dangernoodle-io / breadboard / 30771128505

97%

push

github

feat: add synthetic core-claimer smoke env to prove httpd steer (#1205)

Adds a Kconfig-gated (CONFIG_BB_SMOKE_CORE_CLAIM, default n) task in
examples/smoke that exclusively claims a core via bb_task_create()'s
core_owning=true path, ordered before bb_http_server_ensure_started()
via a new order=0 pre_http-tier manifest entry. Nothing in the tree
previously called bb_wdt_claim_core_exclusive() before httpd started,
so the httpd worker-core steering path added in bb_http.c had never
executed anywhere -- only exercised by isolated host tests.

A second manifest entry logs httpd's actual worker-task core (FreeRTOS
xTaskGetHandle("httpd") + xTaskGetCoreID()) on serial once the server
is up, giving a human-readable on-target observable for the steering
decision. It is wired at tier=regular rather than a requires=http_server
edge on tier=pre_http: render_source excludes the http_server-providing
entry from the tier=pre_http loop entirely, renders every OTHER
pre_http entry first in full, and only after that loop finishes emits
the captured __auto_type handle line for the excluded entry -- never
before. So a requires=http_server entry sitting IN tier=pre_http is
still rendered ahead of the capture line and its runtime guard is
always false -- the call never fires. tier=regular is unconditionally
rendered after the capture line, which is the only marker shape that
actually runs post-httpd-start.

The claimer spins a bounded, CPU-heavy burst then yields via
bb_task_delay_ms(1) (~98% duty cycle) rather than an unbounded loop --
representative CPU-bound contention without permanently starving its
claimed core; it never subscribes to the hardware Task WDT itself, and
core_owning=true excuses the claimed core's idle-task WDT check before
the task is created, so it cannot trip the WDT directly or indirectly.

New esp32-core-claim smoke env wired into the Makefile and CI matrix;
every existing env (knob off by default) is unaffected. Unicore
degrade-sanely behavior ... (continued)

8222 of 8786 branches covered (93.58%)

Branch coverage included in aggregate %.

13917 of 14072 relevant lines covered (98.9%)

2664.53 hits per line