Repository navigation
Description
ArSonMix P0 cross-thread integration contract — checkpoint 2026-11-10
Purpose: prevent independent development threads from overwriting Harmony/PCM reliability when integrating Vocal Throw, Player, Compressor, WIRE, ASIO device recovery, updater, .smix/session, and release workflow. This issue is a coordination record, not a branch-protection rule and not a hardware certificate.
Verified live repo state
- Authoritative main inspected at
6bcf88af30629d2e20026bfa8e0a84163c06c29e(2026-11-10). Revalidate before every merge, because other threads push in parallel. - Harmony perf(harmony): P0–P2 realtime pitch, phase-safe analysis and rate-stable tone #487 (D4 head
f02f35e) and M14 integration DRAFT M14 — 80/80 Windows FAT qualified Vocal Throw + Player + Harmony (hardware trial) #499 (headc717be0) were both marked merged at2026-11-10T00:46:21Z, via merge commit6bcf88af. This is a repository fact, not proof of physical ASIO or release authorization. Do not revert whole merge without a surgical compatibility review. - Old QA fix(harmony): bounded Music guide key retarget under moderate confidence #503 (
f3a29d66) is draft over M14; fix(harmony): fresh Music guide reacquisition after ChannelId dropout #506 (db2bcf8c) is a draft follow-on over fix(harmony): bounded Music guide key retarget under moderate confidence #503; DRAFT EXACT-SHA Windows FAT qualification — Harmony Music/bypass on preserved M14 QA integration (DO NOT MERGE) #504 and DRAFT QA ONLY — exact-SHA Harmony guide dropout Windows FAT (#506) #508 are main-based CI mirror drafts, NEVER MERGE. Whole-branch merges from these historical lineages could revert or duplicate independently shipped changes. - #506 exact-SHA Windows FAT run #42730552143 was in progress during audit. Do not conflate its SHA with main or new work.
Current forward integration owner
PR #509 based on merge SHA 6bcf88af, current head 618cb67c476ffe41904175cc87890f075c9c3b21, DRAFT/NO AUTO-MERGE, only:
engine/builtin_fx/ChannelHarmonyProcessor.h: restore correct 8ms FxInstance bypass rate metadata during control-side JUCE prepare, plus one-shot reset of stale Music guide detector on >2-second missing-source timeout.tests/harmony_bypass_reliability_test.cpp: actual native FxInstance fade length at 48/96 kHz and 64/128/256 host partitions; maintains 9-way wet/space warm bypass PCM test.tests/harmony_music_bank_test.cpp: live 220Hz singer remains active during missing guide, no stale C-major flash upon reconnection, fresh A-minor acquisition and unchanged source PCM.
Do not copy complete files from #503/#506 into new main: main already contains other Music detector and bypass fixes. #509 is a focused, reviewed forward-port retaining latest main behavior.
Required merge/release gates
- Check branch diff relative to latest main, ensure only intended Harmony files changed and no overlapping thread owns them. Never force-push/rebase other threads' branches or remove P0 test coverage.
- Exact-SHA native build + Windows full MSVC/CTest must pass, especially actual JUCE
harmony_music_bank_testandharmony_bypass_reliability_test; CI fast alone does not test them on Windows. - FAT ZIP/Inno provenance at same exact qualified SHA, SHA-256/CRC verified; do not reuse earlier M14 or fix(harmony): fresh Music guide reacquisition after ChannelId dropout #506 installer as evidence for fix(harmony): reconcile approved M14 into current main without losing dropout/bypass reliability #509.
- Hardware owner trial: ASIO4ALL/device restart, 64/128/256 callback windows, 48/96k, sustained vocal, guide silence/disconnect/reconnect and key transition, rack/slot bypass and click/XRUN observation; verify audible behavior and latency. UI/session/WIRE regression checks unchanged.
- Never label a public release as clickless/zero-XRUN or natural-singer certified on CI alone. No automatic merge of QA CI mirrors DRAFT EXACT-SHA Windows FAT qualification — Harmony Music/bypass on preserved M14 QA integration (DO NOT MERGE) #504/DRAFT QA ONLY — exact-SHA Harmony guide dropout Windows FAT (#506) #508.
Thread handoff rule
Anyone touching ChannelHarmonyProcessor.h, HarmonyMusicKey.h, FxBypassEnvelope.h, LiveChannelFxGraph.h or native PCM routing must: (a) check newest main and PR #509, (b) explicitly show before/after output-latency and DSP PCM tests, (c) preserve warm bypass even when rack/slot OFF, (d) avoid duplicate graph/audio callback or GUI rebind gate, and (e) reference this issue in their PR.
No GitHub hard branch lock was configured by creating this issue. Owners should enable required checks/reviews for main and use staged integration where appropriate.
Activity
masarray commentedon Nov 10, 2026
Supplemental risk found: M15 Vocal Throw PR #507
Open draft #507 head
aca0441dis 7 ahead / 12 behind currentmainat6bcf88af; its PR base metadata still points to old9bd81a1b. The PR changed-files presentation includes many older Harmony files, while direct current-main SHA comparison shows 9 true M15 delta files. Critically the M15 branch snapshot includes oldHarmonyMusicKey.hmissingacceptMusicKeyCandidate, whereas current main contains it. Three-way reconciliation is required, never blind whole-file copy of M15 Harmony/PCM files. A safety/reconcile comment was added to #507. #509 remains the only current-main-focused missing Harmony fix branch, draft, and no CI-only mirrors are merge candidates.masarray commentedon Nov 10, 2026
Hardware owner report / Harmony continuation (2026-11-10)
Owner says Vocal Throw audio quality is still unacceptable. Treat M15 #507 as HOLD / DO NOT MERGE OR RELEASE, separately from Harmony. This reported quality failure is not license to alter native mixer PCM graph, WIRE, ASIO direct-device recovery, Harmony DSP or
.smixpersistence. Added explicit owner-field blocker to #507.Current-main surgical Harmony #509 remains only 3 files / source
618cb67c476ffe41904175cc87890f075c9c3b21. The previous combined QA #506 80/80 Windows CTest PASS, FAT/Inno artifacts verified by workflow, but not a test of #509 source. A single exact-SHA Windows MSVC/CTest + FAT/Inno #42735079669 has been launched; preflight SUCCESS, native Windows job running. Physical ASIO/singer and true no-crackle performance still not certified. Never confuse code-contract tests with release acceptance.masarray commentedon Nov 10, 2026
2026-11-10 P0 shared integration checkpoint — forward-port branch in progress
Canonical integration candidate: PR #512
fix/p0-reconcile-native-asio-harmony-20261110at exact head378ffbc2aba4b3de421f723270951365c563b981, based onmain@6bcf88af30629d2e20026bfa8e0a84163c06c29e.Do not merge or overwrite it with original old-base #503/#506/#507 or #511 wholesale.
5d0067d4: selectively forward-ported P0: stop fake native first paint + verified same-driver ASIO recovery (alpha.8 regression) #511's same registered ASIO vendor/pin restart and pristine SOURCE NONE rebind, while excluding its GUI startup hide overlay. Ported isolated 3-file fix(harmony): reconcile approved M14 into current main without losing dropout/bypass reliability #509 Harmony device-rate 8ms bypass/stale guide reset, native PCM tests and JUCE bridge handshake (no Pages-only-ready).378ffbc2: discovered true M14→M15 delta is only 9 paths, not the 99-file old-base PR; selectively carried DRAFT M15 — Smart end-of-phrase Vocal THROW + spatial ping-pong + first-use encoder #507 M15 phrase-end cue/ping-pong and tests, plus only onesrc/Main.cppproperty for native cue readback. No broad engine replacement. Current main Compressor/Player/WIRE/title/updater paths preserved.mainbefore any merge. No force-push of sibling heads and no automatic alpha release.Integration work is real code in draft PR #512; this note is an anti-regression checkpoint, not merge approval.
masarray commentedon Nov 10, 2026
2026-11-10 Harmonice P1 thread checkpoint (safe independent branch, DO NOT AUTO-MERGE)
6bcf88af30629d2e20026bfa8e0a84163c06c29eduring this audit.618cb67c476ffe41904175cc87890f075c9c3b21: Windows full native run #42735079669 80/80 CTest PASS; FAT and Inno artifacts tied to that SHA. Draft/unmerged.481e1b2e2a3c19440e8e37ebe215e89c0f94faa1. ShippingEngine::prepare()still uses 32ms PDC; 24/16ms candidates are explicitly offline-only, plus native PDC/PCM partition safety matrix, source-aligned WAV comparison, and p99.9 synthetic DSP CPU reports. Linux fast exact-head run #42753258905 PASS; Windows draft policy skips full MSVC/CTest, so no new Windows qualification claim.ChannelHarmonyProcessor.hand two Harmony tests; PRs P0: stop fake native first paint + verified same-driver ASIO recovery (alpha.8 regression) #511/DRAFT P0 Milestone A — device-independent native ChannelId logical session #513 separately own live mixer/device/session. DRAFT Harmonice P1 — isolated 32/24/16 ms latency experiment, PCM/PDC and aligned WAV A/B #514 touches NONE of those shared owner files; DRAFT Harmonice P1 CPU — sample-exact PSOLA epoch scan and same-runner tail profiling #515 CLOSED. Do not cherry-pick historical integration branches or apply DRAFT Harmonice P1 CPU — sample-exact PSOLA epoch scan and same-runner tail profiling #515 by default.Next legitimate stage: genuine vocal WAV A/B via
tools/harmony_offline_ab --latency-study, human auditory evaluation, Windows exact-head CI when gated candidate is approved, hardware ASIO test.masarray commentedon Nov 10, 2026
2026-11-10 Harmonice Milestone D/E integration checkpoint (same-main, exact patch audit)
Live main remains
6bcf88af30629d2e20026bfa8e0a84163c06c29eat inspection. I compared every line of the three #509 changed-file patches with PR #512. All three patches are byte-identical, includingChannelHarmonyProcessor.h, bypass PCM and Music ChannelId PCM regression. Hence if #512 merges first, #509's P0 forward-port is already represented and must NOT be separately duplicated; rebase only the independent P1 study on the then-current main and rerun relevant exact-SHA tests. No changes made to #512 or shared mixer/device/GUI files.The independent P1 study #514 head
ffe429ff992314ffbf1674469b783f487b19b00fretains 32ms live; source-aligned offline 24/16ms, four-case synthetic WAV/JSON corpus, and native PSOLA causal low-note gate. Physics and 44.1/48/96k real PCM show 16ms cannot produce PSOLA-ready samples at 65Hz (needs 18.46ms minimum), so REJECT 16ms as full-range candidate; 24ms only research until human singer A/B. Fast Linux Harmony CI #42756943860 PASS.The DO-NOT-MERGE QA mirror #516 is exact same
ffe429ffhead, based on main to request owner-authorized one-time full Windows #42757047991. Check final conclusion and all CTest results; no Windows PASS is claimed until verified. Remain drafts; no Vocal Throw, no ASIO hardware/audio quality certification.masarray commentedon Nov 10, 2026
Harmonice visible-brand contract — PR #517 (2026-11-10)
User requests Harmonice in built-in FX Picker and native GUI header instead of ArSon Harmony. Public name search found no exact VST/VST3 product collision in checked sources; this is not trademark clearance. Alternative Harmonis only if a real competing plugin is found.
Scoped PR #517 exact head
1d62d62a5cca6d7b9aea15429a58673cd6b24847based on currentmain@6bcf88af. Changes only UI/canonical display label, not engine DSP or device/routing. FX Picker, native JUCE getName, installed lease name, bridge catalog/fallback, GUI top title and engraved HARMONICE label are consistent.NativeMixerSessionFxRestorecanonicalizes only the built-in Harmony instance's display name on old.smixrestore to eliminate persistent 'ArSon Harmony' user-facing text, without rewriting the session or affecting VST3/preset/binary state. Source ID remainsarmix.builtin.arson-harmony.v1so no session and editor control bridge ABI break. Browser preview opener accepts both old/new names.Current exact-head UI quick run 42760956302 and Windows preflight 42760956304 PASS. One actual Windows CTest was explicitly dispatched at exact head 42761094407; full Chromium branding/editor test dispatched on earlier UI-identical head 42760463223. At checkpoint both expensive runs in progress: don't claim PASS prematurely.
Coordination: PR #512 and other mixer work modifies native Harmony, LiveAudioEngine, and session paths. On integration preserve #517's isolated display label + restore rule and #512's authoritative ASIO/mixer fixes, without renaming stable catalog ID or merging this draft automatically while full native checks pending. No physical ASIO claims.
masarray commentedon Nov 10, 2026
2026-11-10 Harmonice integration checkpoint — source-authoritative results
main@dc3540af8aafc3a113c37ef7c5a6f2593a1769fd: FX Picker/GUI/internal native lease/current .smix restore all presentHarmonice, immutablearmix.builtin.arson-harmony.v1. Exact PR SHA Windows 42761094407 80/80 CTest PASS.No unapproved ASIO/WIRE code modifications, Vocal Throw integration or public VST3-standalone artifact claim. Do not claim physical hardware qualification from synthetic CI.
masarray commentedon Nov 10, 2026
Harmonice reconciled main checkpoint — 10 November 2026
Verified final canonical
mainHEAD:70cd7604d808c6152803bc377839ff6c8353e393. Three completed merges: #517 native catalog/picker/editorHarmonicedisplay and backward-compatible.smixrestoration (stablearmix.builtin.arson-harmony.v1ID), #519 cold-start Chromium fixture updated to current acceptedAudio connection lostUI (exact-head full browser FAT SUCCESS), and #518 correct 8ms JUCE rate-prepared bypass + stale 2s Music-guide reset (exact-head Windows CTest 80/80 SUCCESS, Linux native DSP PASS). All are merged with GitHub acknowledgment, not merely draft.No loss and no duplicate integration: old #509 closed unmerged after byte-for-byte compare of 3 files against new main (ChannelHarmonyProcessor difference only approved name Harmonice). Old P1 #514/#516/#520 all closed unmerged and superseded by #521 current-main 9-file experimental draft at
8550ca7b613a90f4110660f7deeadc427b0dfa79; its fast Linux DSP 42783711206, native 42783711039, UI 42783711077 all PASS. No 24/16ms live latency change; 16ms disqualified for 65 Hz causal PSOLA; user singing & physical ASIO unverified.Still open outside this finished Harmony P0 scope: #512 ASIO/device/mixer architecture draft. Must reconcile separately against canonical main, exclude already merged Harmony hunks, and qualify hardware, not merge as-is. New main post-merge Windows 42783209892 was in progress when checked; do not claim post-merge job PASS until completed. This checkpoint preserves validated statuses, future integration rules, and target SHAs so other workstreams can proceed without regressions.
masarray commentedon Nov 10, 2026
Administrative closure (2026-11-10)
This issue is superseded as a coordination ledger, not "hardware completed." The historical draft references (#509/#503/#506 and QA mirrors) are retired; branded current-main repairs reached main via merged #517/#518. Its still-operative guards have been copied into the documentation in merged PR #524 and onto the concrete ownership issues #490/#412/#192.
Keep one authoritative JUCE PCM callback, off-thread prepared graphs, real sample-rate 8ms Harmonice warm bypass, 2s stale Music-guide recovery, unchanged immutable catalog ID, native intent/ACK/snapshot, exact-SHA Windows tests and separate physical singer/ASIO FAT. DO NOT use closing this coordination issue to waive any of those gates. #450 stays open as the release controller.