dangernoodle-io / breadboard / 30672282110

97%

push

github

fix: fold route-registry-cap audit into registration path (#1167)

bb_http_route_audit_init() was a boot-time bounds check wired as a
bbtool:init tier=regular server=true entry, meant to run last among
regular-tier init calls after every route is registered. It has never
fired: codegen's collect_entries walks components dependency-first, so
bb_http_server.h always parses before its dependents, and the tie-break
(order or +inf, parse_index) sorts unordered entries at +inf -- the
maximum key, so order= cannot move the audit after them. Proof: in
examples/smoke/main/generated/bb_app_init.c the audit was emitted at
line 198, before all 19 route registrations at lines 201-264, so it
always counted a near-empty registry and passed. In hand-wire mode
floor never called it at all (its only call site was the generated
file, documented as compiled-but-never-called).

Moves both behaviors (STRICT overflow escalation, cap-8 high-watermark
warn) into registry_add() in route_registry.c, mirroring the pattern
bb_dispatch_api_add() already uses for the sibling s_dispatch[]
registry. Folded into the registration path itself, the check is
order-independent by construction -- it runs at the write that actually
crosses the threshold, in both codegen and hand-wire composition, with
zero wiring.

CONFIG_BB_HTTP_ROUTE_REGISTRY_STRICT is bridged with a real C default
(matching BB_ROUTE_REGISTRY_CAP's existing bridge), not a bare #ifndef
shadow. Its assert() compiles out under -DBB_HTTP_TESTING (same
BB_*_TESTING convention as bb_data_scratch_acquire()), so host tests can
drive a real overflow through registry_add() without aborting the Unity
binary. Two host-testable seams cover it, both structurally excluded
from production via #ifdef BB_HTTP_TESTING (not left to incidental
--gc-sections dead-stripping -- confirmed by nm on the esp32 firmware.elf:
only bb_http_route_registry_strict_would_fire, which production also
calls, is present; the four test-only accessors and their b... (continued)

8103 of 8676 branches covered (93.4%)

Branch coverage included in aggregate %.

24 of 24 new or added lines in 1 file covered. (100.0%)

13743 of 13898 relevant lines covered (98.88%)

2628.83 hits per line