dangernoodle-io / breadboard / 27921360745

100%

push

github

fix: bb_pub status-read locking and buffer_replay non-reentrancy (B1-292 follow-up) (#491)

Four correctness fixes following the B1-292 bounded-wait restructure that
released s_tick_lock before the sink fan-out:

- bb_pub_get_status(): acquire s_tick_lock around reads of s_last_publish_ok,
  s_last_publish_ms, and s_published_ever to prevent torn reads vs the Phase 3
  writer; s_source_count and s_sink_count stay unlocked (registration-time only).
- bb_pub_tick_once(): snapshot interval_ms under s_tick_lock during Phase 1
  as snap_interval_ms; Phase 2 uses the snapshot instead of re-reading
  s_interval_ms without any lock, eliminating the race with
  bb_pub_set_interval_ms() from an HTTP handler.
- buffer_capture / buffer_replay comments: replace stale "called under
  s_tick_lock" with the real invariant — single-worker guarantee; on-failure
  path runs with no lock held.
- buffer_replay: drop static from s_replay_buf, making it a stack local
  (~450 bytes, well within the >= 4096-byte worker stack). The static was safe
  only under the single-worker guarantee, but stack allocation is clearer and
  avoids BSS waste now that the call site is outside any lock.

Pause/condvar wait logic is unchanged.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>

1718 of 1718 branches covered (100.0%)

Branch coverage included in aggregate %.

2845 of 2845 relevant lines covered (100.0%)

1227.49 hits per line