dangernoodle-io / breadboard / 31109100871

97%

push

github

feat: wire a real second caller to bb_data_http's shared EVENT ring (B1-1451) (#1240)

- bb_data_http_notify_push() (ESP-IDF, new): push_pump() + an immediate
  task-notification wake of the broadcaster, so a producer's EVENT push
  no longer waits for the next scheduled sweep to be drained.
- broadcaster_task() now waits on a bounded task notification
  (ulTaskNotifyTake, timeout = BB_DATA_HTTP_SWEEP_INTERVAL_MS) instead of
  an unconditional delay, preserving the periodic sweep for every
  attached EVENT key regardless of notification traffic.
- examples/smoke: CONFIG_BB_SMOKE_DATA_NOTIFY-gated proof harness
  (smoke_data_notify.c) with three POST routes (notify-push, tick-push,
  notify-stress) for off-board latency and two-task ring-contention
  verification against a new EVENT-kind smoke_notify key.
- the esp32-data-notify env is wired into the Makefile smoke: aggregate
  and build.yml's smoke matrix, so CI compiles the harness rather than
  leaving it runnable only by hand.

Hardware validation (ESP32, app-partition flash, hash-verified):
- boot clean, zero error lines; bb_ws registered /ws/events and /ws. The
  prior image failed all three WS route registrations with 262
  (ESP_ERR_NOT_SUPPORTED) because variant envs do not chain the base
  sdkconfig.defaults.esp32, so bb_ws_server compiled to its stub.
- /ws/events handshake returns 101 Switching Protocols; was 400.
- contention stress: 344 frames delivered, seq 1 -> 1518, strictly
  monotonic (no stale render ever committed), zero malformed frames.
- latency A/B: notify-push mean 180ms / median 177ms vs tick-push mean
  213ms / median 222ms, consistent with eliminating a uniform 0-50ms
  sweep wait. Per-round samples overlap; only the aggregate separates.

Host coverage does not measure smoke_data_notify.c or the espidf path at
all, so the hardware run above is their only behavioral verification. An
automated assertion for the harness is tracked in B1-1457.

Claude-Session: https://claude.ai/... (continued)

8534 of 9074 branches covered (94.05%)

Branch coverage included in aggregate %.

14279 of 14434 relevant lines covered (98.93%)

2600.57 hits per line