POWER-FINAL-26 — Bluetooth codec auto-selection

Status: INSTALLED, BLUETOOTH CODEC CHECK OPEN (2026-10-13; archive 121). Package out/y2linux-platform-v1-power-final-candidate-26/ (fallback = 25), receipts out/deep-sleep-final/build26-receipts/. BOOTIMG f4ca442b185a6a8561ef12901d68c9662c630a5be65f554776fbe885fe1d585a, Y2ROOT 66efd3ed0d4f727edeffa9dd6571f3d9c3e4ebe02647d8074b04c7cf09c2405f. Identity: build Y2LINUX-PLATFORM-V1-POWER-FINAL-26, kernel 6.18.56-y2linux-platform-v1-power-final-26, root 2026.08-platform-v1.56, release 1.0.0-platform-v1-power-final-candidate.26; Linux 64b3cc4, Reborn 1c9edd4. Platform as 25.

25 on the device (first check, archives 119–120)

Boot f0cc8b34 healthy, taint 0, DSI hand-over dark (backlight off until the first Linux frame → restored). The owner's card now scans complete: 311 files discovered, 0 failures (24: "11 traversal failures"). Cover Flow frame rate not measured (the owner filmed it instead; the sampler caught only one 60-fps burst). The framebuffer stream tool reads ≈10 fps whatever the transport (raw 9.8, gzip 9.3; SSH chacha20 carries ≈19 MB/s), so it cannot show a 60-fps glide.

Why only SBC was chosen

The platform already starts BlueALSA with every compiled codec (`--codec=AAC --codec=aptX --codec=aptX-HD --codec=LDAC,/run/y2/codec-runtime.json). Liveorg.bluealsa.Manager1.Codecs=A2DP-source:SBC,…:AAC`, …:aptX, …:aptX-HD, …:LDAC. Reborn accepted only the lowercase prefix a2dp-source:, so every codec, SBC included, was runtime_enabled=false; the policy failed (no_eligible_codec_PHYSICAL_GATE_or_distribution_gate), the codec choice (Auto included) stayed hidden and the saved preference was the old default, SBC. The AirPods Pro were on AAC only because BlueZ negotiated it at connect.

What 26 changes (Reborn `1c9edd4`)

The A2DP-source profile in Codecs matches case-insensitively (a2dp_source_codecs).
Default codec preference is Auto: LDAC → aptX-HD → aptX → AAC among mutually usable codecs, SBC fallback, requested
only while the PCM is idle (unchanged policy code).
Sessions saved before codec_revision 1 that hold SBC (the old default; the choice was never offered) move to Auto
once; another saved codec, or SBC chosen afterwards, is kept.

Tests

Reborn fmt, clippy -D warnings, workspace tests pass; the discovery test now uses the BlueALSA 5.0.0 names seen on the Y2, and both codec tests fail without the fix. Platform: validate.py PASS, ELF closure 354 PASS, installed ARM PASS, privacy clean, inventory 108 packages, container suite 693 OK.

Owner checks after the flash

Settings → Bluetooth codec shows Auto, SBC, AAC for the AirPods Pro; Auto selected.
Status bluetooth.codec_policy: runtime_enabled true for all five, selected/pcms[0].codec AAC after a
reconnect with playback stopped.
aptX / aptX-HD / LDAC need headphones that offer them (NOT_RUN).

26 on the device (archive 121, boot `9f7fe545`)

Kernel …power-final-26, taint 0, Reborn running with the new session (codec_preference Auto, codec_revision 1: the SBC → Auto migration worked). The owner could not play music: Reborn's output was still the AirPods Pro (14:14:7D:AA:34:47) and every play press failed with operation_failed "connected A2DP peer required". The AirPods were paired, bonded and trusted but never connected this boot: the platform reconnect (/run/y2/bt-reconnect.json) made two ConnectProfile calls, both ended without a reply (completion_unknown_user_retry_required), so automatic reconnect is Inhibited until a manual connect (tools/connectivity/reconnect.c, unchanged since 25). Reborn metrics: 0 Bluetooth connects, 0 disconnects; pcms empty, codec_policy none; codec history unchanged since 25 (last negotiated AAC). So 26's codec code never ran; the likely cause is the AirPods being in their case, out of range or held by the phone at boot.

Still open (device was unplugged for the overnight test before a manual connect was checked):

Manual connect (Settings → Bluetooth → AirPods Pro → Connect) works and lifts the inhibit.
Auto then requests AAC (26 is the first image that actually issues SelectCodec on connect) **without dropping the
link**; watch metrics.bluetooth_disconnects and the log right after the connect. If the request drops the link: skip SelectCodec when the negotiated codec already equals Auto's choice.
Design gap regardless of 26: when the saved Bluetooth output is not connected, Reborn only reports "connected A2DP
peer required"; it neither retries the connect on play nor falls back to the Y2's own output.

Overnight test start: refused sleep looked like a self-wake (archive 122)

From 20:16 to 21:17 every sleep request was refused because the cable was attached (suspend_refused_charger_or_usb_active, once a minute). The owner unplugged at ≈21:18:38; the next attempt read /sys/power/wakeup_count, quiesced the radios (≈4–5 s), and at 21:18:42 the charger noticed the missing source (status=2, NO_SOURCE) and published a BAT0 change (the only wakeup source with wakeup_count 1). At 21:18:44 the count store failed, so the helper refused (product_refused, wake_event_in_progress, no kernel entry) and restored the screen and radios: the "self-wake". The same refusal appears twice in the archived journals, both on 13; none on 14–25. The platform reconnect also tried the AirPods a third time after the radio restore and failed again (failures 3, still Inhibited).

Proposed platform fix (not made; needs the owner's OK and a 27 build): when the count store fails and neither the Power nor the RTC interrupt count changed, start a new handshake with a fresh count (bounded, e.g. 3 tries), as the maintenance-wake path already does after its BAT0 tick, instead of restoring the screen.