[Birfy]

Written-by: Dev

Fixes #2634

A timed-out room turn of an external participant now ends its child process instead of leaving it running, so the next turn starts a fresh process and the late answer to the timed-out turn cannot settle it.

Owner: Dev
Reviewer: Reviewer

Changes

• core/src/host/room-participant-external.js — the timeout branch of runTurn() calls endChild(entry) before finish({ missed: true }), exactly as the abort branch already does. The timeout room report and the { missed: true } result are unchanged: a timeout is still silence, not a model failure.
• core/test/room-external.test.js — a child that holds the first turn's answer until a marker appears and handles turns serially. The second turn must get its own answer or { missed: true }, never the first turn's text, and must reach a different process.
• docs/subsystems/core-host.md — the host page says a timed-out participant is ended and the next turn starts a fresh process.
• changelog.d/2634.fixed.md.

Ran at this tip:

$ cd core && TZ=UTC node --import ./test/harness-resolve.mjs --test test/room-external.test.js
$ node scripts/ds-bot-patches.mjs check
$ node scripts/ds-bot-patches.mjs canonical
$ node core/scripts/build-client.mjs --check
$ npm --prefix core run lint

All green: 14/14 tests at this tip; with the timeout-path endChild(entry) reverted the new test fails with the second turn settled by the timed-out turn's answer to one. ds-bot-patches check and canonical are green, build-client --check is green, and lint reports 85 pre-existing warnings, 0 errors, with arch-docs: OK (6 changed file(s)).

Not run: the full npm run verify (another job was using this machine), the smoke test (no core/index.js or driver change) and the plugin suites (no plugin change).

Checkpoint

{
  "pr": 2635, "claim": "Dev", "issue": 2634,
  "summary": "A timed-out external room turn now ends its child process before the turn is finished, exactly as the abort path already did. Without that the next turn reused the still-running child and its late answer to the timed-out turn settled the new turn with the previous text. The timeout report and the { missed: true } result are unchanged, and the next turn starts a fresh process. A new test in core/test/room-external.test.js fails when the timeout-path endChild is reverted.",
  "files": ["core/src/host/room-participant-external.js", "core/test/room-external.test.js", "docs/subsystems/core-host.md", "changelog.d/2634.fixed.md", "patches/ds-bot/files/src/host/room-participant-external.js", "patches/ds-bot/files/test/room-external.test.js"],
  "checks": { "ran": "cd core && TZ=UTC node --import ./test/harness-resolve.mjs --test test/room-external.test.js; node scripts/ds-bot-patches.mjs check; node scripts/ds-bot-patches.mjs canonical; node core/scripts/build-client.mjs --check; npm --prefix core run lint", "result": "green", "at": "09786e6b9e5b7426cc9d77b6f198861cb5dc415e" },
  "next": "none — ready for review"
}

[Birfy]

Verdict: PASS at 09786e6
Written-by: Reviewer
Attestation: 09786e6 role=Reviewer

Ran at this tip:

$ cd /data/dsh/bot/workspace/r2635/tip/core && DSH_HARNESS=/data/dsh/bot/workspace/d2607/repo/core/dev/harness TZ=UTC node --import ./test/harness-resolve.mjs --test test/room-external.test.js
# tests 14  # pass 14  # fail 0

$ cd /data/dsh/bot/workspace/r2635/red/core && DSH_HARNESS=/data/dsh/bot/workspace/d2607/repo/core/dev/harness TZ=UTC node --import ./test/harness-resolve.mjs --test test/room-external.test.js
# tests 14  # pass 13  # fail 1
# not ok 9 - a timed-out turn never settles the next one with its late answer (#2634)
#   error: "the next turn was settled by the timed-out turn's late answer"
#   expected: 'answer to one'  actual: 'answer to one'  operator: 'notDeepStrictEqual'

$ cd /data/dsh/bot/workspace/r2635/tip && node scripts/ds-bot-patches.mjs check
# ok 10 patches apply cleanly in order
# ok core/ equals ds-bot/ + patches/ds-bot/

$ cmp core/src/host/room-participant-external.js patches/ds-bot/files/src/host/room-participant-external.js && cmp core/test/room-external.test.js patches/ds-bot/files/test/room-external.test.js
# both byte-identical (md5 ca61861… / 7aaca0e…)

Not run: full npm run verify, the smoke test, the plugin suites (out of scope for this line).

Red first: the tip's core/src/host/room-participant-external.js with only the timeout branch's endChild(entry) line deleted fails exactly as claimed (second turn settled by answer to one), and the restored tip is 14/14 green on the same machine.

Fix location and side effects: endChild(entry) sits before entry.state = 'timeout' / finish({ missed: true }), same as the abort branch. endChild sets entry.stopping = true and state = 'stopped', and reportExit returns early on either (core/src/host/room-participant-external.js:167), so no extra crashed/exited room line follows the single timeout report; { missed: true } is unchanged.

Fresh process is asserted, not just text: the child records start <pid> and turn <pid> <text>, and the test asserts starts()[1] !== starts()[0] plus answer !== 'answer to one'.

Scope: 6 files, exactly the change set (core/, its two patches/ds-bot/files/ mirrors, the owned docs/subsystems/core-host.md, and the changelog fragment); base 2e93b8c09174ddb6e843593c8e7b203348a7cb41. Docs: core/src/host/ is an owned path of core-host.md and the updated sentence matches the implementation (timed-out turn ends the child; next turn starts fresh). Checkpoint block: 1266 chars, files all in the diff, checks.at is this tip.

CI at the tip: tests-ok, merge-gate, Node 24, plugins, package green; enqueue-guard red only for "no Verifier report names the tip 09786e6" (this report is that one).

[Birfy]

Written-by: Enqueue

Merged through the merge queue: squash commit 6a867a9d02accde5811e0fcec02d0152f2485360 at 2026-11-07T19:42:37Z.

main head is that commit, and Test on the push to main is success (run 42407530148). Queue entry at 09786e6b9e5b7426cc9d77b6f198861cb5dc415e: merge-gate, enqueue-guard and Test on merge_group all success; every check at the tip terminal, no failure.