[laniol8926]

Debug APK

Fork PR — CI builds the debug APK on ft8af/** changes but can't post a comment here (read-only token). See the Debug APK for manual testing job's summary page on this PR's checks tab once it finishes.

Device test

• Needs device test (labeled needs-device-test): changes MicRecorder's capture format (float → PCM16) and its reinitialize/restart lifecycle, which affects USB-audio and standard-mic RX on every rig, not just the one it was found on.
• Emulator is enough

Manual test checklist

1. Connect a rig over USB audio (CAT rig of your choice) and start RX. Expected: waterfall/decodes appear as before.
2. In Settings, switch the audio input device while RX is running. Expected: capture actually switches to the new device (previously it silently kept using whatever it opened with until the next USB reconnect/app restart).
3. Unplug/replug the USB audio device a few times in quick succession. Expected: capture keeps recovering every time, including after picking a device right after a replug (previously one transient AudioRecord creation failure right after a replug could permanently stop capture for the rest of the app session).
4. Confirm decode quality/latency on at least one non-QMX rig (e.g. IC-7300/705, a CM108/Digirig-style interface, or a Bluetooth SCO mic) is unchanged from before this PR — this is the main open question this PR exists to answer.
5. Pick the USB rig as the input device in Settings, then power-cycle the radio (not just unplug/replug). Expected: the picker still shows the rig selected (not "Default") after reopening Settings, and RX keeps working without having to manually re-pick it.

What changed

Split out of PR #925 (QRP Labs QMX via hamlib) per review feedback: that PR found and fixed three MicRecorder bugs during an all-night real-hardware RX investigation (full narrative in that PR's comment thread), but none of the three are QMX-specific — they affect every rig's RX audio capture — so they're reviewed/tested here independently of the QMX catalog change.

1. Picking a new input device in Settings never reached the live capture. The device-picker's onSelect persisted the new device id but never told the already-running MicRecorder to rebuild with it — it only reads that field at construction time. Fixed by calling the existing reinitializeAudioInput() from the picker, mirroring the RX-channel-change and Bluetooth-SCO-connect paths, which already did this for the same reason.
2. A transient AudioRecord creation failure permanently stopped capture for the rest of the app session. reinitialize() decided whether to restart by reading isRunning, but a transient failure (observed right after a USB replug/power-cycle, before the audio HAL caught up) sets isRunning = false as a side effect with no owner action — so every later reinitialize() (even the operator picking a perfectly valid device afterward) saw isRunning already false and silently skipped restarting. Fixed with a separate shouldBeRunning intent flag, touched only by start()/stopRecord(), with the restart decision keyed off that instead (shouldRestartAfterReinitialize, now unit-tested).
3. Requesting ENCODING_PCM_FLOAT silently breaks capture on at least one real MediaTek USB audio HAL. Found via a precise real-hardware A/B (same running app process, same CAT mode, same device — only the HAL's negotiated format differed across a USB re-enumeration): every float-format open decoded nothing, every PCM16-format open decoded normally. The HAL's own log explains why: it falls back to a 16-bit word size internally while still being asked to hand back 4-byte floats, producing unusable samples with no error surfaced anywhere. Fixed by requesting ENCODING_PCM_16BIT and converting to float in software in the reader loop (int16ToFloat).
4. A USB device's AudioDeviceInfo id is not stable across a reconnect — confirmed on real hardware when a QMX power-cycle alone reverted the Settings input picker to "Default". MicRecorder.findAudioDeviceById, FT8TransmitSignal.findAudioDeviceById, and AudioDeviceSpinnerAdapter.getPositionByDeviceId all matched only by exact numeric id, so a stale id after any USB re-enumeration silently fell back to the default device/display, even though the app already persists the device's VID:PID on every pick. Fixed with UsbAudioDevice.resolveDeviceIndex: exact id match first, else fall back to the product name of the live USB device whose VID:PID matches what was persisted (product name doesn't change across reconnects). Applied symmetrically to both input and output.
   • Two follow-up bugs surfaced via real on-phone dumpsys data while testing Fix 26 bugs: crashes, resource leaks, encoding errors, RTL support #4: (a) picking a framework AudioManager USB entry (as opposed to the separate USB-direct list) always persisted VID:PID 0:0 — nothing populated it for that path — so the fallback above had nothing to fall back to; fixed by looking up the matching raw UsbDevice by product name at pick time (AudioDeviceSpinnerAdapter.getUsbFrameworkDeviceProductName + UsbAudioDevice.matchByProductName) and persisting its VID:PID, in both the Compose picker and the legacy ConfigFragment. (b) Even with VID:PID persisted correctly, UsbDevice.getProductName() ("QMX Transceiver") never equals the AudioDeviceInfo list's name for the same card ("USB-Audio - QMX Transceiver") — the generic USB-Audio HAL decorates the raw name — so the match needed to be case-insensitive containment, not equality.

Risks

Medium. This changes the capture audio format and the restart/reinit lifecycle for every rig's RX path, not just the one it was diagnosed on (a Motorola Edge 2022, Android 15, QMX over USB Audio Class via MediaTek's audio HAL). PCM16 is the format Android's AudioRecord guarantees support for, and 24-bit-capable USB codecs losing precision to 16-bit is not meaningful for FT8/FT4 decoding — but this hasn't been smoke-tested against other USB audio interfaces (IC-7300/705, CM108/Digirig) or Bluetooth SCO mics, which is exactly what this PR's own manual test checklist (item 4) and QA pass should confirm before merge.

Tested automatically

• MicRecorderInt16ToFloatTest (new): the PCM16→float conversion applied to every capture buffer — full-scale positive/negative, silence, and partial-length reads.
• MicRecorderRestartDecisionTest (new): shouldRestartAfterReinitialize — restarts when a transient failure left isRunning false without stopRecord(); does not restart after stopRecord() itself cleared shouldBeRunning.
• UsbAudioDeviceResolveIndexTest (new): resolveDeviceIndex — exact id match, product-name fallback when the id isn't found (incl. the real decorated-name mismatch), no match when there's nothing to fall back to, null product-name entries don't throw.
• UsbAudioDeviceNameMatchTest (new): unambiguousNameMatchIndex (the pure decision behind matchByProductName) — single containment match, no match, ambiguous match (two candidates) refuses to guess, blank/null input, null entries in the candidate list.

Code review

• Reviewer used: not yet requested — opening this PR first per the split Patrick asked for on PR Add QRP Labs QMX via the hamlib backend #925's review.
• Every finding addressed or answered

QA evidence

(to be filled in by QA)

🤖 Generated with Claude Code

https://claude.ai/code/session_01SaooA8HLsiETQ12Tbwwyzf

[laniol8926]

Couldn't self-apply the needs-device-test label from this fork (no permission on the upstream repo) — flagging it here instead: this touches USB-audio capture format and the input-device reinit lifecycle for every rig, so it needs real-hardware testing per AGENTS.md's "Device testing" criteria, not just the emulator.

[codecov]

Codecov Report

❌ Patch coverage is 0% with 27 lines in your changes missing coverage. Please review.
✅ Project coverage is 41.92%. Comparing base (f702100) to head (fd03b07).

Files with missing lines	Patch %	Lines
...dio/ks3ckc/ft8af/ui/settings/RadioAudioSettings.kt	0.00%	27 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##                dev     #949      +/-   ##
============================================
+ Coverage     41.70%   41.92%   +0.22%     
  Complexity      252      252              
============================================
  Files           280      280              
  Lines         34002    34105     +103     
  Branches       4076     4088      +12     
============================================
+ Hits          14180    14299     +119     
+ Misses        19531    19494      -37     
- Partials        291      312      +21     

Flag	Coverage Δ
android	18.18% <0.00%> (+0.51%)	⬆️
native	12.70% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
...dio/ks3ckc/ft8af/ui/settings/RadioAudioSettings.kt	0.25% <0.00%> (-0.01%)	⬇️

... and 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.