DeviceConfig — env-free library, config struct at the factory

[josephnef]

What

The library read 46 distinct DEVOURER_* environment variables at 72 call sites to select runtime behaviour — bring-up gates, beamforming arming, calibration skips, debug dumps, USB tuning. A library should not read CLI env vars to decide its behaviour; consumers get an explicit API instead.

src/DeviceConfig.h — construction-time configuration struct (rx / tx / bf / tuning / debug / usb sections), passed as CreateRtlDevice's defaulted fourth argument, so existing callers compile unchanged:

devourer::DeviceConfig cfg;
cfg.rx.keep_corrupted = true;      // FCS-failed frames to host
cfg.bf.arm_sounder = true;         // beamformer sounding engine
cfg.debug.log_writes = true;       // <wlog> register-write trace
auto dev = driver.CreateRtlDevice(handle, ctx, lock, cfg);

Every field is doc-tagged with its env-var spelling and value grammar. A MacAddr POD + parse_mac() replaces four copies of inline sscanf MAC parsing. Knobs that change mid-session remain runtime setters on IRtlDevice (SetTxMode, SetTxPowerOffsetQdb, SetRxPathMask, SetCcaMode, ...).

Plumbing — each component (EepromManager, RadioManagementModule, HalModule, FirmwareManager, HalJaguar2, HalJaguar3, RadioManagementJaguar3, Halrf8822e, RtlUsbAdapter, HopProf, ToneMask) takes the config at construction. Static-cached getenv lambdas (TX_EP, TX_TIMEOUT_MS, LOG_WRITES, TX_NDPA) become plain members — per-process knobs become per-device. parse_tx_mode_env leaves the library; BfReportDetect.h (demo-side per-packet debug hook) moves to examples/common/; the TMPDIR read in UsbDeviceLock becomes usb.lock_dir.

examples/common/env_config.{h,cpp} — the demos' env translator: every library-level DEVOURER_* var maps onto a DeviceConfig field there, so all env-driven test scripts run unchanged. The gate holds: grep -rn getenv src/ hal/ returns nothing.

Semantic note: per-channel-set re-read knobs (DUMP_CANARY, SKIP_TXPWR, FORCE_IQK, ...) are fixed at construction. Scripts export env before launch, so none break.

Validation

• Full all-chips build + two gated per-chip configs; ctest green on each
• txpwr_offset_regcheck: 49 PASS / 0 FAIL — bring-up register canary byte-identical to master on all 7 plugged DUTs (8812AU, 8821AU, 8814AU, 8822BU, 8811CU, 8822CU, 8812EU), the key no-regression invariant
• On-air d/d cell 8100/8000 hits (8822CU→8812EU, ch36)
• tone_mask_regcheck 6/6 (CSI mask + NBI land on every generation incl. 80 MHz)
• rx_path_mask_check 2/2 (mask spec incl. toggle grammar; functional + sticky)
• Jaguar2 single-radio BF self-sounding: 85k self-captured reports, armed 880 vs unarmed 0 (exercises enable_with_tx + sounder + NDPA in one process)
• DEVOURER_HOP_PROF emits per-hop lines with the flag, silent without

The regcheck's first run caught a real bug this PR also fixes: RtlJaguarDevice initialised _eepromManager/_radioManagement from _cfg before _cfg's declaration-order construction (UB, flaky per-process). _cfg is now declared first, with a comment pinning the ordering contract.

🤖 Generated with Claude Code