is one whose previous
 	# operation never finished; this one supersedes it with a NEW generation,
 	# which is precisely what refuses the old finalizer if it ever wakes up.
 	# The old operation id is named rather than lost.
 	if [ "$lcb_now" = SWAPPING ]; then
-		lcb_old="$(LANES_NO_FETCH=1 "$LANES_EDIT" lane-state "$lane" 2>/dev/null \
-			| awk -F"\t" '$1 == "operation" { print $2; exit }')"
+		# Read whole, then parsed, for the reason given above (#280); its status
+		# was never read, and the operation is only named in the note below.
+		lcb_old_raw="$(LANES_NO_FETCH=1 "$LANES_EDIT" lane-state "$lane" 2>/dev/null)"
+		lcb_old="$(printf '%s\n' "$lcb_old_raw" | awk -F"\t" '$1 == "operation" { print $2; exit }')"
 		note "lane $lane is still recorded SWAPPING from operation ${lcb_old:-unknown}, which never finished: this handoff SUPERSEDES it with a new generation, and that is what refuses the old finalizer if it ever returns. Nothing of that operation is undone here."
 	fi
 	lcb_out=""; lcb_src=0

tests/test_triad_lane_tooling_handoff.py

@@ -96,9 +96,9 @@
 
 
 def handoff(b: Bench, lane: str, cwd: Path, why: str = "t018 handoff",
-            *flags: str) -> subprocess.CompletedProcess:
+            *flags: str, env: dict | None = None) -> subprocess.CompletedProcess:
     return b.tool("lane-handoff", "--lane", lane, "--transcript", "none", "--agent", "claude",
-                  *flags, why, cwd=cwd)
+                  *flags, why, cwd=cwd, env=env)
 
 
 def writers_file(b: Bench, lane: str, rows: dict) -> Path:
@@ -789,6 +789,138 @@
     assert inventory(b, "triad-1") == want
 
 
+# ===================================================================== #280
+
+#: A `lanes-edit.sh` that is the bench's own for every verb but `lane-state`, and
+#: answers that one in PIECES - the `state` line, 1.5 MiB of padding, then the
+#: rest of the snapshot - so that its reader has gone before it has finished,
+#: WHATEVER THE TIMING AND WHATEVER THE AWK. A pipe holds 64 KiB, so a producer
+#: with 1.5 MiB to write is still writing when a reader that stops at the first
+#: line closes; its next write gets SIGPIPE and the wrapper dies of it, as the
+#: real one can when a reader is quick enough to leave between its writes (its
+#: builtin `printf` writes a multi-line snapshot a line at a time). A pause
+#: instead of the padding was a race of its own: an awk that stalled for longer
+#: than the pause read it all. mawk reads ahead until its buffer is full - the
+#: review of fc061fe measured the end of input at 255 KiB or less and an early
+#: stop at 320 KiB or more - so the padding stays far above that. Its key, `pad`,
+#: is no field `lane-state` prints, so nothing in `lane-handoff` takes it for
+#: one. `real` is the bench's copy, `calls` the file each `lane-state` call is
+#: logged to, `pad` one padding line (a real TAB) and `pad_lines` how many.
+SLOW_LANE_STATE = """#!/bin/sh
+real='{real}'
+if [ "$1" = lane-state ]; then
+    printf '%s\\n' "$*" >> '{calls}'
+    out="$("$real" "$@")"
+    rc=$?
+    if [ "$rc" != 0 ]; then
+        [ -z "$out" ] || printf '%s\\n' "$out"
+        exit "$rc"
+    fi
+    first="$(printf '%s\\n' "$out" | sed -n 1p)"
+    rest="$(printf '%s\\n' "$out" | sed 1d)"
+    printf '%s\\n' "$first"
+    yes '{pad}' | head -n {pad_lines}
+    printf '%s\\n' "$rest"
+    exit 0
+fi
+exec "$real" "$@"
+"""
+#: One line of padding, and how many: 32768 lines of 48 bytes are 1.5 MiB.
+PAD_LINE = "pad\tno lifecycle field is spelled like this one"
+PAD_LINES = 32768
+
+#: `lane-handoff`'s lifecycle read as it was written before #280: the producer
+#: piped into an `awk` that exits at its first match, under `pipefail`.
+OLD_LIFECYCLE_READ = r'''set -uo pipefail
+rc=0
+now="$(LANES_NO_FETCH=1 "$1" lane-state "$2" 2>/dev/null | awk -F"\t" '$1 == "state" { print $2; exit }')" || rc=$?
+printf '%s %s\n' "$now" "$rc"
+'''
+
+
+def slow_lane_state(b: Bench) -> tuple:
+    """THE PRODUCER THAT OUTLIVES ITS READER: the wrapper (a `$LANES_EDIT`), and
+    the file it logs each `lane-state` call to."""
+    wrapper = b.root / "slow-lane-state" / "lanes-edit.sh"
+    wrapper.parent.mkdir()
+    calls = wrapper.parent / "calls"
+    wrapper.write_text(SLOW_LANE_STATE.format(real=b.bin / "lanes-edit.sh", calls=calls,
+                                              pad=PAD_LINE, pad_lines=PAD_LINES))
+    wrapper.chmod(0o755)
+    return wrapper, calls
+
+
+def lane_state(b: Bench, lane: str) -> dict:
+    proc = b.tool("lanes-edit.sh", "lane-state", lane)
+    assert proc.returncode == 0, proc.stderr
+    return dict(ln.split("\t", 1) for ln in proc.stdout.splitlines() if "\t" in ln)
+
+
+@NEEDS_UPSTREAM
+def test_280_a_lifecycle_read_whose_producer_writes_after_the_filter_exits_is_still_a_read(triad):
+    """#280: `lane-handoff` read the lane's lifecycle as `lane-state | awk '$1 ==
+    "state" { print $2; exit }'` under `set -o pipefail`, and that `awk` exits at
+    the first line. A producer that writes once more after that dies of SIGPIPE -
+    the real one's builtin `printf` hands a multi-line snapshot to the pipe in
+    pieces, and a reader quick enough sees the first before the last is written -
+    so the substitution's status was 141, the handoff said "the lane lifecycle
+    could not be read" and recorded no RUNNING -> SWAPPING -> SWAPPED transition
+    (CI run 39322387035, which passed one prefix down and in every run before
+    it). The race is held here EVERY time, by a `lane-state` that prints its
+    `state` line, 1.5 MiB of padding and then the rest: a pipe holds 64 KiB, so
+    that producer is still writing when a reader that stops at the first line
+    has closed, whatever the timing and whatever the awk. THE PREMISE FIRST: the
+    old shape of the read, run on that producer, gives `RUNNING 141`. Then the
+    handoff reads the snapshot whole and takes the status from `lane-state`
+    alone, so it records the transition as it does when the producer is quick."""
+    b, t = triad
+    a = t["assembly"]
+    wrapper, calls = slow_lane_state(b)
+    old = b.run("bash", "-c", OLD_LIFECYCLE_READ, "_", wrapper, "triad-1")
+    assert (old.returncode, old.stdout) == (0, "RUNNING 141\n"), (old.stdout, old.stderr)
+    calls.write_text("")        # the premise logged a call of its own
+
+    proc = handoff(b, "triad-1", a, "t280 handoff", env={"LANES_EDIT": str(wrapper)})
+    assert proc.returncode == 0, proc.stdout + proc.stderr
+    assert "lane-state triad-1" in calls.read_text().splitlines(), \
+        "the handoff never read the lifecycle through the wrapper"
+    assert "the lane lifecycle could not be read" not in proc.stderr, proc.stderr
+    assert "exited 141" not in proc.stderr, proc.stderr
+    assert "lifecycle: RUNNING -> SWAPPING (generation " in proc.stderr, proc.stderr
+    assert "lifecycle: SWAPPING -> SWAPPED (generation " in proc.stderr, proc.stderr
+    assert "lifecycle: SWAPPED" in proc.stdout, proc.stdout
+    assert lane_state(b, "triad-1")["state"] == "SWAPPED"
+
+
+@NEEDS_UPSTREAM
+def test_280_a_swapping_lane_read_the_same_way_still_names_the_operation_it_supersedes(triad):
+    """#280, a lane left SWAPPING: the FIRST read - the state - goes through the
+    same producer and must still see SWAPPING, so the handoff takes the lane
+    over, and its note names the operation that never finished. THAT IS WHAT
+    THIS PINS: the note's content, and the first read on a SWAPPING lane. It
+    cannot pin the SECOND read's own fix, the operation read once more for that
+    note: its status was never read, so reverting that site alone still passes."""
+    b, t = triad
+    a = t["assembly"]
+    wrapper, calls = slow_lane_state(b)
+    took = b.tool("lanes-edit.sh", "set-lane-state", "triad-1", "SWAPPING", "--expect", "RUNNING",
+                  "--operation", "op-stuck-280")
+    assert took.returncode == 0, took.stderr
+    assert lane_state(b, "triad-1")["operation"] == "op-stuck-280"
+
+    proc = handoff(b, "triad-1", a, "t280 handoff", env={"LANES_EDIT": str(wrapper)})
+    assert proc.returncode == 0, proc.stdout + proc.stderr
+    assert "lane-state triad-1" in calls.read_text().splitlines(), \
+        "the handoff never read the lifecycle through the wrapper"
+    assert "the lane lifecycle could not be read" not in proc.stderr, proc.stderr
+    assert ("lane triad-1 is still recorded SWAPPING from operation op-stuck-280, which never "
+            "finished") in proc.stderr, proc.stderr
+    assert "lifecycle: SWAPPING -> SWAPPING (generation " in proc.stderr, proc.stderr
+    assert "lifecycle: SWAPPING -> SWAPPED (generation " in proc.stderr, proc.stderr
+    after = lane_state(b, "triad-1")
+    assert after["state"] == "SWAPPED" and after["operation"] != "op-stuck-280", after
+
+
 # ===================================================== the single repository
 
 def single_repository_decoys(bench: Bench) -> list: