watchsettings.c: resync in-stage BGM (track 1) on Music commit and level entry (D355)

docs/dev/findings.md

@@ -640,6 +640,7 @@ covers D24–D69; the log continues in §H (D32 procedure, D70–D121).
 | D352 | **D350 review of 1303c0eb: stage transitions silently drop queued F10 setting changes (queue drained before the `stageActive` check); the UI thread's chooser state is unsynchronized; rapid slider drags push one RSP command per detent and re-persist the same value each frame; persist failures are silent (2026-09-27).** — full `## D352` entry at file tail | FIXED (port-only, `port/src/watchsettings.c`): hold-not-drop at stage transitions, spinlock on chooser state + front-screen guard, per-field coalesced queue commits, warning logs on all no-op paths, `PersistField` made static. NTSC build clean, headless menu smoke clean. |
 | D353 | **Turok-style settings regrouping: the D346 VIDEO mega-section (14–15 rows, at the MAX_PROWS limit) re-split into DISPLAY + GRAPHICS, the "AUDIO (BOND FILE)" hack header dropped (music/FX sliders fold into BOND FILE), all remaining rows converted to designated initializers, `buildPages` overflow now logs, and the Bond-file page title annotates the selected file (2026-09-27).** — full `## D353` entry at file tail | FIXED (port-only, `optionsoverlay.c`/`frontoptions.c`): 6 sections of 3–11 rows, all rows designated-initialized, `optionsRowIsBondChooser` accessor, build clean + headless smoke clean. Master-volume row, per-section reset (Gate E) and key rebinding remain in the M3 backlog. |
 | D354 | **Front-end F10 Bond rows silently disabled: `watchSettingsAvailable()` mis-routed F10-on-file-select through the stage path (returned 0), so with the hidden chooser stuck at "none" the volume sliders read "Select file" and did nothing; maintainer-reported (2026-09-27).** — full `## D354` entry at file tail | FIXED (port-only, `port/src/watchsettings.c`): all non-stage contexts route through the front path with a lazily-resolved default folder (explicit pick > `selected_folder_num` > first valid), one-shot game-thread `fileBuildWriteNewSave(FOLDER1)` fallback, GE_WSPROBE/GE_WSPROBE_FRONT env probes headless-verified (stage: live+save changed; front: default file 1 + save changed). |
+| D355 | **F10/options Music slider inaudible in-stage: the slider drives only the X-track (track 2) via `set_mTrack2Vol`, but in-stage BGM is track 1, which the game pins to `VOLUME_MAX` at level start (lv.c:364) and never rescales from `mTrack2Vol` in solo -- so neither the N64 watch nor the F10 could change the music the player was hearing; maintainer-reported (2026-09-27).** — full `## D355` entry at file tail | FIXED (port-only, `port/src/watchsettings.c`): the Music commit now also calls `musicTrack1ApplySeqpVol(v)` (the exact resync the MP path already does, mpmenu.c:354) and a once-per-activation resync runs on stage entry; by-ear confirmed by the maintainer. N64 watch path untouched. |
 
 Phase 2 replaced the Phase-1 demo loop with the real `mainproc()` on real OS
 threads, compiled GE's real `src/sched.c`, and brought in PD's fast3d software
@@ -13834,4 +13835,19 @@ Maintainer report: the F10 volume settings "don't apply", while the watch and th
   - `watchSettingsChooseFile` guard flipped to "refuse in stage context" (front-end control only); "none" remains reachable by explicit cycling.
   - `GE_WSPROBE_FRONT=<tick>` probe verifies the same front commit headlessly: default boot (menu) resolves folder 0 (file 1) with no interaction, commit 4096 → save.music 255→32.
 
-Verification: NTSC build clean; both probes pass headless; in-stage behavior unchanged (probe-identical). The D350 threading gate is unchanged in effect: the front path's `fileWriteSave` (EEPROM) now also runs from the F10 input thread in front contexts, but only one settings surface is ever active at a time and the game thread performs no save writes on menus, so writes stay serialized; reads (`saves[]`/CRC) are stable menu state. Manual by-ear check owed: F10 music/FX on file select, and default-file behavior after erase.
+Verification: NTSC build clean; both probes pass headless; in-stage behavior unchanged (probe-identical). The D350 threading gate is unchanged in effect: the front path's `fileWriteSave` (EEPROM) now also runs from the F10 input thread in front contexts, but only one settings surface is ever active at a time and the game thread performs no save writes on menus, so writes stay serialized; reads (`saves[]`/CRC) are stable menu state. Manual by-ear check owed: F10 music/FX on file select, and default-file behavior after erase. (By-ear confirmed by the maintainer in the D355 session: FX applies live; the music follow-on gap became D355.)
+
+## D355 — F10/options Music slider inaudible on in-stage BGM (2026-09-27)
+
+Maintainer report (after D354): F10 sound-FX changes apply in real time, but the Music slider does not; opening the in-game watch menu seemed to be what "applied" it. Diagnosis (port-only):
+
+- **Root cause: the slider and the audible music are on different tracks.** The Music row drives `mTrack2Vol` → `set_mTrack2Vol` → the **X-track (track 2)** CSP player (cutscene/stinger music). In-stage BGM, however, is **track 1**: solo level start plays `lvlPlayMusicTrack1(getmusictrack_or_randomtrack(...))` (lv.c:304) and the level-start reset pins **all three** players to `VOLUME_MAX` (lv.c:364–366) — it never rescales track 1 from `mTrack2Vol` in solo. The MP path is the exception: `mpmenu.c:354` explicitly feeds `get_mTrack2Vol()` into `musicTrack1ApplySeqpVol` at level entry, and `mp_music.c` mission-state transitions do the same before each `musicTrack1Play`.
+- **Why the watch "fixed" it:** the in-game watch re-applies `set_mTrack2Vol` every frame while its volume screen is drawn (`draw_music_volume_slider`), so the watch-screen/stinger music (X track) audibly tracked the slider — the in-stage BGM never did. A one-shot F10 commit has the same effect as the N64 watch: correct value stored, inaudible on BGM.
+- **Fix (`port/src/watchsettings.c`, port-side only):**
+  - `applyValue(WATCH_SETTING_MUSIC)` now calls `musicTrack1ApplySeqpVol(value)` after `set_mTrack2Vol(value)` — the *same* resync the game's own MP path performs (mpmenu.c:354), so F10/options Music edits are audible on the BGM player in real time (mirroring how FX always was via the global SFX scaler).
+  - **Once-per-activation resync** in `watchSettingsGameTick`: level start pins track 1 to `VOLUME_MAX` before `init_watch_at_start_of_stage` (lv.c:474, level load) re-applies the saved music value to the X track; the hook resyncs track 1 to `get_mTrack2Vol()` on the first active tick of each activation (`wsStageWasActive` reset whenever the stage is inactive), so the player's setting survives level starts, not just live edits. The `!=` guard keeps it a no-op when the game already set an equal value.
+  - Track 3 (MP BGM) is intentionally untouched — the game drives it from its own volume, and the PC port is solo.
+  - The N64 watch-menu path is **unchanged** (ground truth: the watch scales the X track only); this is a PC-settings-surface extension, in the same category as D350's existing `set_mTrack2Vol` call from the port.
+- GE_WSPROBE logs extended with the track-1 volume (`t1=`) so the resync is visible headless.
+
+Verification: NTSC build clean; **by-ear confirmed by the maintainer** — F10 Music now changes in-stage music in real time. The front-end commit path (input thread) now also calls `musicTrack1ApplySeqpVol`, which wraps `alCSPSetVol` on the track-1 player — same thread pattern as the pre-existing track-2 call, so the D350 threading gate is unchanged in effect.