Repository navigation
PASS_WITH_CONDITIONS
Baseline. This re-reviews b4f415e6 against the PASS_WITH_CONDITIONS on 0eecd2c9 (this PR), on arch 109541a4. The checkout has no arch-map.md, so I started from the lines the baselines cite. No findings.json was staged. I checked that since.diff equals git diff 0eecd2c9 b4f415e6 and that pr.diff equals git diff 64d33586 b4f415e6. The compare is a merge of main (a926b2c8, empty --remerge-diff, so main's content only) and one PR commit, b4f415e6.
Governs: architecture.md line 495 (The box record and its states): stop order, --force kills at once, the stored ending, and the restart flip. Also line 497 (Retention and reaping) and the Box Record glossary entry at line 21.
Ruling. b4f415e6 makes a forced stop that arrives during a graceful stop end the grace at once. BoxHandle::stop_box sets the shared StopForce flag before the message queues, so the actor, busy in the graceful stop, still sees it. stop_box_processes checks the flag at its start and on every poll. stop_box_inner clears it after the signals on all three arms, before the record write. A forced message that queues behind carries force itself, so the clear loses nothing. This conforms to line 495 (--force kills at once). The commit also moves mark_name_stopped ahead of the record-write check in the StopBox arm. The processes are dead either way, so the name now fails closed (rule 7), and a retried stop marks it again. The new tests drive the real stop_box_processes and the real StopBox arm, so they test the cited rule (rule 3). BOX-013 and BOX-018 conform. BOX-154 keeps its baseline status: the flip exists and is tested, but it is not called at daemon start. It is filed as #2393, so C1 stays open at merge as that residual.
Departures. None. The why text of the earlier rulings on this PR is truncated in prior-rulings.json, so I could not read their condition lists word for word. I kept the BOX-154 deferral that the PR body and the code both name. I added no new condition on lines the compare does not touch.
Not checked. I did not build or run the tests, as the job requires. The PR body reports them green. I did not check whether STOP_GRACE (a 10 s constant in box_actor/host.rs) needs a configuration surface for line 495's "unless it is configured otherwise". The compare does not touch it, so the baseline status stands.
Conditions
- C1: In #2393 (T6), call
stop_orphaned_running_recordsfromBoxManager::initincrates/minimald/src/box_manager.rsonceBoxActor::runcan host a stopped box, and remove itsexpect(dead_code)attribute. Until then arunningrecord left by a dead daemon does not becomestoppedat restart. This can stay open at merge as the filed residual #2393.
Ruled at b4f415e against gominimal/arch@109541a48c31 by claude-opus-5-5; arch_rulings 556, job 19f223a9-3018-44d4-ba1c-47dd00348882.