dangernoodle-io / breadboard / 28564007362

100%

push

github

refactor(bb_pub): back store-forward ring with bb_pool in a private arena (B1-478 PR D, B1-419) (#655)

migrates bb_pub's store-and-forward ring buffer off the lazily
heap-allocated bb_ring_t (never freed from the global heap — B1-419)
onto a bb_pool FIFO carved from a bb_pub-private static-BSS bb_arena
(s_ring_arena_buf). No shared "app-pool" component is introduced; each
bb_pub consumer of bb_pool owns its own arena per B1-478's design
decision (SSE gets its own in PR E).

- ring_pool_get() replaces buffer_get(); lazily creates a bb_pool FIFO
  (EVICT_OLDEST) from the private arena on first enqueue (on-failure
  mode) or eagerly at init (always-on mode via bb_pub_buffer_init_eager)
- bb_pub_buffer_stats: count/dropped sourced from bb_pool_count/dropped;
  truncated is always 0 (buffer_capture already rejects oversized
  entries before ever calling bb_pool_push, same as the pre-migration
  bb_ring behavior)
- bb_pub_test_reset: bb_arena_reset + recreate instead of destroy+
  recreate, preserving the same dropped-counter-resets-to-zero semantics
  with zero heap churn in tests
- idle-tick bookkeeping (s_buffer_idle_ticks) is retained purely so the
  bb_pub_test_set_idle_free_ticks() test hook keeps a stable API/ABI
  surface; reaching the threshold no longer destroys/recreates storage
  since there is no heap allocation left to reclaim once the ring lives
  in a permanent static-BSS arena
- JSON cache: NOT added. bb_pub's current serialize-once-per-tick path
  already frees the heap JSON string within the same tick (no
  cross-phase heap hold), so there is nothing here to migrate onto
  bb_pool. Externally-consumed JSON output is unchanged (same
  bb_json_serialize/bb_json_free_str call sites, unmodified).

net RAM: store-and-forward is opt-in (CONFIG_BB_PUB_BUFFER_ENABLE=n
default) and unchanged for boards that don't enable it. For boards that
do: the ring's ~8-9 KB moves from an on-failure-lazy heap allocation to
a permanent static-BSS reservation (... (continued)

1840 of 1840 branches covered (100.0%)

Branch coverage included in aggregate %.

3167 of 3167 relevant lines covered (100.0%)

1301.34 hits per line