dangernoodle-io / breadboard / 29218561166

95%

push

github

feat: add condvar wait/signal/broadcast to bb_lock (#840)

- bb_lock_cond_t composes with the caller's existing bb_lock_t: wait()
  atomically releases the lock and blocks (timeout_ms, never raw ticks;
  BB_LOCK_COND_WAIT_FOREVER sentinel), re-acquiring it before returning;
  signal() wakes one waiter, broadcast() wakes ALL — mandatory for any
  multi-peeker consumer (the prerequisite for B1-821's bqueue mailbox).
- host backend: real pthread_cond_t paired with bb_lock_t's own
  pthread_mutex_t (via new platform/host/bb_core/bb_lock_impl.h, factored
  out of bb_lock.c for reuse). CLOCK_MONOTONIC via
  pthread_condattr_setclock() on Linux/glibc/musl; macOS lacks that call,
  so there this uses the Darwin-only pthread_cond_timedwait_relative_np()
  (a relative timeout, immune to wall-clock steps by construction) instead.
- ESP-IDF backend: bb_lock_t wraps a raw FreeRTOS mutex semaphore there
  (not a pthread_mutex_t, for priority-inheritance reasons predating this
  change), so ESP-IDF's own pthread_cond_t cannot pair with it. Instead this
  is a FreeRTOS-native waiter-list implementation that deliberately mirrors
  the algorithm ESP-IDF's own components/pthread/pthread_cond_var.c uses
  (a proven design, not a novel one) — zero heap in the wait() hot path.
- caller-owned storage, no heap, no stats/instrumentation (deliberately out
  of scope, tracked under B1-524).
- host tests cover real block/wake, broadcast-wakes-all (the primitive's
  reason for existing), signal-wakes-exactly-one, timeout expiry, genuine
  lock release/re-acquire, and NULL-arg safety — 100% line+branch on the
  new host backend file.

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>

3987 of 3987 branches covered (100.0%)

Branch coverage included in aggregate %.

84 of 84 new or added lines in 3 files covered. (100.0%)

6456 of 6456 relevant lines covered (100.0%)

8068.02 hits per line