rasprover: add ROS 2 CDR golden-bytes tests, fix BatteryState health enum

[rosterloh]

Two commits. The second one is a real bug fix and is the reason this PR matters.

SHA	Subject
790c59a	rasprover: add Ztest suite for ROS 2 CDR encoding
5061147	rasprover: fix BatteryState health enum (was OVERHEAT, not GOOD)

The bug

app_ros_encode_battery_state() wrote 2 for power_supply_health under a /* POWER_SUPPLY_HEALTH_GOOD */ comment. In sensor_msgs/BatteryState the constants are POWER_SUPPLY_HEALTH_UNKNOWN = 0, GOOD = 1, OVERHEAT = 2 (mirroring the Linux power_supply enum).

Every battery message the rover published claimed the pack was overheating. Nothing in firmware reads the value back, so it was only ever visible to a ROS 2 subscriber — exactly the class of defect that is invisible from the firmware side and that golden-bytes tests exist to catch. It was found while deriving expected wire bytes from the message IDL, not by running the code.

power_supply_status = 2 (DISCHARGING) and power_supply_technology = 0 (UNKNOWN) are both correct. The neighbouring writes now spell out their numeric values too — a bare 2 sitting next to a constant name is what let this through.

app_zenoh.c:298 is the only caller and just publishes the buffer. Message length unchanged at 73 bytes.

The suite

10 tests over app_ros_cdr.c, which is pure (writes into a caller-supplied buffer, no kernel/net/fs) and was previously uncovered for BatteryState.

• Golden byte arrays for BatteryState (73 B) and a one-joint JointState (72 B), derived from the message IDL and the CDR spec with the derivation shown in comments — not generated by running the code under test, which would assert nothing.
• Alignment/padding transitions, which is where hand-rolled CDR actually breaks: the 3-byte pad for the f32 after an empty frame_id, the absence of a pad for the u32 after four u8 flags (rel 48 already aligned), the 4-byte pad for the f64 after the velocity count. Alignment measured from the post-encapsulation origin, as FastCDR expects.
• Buffer-bounds behaviour at the trust boundary.
• The existing byte layout is otherwise correct — field order and alignment both match the IDL.

Platform: qemu_cortex_m3

unit_testing would be the natural choice for pure code, but it does not build on macOS: ztest registers suites via STRUCT_SECTION_ITERABLE → __attribute__((section(...))), which Mach-O rejects (mach-o section specifier requires a segment and section separated by a comma). native_sim is Linux-only. qemu_cortex_m3 ships with the Zephyr SDK and runs on any host; native_sim stays in platform_allow for Linux CI.

Verification

1 of 1 executed test configurations passed (100.00%), 0 failed, 0 errored
10 of 10 executed test cases passed (100.00%)

• mise run agent-build rasprover --sysbuild — succeeds.
• Pre-existing tests/rasprover/test_ros_cdr.py — 3 passed, unaffected (JointState only; it never covered BatteryState, which is why this gap existed).

Deliberately out of scope

• tests/rasprover/test_ros_cdr.py left in place despite some JointState overlap — consolidating test infrastructure is not this PR's job.
• No repo-wide twister task added: west twister -T applications currently fails on motor_controller (error: 'usb_dc_detach' is deprecated [-Werror=deprecated-declarations], 4 configs), so such a task would be red on arrival. That failure is pre-existing and worth its own look.

🤖 Generated with Claude Code