Skip to content

fix(harmony): fresh Music guide reacquisition after ChannelId dropout - #506

Closed
masarray wants to merge 3 commits into
fix/harmony-p0-bounded-music-retarget-20261110from
fix/harmony-guide-dropout-fresh-reacquire-20261110
Closed

masarray wants to merge 3 commits into
fix/harmony-p0-bounded-music-retarget-20261110from
fix/harmony-guide-dropout-fresh-reacquire-20261110

Conversation

@masarray

@masarray masarray commented Nov 9, 2026 •

Copy link
Copy Markdown
Owner

P0 follow-on — no stale key flash after source dropout

Base: #503 exact head f3a29d66f95edd103e76d2fc94ba624b6ce0d450 (including its Music detector budget, bypass sample-rate fix and PCM matrix), stacked for safe isolation; #499 remains DRAFT verification only, NEVER MERGE.

Reproduction and root cause

The audio callback aged missing ChannelId guide PCM out after 2 seconds and fell back to Manual Key but did not reset MusicKeyDetector. On the first resumed PCM block, observeMusicTrack() called estimate() and immediately republished the stale pre-disconnect key (before acquiring new musical evidence). That creates a premature tonal retune and can cause crackles or an incorrect harmony burst after device/source recovery.

Fix

On the single transition across the existing 2s no-guide timeout, reset only guide analysis state and publish Manual fallback. No recurring reset each missing-source block, no new audio callback, no heap/lock, no extra audio route, no JUCE graph/device restart, no changes to the Harmony backing sound or declared PDC. On resumption guide must re-acquire fresh tonal evidence.

Regression

NativeHarmonyMusicFromOtherChannelIdAndStateMigration now sends real ChannelId music PCM, loses it for >2s while vocal FX continues, checks Manual fallback, resumes with different A-minor PCM and rejects stale C-major reactivation on first block, then verifies correct A-minor acquisition and unchanged source PCM.

Not physical certification: Full Windows native CTest and FAT/Inno + ASIO4ALL/real singer acceptance still needed. Retain this PR draft until exact-SHA evidence.

Live-singer extension (exact SHA db2bcf8c2f2091db061309de1775fe3aee838537)

The source-loss regression now drives 220Hz vocal PCM throughout the >2-second missing-guide interval, checking finite stereo output and a nonnegative observed vocal note. This rules out a test that passes merely because the singer is silent. Linux fast at #42723858125 SUCCESS. Windows PR quick at #42723858342 must be checked separately; draft Windows MSVC full CTest is disabled by repository policy. Do not count the older #503 FAT as qualification of this SHA.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Exact-head checkpoint 8ac12f471ba18226e9fb9cef96ff2bae6bd8ea72

Release gate remains closed: Need exact-SHA Windows full CTest + one optional FAT/Inno, hardware ASIO4ALL and singer playback before merge/promotion. Preserve #499 draft, main, and all independent workstreams.

@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: db2bcf8
Status: ✅  Deploy successful!
Preview URL: https://7055deaf.arsonmix.pages.dev
Branch Preview URL: https://fix-harmony-guide-dropout-fr.arsonmix.pages.dev

View logs

Copy link
Copy Markdown
Owner Author

P0 Windows exact-SHA qualification requested (2026-11-10)

CI-only mirror #508 was created against current main to satisfy the repository's owner-comment dispatcher policy, not to merge #499's integration history into main. Authorized comment /arsonmix-ci fat db2bcf8c2f2091db061309de1775fe3aee838537 succeeded in dispatcher #42730532660, and dispatched one exact-SHA Windows MSVC/CTest+FAT/Inno run #42730552143. Fast preflight passed; Windows job in progress. Do not claim installer artifacts or pass until run completes. Preserve both #506 and #508 as draft, do not merge to public main without native+physical ASIO and singer checks.

Copy link
Copy Markdown
Owner Author

Reconciled continuation: Current main 6bcf88af already contains some #503 Harmony work and M14 integration. Do NOT merge the entire old QA lineage from #506 into main. The focused current-main forward-port is draft #509 (three Harmony files only), tracked by integration contract #510. #506 still retains source-lineage exact-SHA Windows validation #42730552143; it does not qualify #509's SHA.

Copy link
Copy Markdown
Owner Author

Repository housekeeping (2026-11-10): Closing this historical/superseded draft without merging. The stale Music-guide reset after 2s dropout and continued singer-source PCM regression are now integrated on branded main via merged #518 (current 70cd760). The remaining separate moderate-confidence retarget / bounded Music detector CPU changes from the stacked predecessor are not declared delivered; preserve that work under still-open #503. This old #499-based stack must not be merged as a whole into current main.

No physical hardware FAT or new public release is implied.

@masarray masarray closed this Nov 10, 2026

Copy link
Copy Markdown
Owner Author

Final clarification after source/blob reconciliation (2026-11-10): PR #503 has also been closed as superseded. Its MusicKey engine and tests are byte-identical to current main@70cd7604; branded and bypass/dropout variants are carried by merged #517/#518. The earlier housekeeping note referring to #503 as still open is now outdated. No unreconciled Harmony branch is a safe independent merge target. Physical singer/ASIO FAT remains tracked in #412.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant