dangernoodle-io / breadboard / 25073845293

100%

push

github

feat(bb_registry): add handler-lifecycle registry; migrate ota_pull, ota_push, log_routes, info (#123)

New bb_registry component holds (name, init_fn) entries that opt-in components self-register at link time via the BB_REGISTRY_REGISTER(name, fn) macro. The app calls bb_registry_init(server) once after bb_http_server_start; the registry walks every entry in registration order and invokes its init.

Migrated four components — bb_ota_pull, bb_ota_push, bb_log_routes (two entries: stream + status routes), bb_info — and removed their public _register_handler / _register_routes symbols. The smoke esp32-wroom-32 main shrinks from 4 register calls to one bb_registry_init.

Why a new component, not folded into bb_http: bb_http already owns a route descriptor registry for OpenAPI metadata; this is a handler-lifecycle registry, different purpose. Co-locating would conflate the two and force every registry user to REQUIRES bb_http even without HTTP routes.

Linker quirk worth knowing: PlatformIO's espidf builder strips ESP-IDF's WHOLE_ARCHIVE flag, so constructor-only translation units get garbage-collected by --gc-sections. The cmake/bb_registry.cmake helper bb_registry_force_register() injects a -u <symbol> link option per registry entry to force-keep the .o. Verified on hardware: registry init reports 5 entries, all four migrated routes respond correctly.

Folding bb_log_routes back into bb_log core is Phase 2.1 (paired with Kconfig) — it can't safely happen until Kconfig can conditionally drop the routes source + bb_http dep, otherwise headless bb_log consumers regress.

272 of 272 branches covered (100.0%)

Branch coverage included in aggregate %.

431 of 431 relevant lines covered (100.0%)

147.89 hits per line