dangernoodle-io / TaipanMiner / 26201763086

90%

push

github

fix(sdkconfig): enable LWIP_TCPIP_CORE_LOCKING to avoid bb_http abort race (#418)

Under SSE + polling load on bitaxe-650 / esp32-wroom32, the device aborts
in tcpip_send_msg_wait_sem (tcpip.c:461) inside the bb_http chunked JSON
emitter. The non-core-locking lwip path posts a tcpip_msg to the tcpip
task mbox and waits on netconn->op_completed_sem; if the peer closes the
socket mid-stream and the netconn is freed before the next chunk, the
LWIP_ASSERT on the (now invalid) sem fires and panics.

/api/pool now emits ~200 chunks per response after the per-pool stats
array landed (PR #413), so the race window per request is wide. Each
chunk is a separate lwip_send and re-runs the assert path.

Enabling CONFIG_LWIP_TCPIP_CORE_LOCKING takes the lwip global lock and
runs the callback inline instead of round-tripping through the tcpip
mbox + sem — the assert path doesn't execute at all. Modest perf cost;
well-supported in IDF.

The durable fix belongs in bb_http (short-circuit chunked writes on
peer-closed netconns) and is filed as breadboard B1-223.

2498 of 3164 branches covered (78.95%)

Branch coverage included in aggregate %.

5061 of 5236 relevant lines covered (96.66%)

157713.76 hits per line