[pranayD21]

Update (Nov 9): rebased onto main's src/ layout as a single commit. ./flight_test passes 26/26 and now includes tests/ethernet; pio run -e bench also builds. Networking is still disabled.

Decision needed before DHCP: #180 proposes a static 192.168.2.81 fallback, but #171 has since moved every other board to a MAC-derived 169.254.x.y recovery address and takes the board ID from the lease. #513 says to copy the other boards. @inez9, which should the flight computer follow? Details are in src/app/ETHERNET-DAQ.md.

---

The flight computer could identify its W5500 but had no Ethernet data path. This adds a bounded UDP driver and a separate network task that sends the existing version-0 BOARD_HEARTBEAT and SENSOR_DATA packets. Sensor/GNSS producers copy selected fields into a nonblocking queue while retaining eMMC logging.

Following issue #180, the disabled default profile now proposes board ID 81, board 192.168.2.81:5005, subnet 255.255.255.0, and server 192.168.2.20:5006. This supersedes the earlier local .20:5006 placeholder. A local unicast MAC is derived from the STM32's full 96-bit UID; the physical board's resulting MAC still needs collision checking and registration.

Discovery-enabled operation broadcasts heartbeats until a valid SERVER_HEARTBEAT or supported SENSOR_CONFIG arrives, pins the responding server, then forgets it after ten seconds of silence. Invalid, unmapped and actuator packets cannot claim or refresh that association. Fixed-server filtering remains available. The task drops stale samples, stops telemetry without configuration, and recovers from link/SPI/send failures. It never dispatches actuator or flight-state commands.

Validation:

• Six Ethernet suites pass with ASan/UBSan and in Release mode, including shared-library wire compatibility, W5500 ring handling, record conversion, production task execution, discovery/identity, and disabled defaults.
• Production-task tests cover learning a different server, returning to broadcast after expiry, queue saturation, send timeout, SPI cleanup and cable loss.
• Clean-rebuild mutation checks prove that broken server pinning, broadcast selection and ten-second expiry fail the new tests.
• STM32H723VET6 PlatformIO flight build passes: 73,204 bytes flash and 45,452 bytes RAM. Diff checks pass.

Deployment remains incomplete; networking stays disabled pending the sign-off requested in #180:

• DHCP, lease renewal/rebinding and the five-second static fallback are not implemented. Discovery of the DAQ server is separate from acquiring a DHCP lease.
• Option A (new FLIGHT_DATA) versus Option B (existing ten-channel SENSOR_DATA) remains an unchecked decision in Flight computer Ethernet: network, channels, DAQ routing (decisions for #513) #180. This PR still implements Option B. Option A needs a shared framing/version/batching contract, decimation and a health-record definition/producer; REC_HEALTH currently has only a type constant.
• The DAQ needs flight-computer routing, the proposed Elodin 0x26 schema, config-broadcast support, and backend/frontend decoding. Unknown boards currently fall back to PT interpretation. The network SELF_TEST handshake stage is also outstanding and is distinct from the radio SelfTest report in Flight computer: health task, IWDG through Idle sleep, SelfTest frame at start-up #179.
• A flight-only bench profile needs an agreed way through config_broadcast's designated-survivor requirement. No live profile or service has been changed.
• Physical MAC registration, network sign-off and hardware validation remain outstanding. No board was flashed or contacted.

ETHERNET-DAQ.md records the implemented behavior, proposed configuration, and remaining work from #180. This PR does not close #180 or STARProject #513.

🤖 Generated with Claude Code