Repository navigation
feat(firmware): add dedicated MR60 USB configuration and readback - #2543
proffesor-for-testing wants to merge 4 commits into
Conversation
proffesor-for-testing
left a comment
There was a problem hiding this comment.
NO BLOCK for the bounded USB producer-contract source review at 94f394242f658e1fab4df1cbd4107207ac4193b5. This is a read-only Codex review from a separate session, using Dragan's GitHub identity; COMMENT, not formal human approval or merge/flash authorization.
The two findings are closed:
- EXECUTED: earlier code accepted a trailing-dot target as parser-OK/claim-valid. The final production parser rejects it.
- EXECUTED + STATIC: earlier code accepted a leading-zero first octet as private decimal10. The actual sender uses inet_pton, and ESP-IDF5.4's pinned esp-lwip implementation delegates to ip4addr_aton, which interprets leading0 as octal (e.g.010 →8). The final parser rejects multi-digit leading-zero octets while retaining single0.
Executed against the final production parser with strict C11/Wall/Wextra/Werror: valid private/single-zero addresses pass; trailing dots, leading-zero octets, repeated dots, out-of-range octets, public/loopback/hostname/IPv6 values reject. Port1023/65536 reject;1024/65535 parse. Wrong device/node/nonce, expired claims and raw/radar port collisions reject. No device/NVS/hardware write was executed by this review.
STATIC checked surfaces: bounded command/token/base64 handling; stable device and saved-node binding; nonce age/one-attempt handling; dedicated-profile capability and generic-command denial; committed-NVS readback before success; explicit uncertain-commit handling; raw/node/other-key preservation; secret-safe replies; firmware-support versus fresh-radar distinction. I found no additional demonstrated blocker in that bounded review.
Limits: I did not execute a dedicated ESP-IDF5.4 image build, physical USB configuration/reboot/network-change acceptance, or sensing/medical qualification. Reported host/GCC/ASAN and6.0.2 image checks belong to the implementation session; new-head CI/build/hardware evidence must remain separately identified. Keep draft until its dependencies and required current-head qualification are complete. A CONFIG ACK alone is not sensor readiness.
The Mac can identify generic ESP32 boards but cannot configure the dedicated MR60 controller’s separate radar destination. Reusing the generic CONFIG command can rewrite its raw CSI port.
Adds a dedicated, nonce/device-ID-bound MR60 USB configuration protocol on the XIAO-C6 radar build. It preserves saved node identity and raw CSI port, writes only the shared receiver IPv4/radar port and explicitly changed Wi-Fi keys, reads committed settings before ACK, and refreshes HELLO from NVS after uncertain writes. Generic builds refuse the radar command; dedicated radar builds refuse generic CSI configuration. No hostname/mDNS claim.
Base: existing radar PR #2528,
5302592e6695bc1ae0a164c9ea6f19a8ef5a7a4c, targetingfix/mr60-carrier-freshness-20261111. Changes are confined to serial onboarding, its host tests/stubs/spec, and the dedicated MR60 profile’s factory raw-port default55063. Existing NVS overrides and generic defaults are unchanged. No parser/sender/antenna/runtime server/workflow edits.Validation executed:
make host_testswith the existing verified cJSON dependency directory.Bounded independent producer review is NO BLOCK (COMMENT, not approval/merge/flash authorization). Current CI has53passing checks,5skips and Rust Workspace Tests pending. Draft: final CI/dependency landing and separately authorized physical USB apply/reboot/fresh-radar acceptance remain open. No device writes, flashing, OTA, main/held-head updates, server-helper replacement or release. Current PR head
94f394242f658e1fab4df1cbd4107207ac4193b5; the parent remains frozen. Follow-up commits keep host checks compatible with the IDF container and reject trailing-dot/leading-zero IPv4 literals.Paired consumer: cognitum-one/ruview-mobile#286, tracking issue#278. Contract:
firmware/esp32-csi-node/test/MR60-USB-CONFIG.md. The app requires a new nonce-bound post-reboot readback and a fresh radar packet; capability/ACK alone is not sensor-ready. Existing node IDs are preserved; this does not add automatic multi-radar ID allocation.