fix: Treat CRC and header errors as RF noise, not radio faults (#91)

[cwill747]

Fixes #91.

Diagnosis

The CRC and header errors are real RF events, not a bug. In the log attached to the issue, the failing packets arrive at RSSI −113 dBm, SNR −7 dB, and the SX1262 reports DeviceErrors=0x0000. The corrupt frame at 16:02:25.326 is the same RESPONSE relayed one more hop, received with a corrupt routing prefix. Decoding the clean copy per Packet.read_from() gives path_len=0x8a (3-byte hashes, 10 hops, 30 path bytes), so its payload starts at offset 32 with length 20 — matching the payload length the dispatcher logged. One further 3-byte hop predicts a 55-byte frame whose payload starts at offset 35, and that is exactly the length the SX1262 reported and exactly where the payload bytes sit. The 20-byte payload is byte-identical to the clean copy. But the frame arrived with header=0x0c and path_len=0x11, which would place the payload at offset 23 with length 32 — inconsistent with its own contents. Four bytes are corrupt (the header, path_len, and two path bytes); the payload came through clean. IRQ=0x0042 is RX_DONE | CRC_ERR with no HEADER_ERR bit, so the LoRa physical header decoded and the payload CRC then failed. The radio is healthy.

The defect is ours: RadioErrorHandler treated every WARNING from a radio-layer logger as a fault. pyMC_core logs reception noise at WARNING, so normal traffic flipped the header badge to "Radio Error" and flooded the console.

Changes

meshcore/logging_setup.py

• classify_rf_noise() separates reception noise (CRC error, header error or corrupted header, empty packet) from real radio errors.
• RadioErrorHandler counts the noise instead of reporting it. Only real errors reach the badge and the toasts.
• RfNoiseCounter keeps per-category totals and writes one summary line per minute, so the rate stays visible without spam.
• RfNoiseFilter on the stderr handler drops the noise from the console. The rotating file handler still records every line for bug reports, and at DEBUG console level the filter steps aside.

ui_gtk/views/settings.py

• New read-only RX Noise row in the Logging panel. It refreshes every 2 s while the panel is mapped.

Tests

10 new tests in tests/unit/test_logging_setup.py. Full suite: 129 passed, 1 skipped. ruff check, ruff format --check, and mypy are clean on the touched files.

Review notes

• I could not render the settings panel. The development environment has no X or Wayland display, and the flake provides no Xvfb. The RX Noise row needs a run on the Pi or in mock mode before merge.
• RX timeout detected stays classified as a real error. It does not appear in the report, so silencing it would be a guess.
• NoiseFloor=-113.8dBm at 62.5 kHz bandwidth is roughly 8-10 dB above a quiet band. That is typical for SPI and display noise inside a uConsole, and it explains why weak packets fail CRC. It is an antenna and shielding matter, not software.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 74da943eaf

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".