Skip to content

[P1][ARCHITECTURE] Long-term maintainability: consolidate mixer/FX host ownership and reduce high-coupling modules #527

Description

@masarray

Milestone B verified triage — 2026-11-17: P1 / architecture / refactor / planned. Classification ≠ completion of product feature or physical FAT.

Scope / release boundary: Long-term single-owner architecture; do not fork realtime callback.

Exit criteria:

  • Audit and report actual remaining mixer/FX host ownership coupling at current main instead of repeating completed M1–M6 milestones.
  • Refactor by bounded subsystem with deterministic native PCM/ACK/.smix regression and no additional device/audio callback; retain independent rollback.

Canonical issue ledger. Preserve historical evidence below; no speculative merging or rewriting approved audio.


M0 workpack and M1 implementation merged

M1 merged: PR #530, main@68e74ee332c3258900c2bbce2a8da635a3d0b62c. The control service owns logical transactions, .smix and ACK/readback. The integration projection is canonical. The built-in FX registry serves live insert, offline prepare, restore and picker metadata. The original WIRE socket ripple follows verified native readback. Windows native qualification passed 80/80 tests on e3f7f627c81a4c2dafb75165a230a66186b31eec. Fast native, fast UI and Linux DSP checks passed. Targeted Chromium covered public and native WIRE impact and the native FX editors. Physical ASIO FAT remains PENDING. M2 will isolate device apply and rollback while preserving the accepted session.


Purpose — long-term professional mixer/VST host maintainability

Audit baseline: main@99f580d76bf20ea994d54f8916098c680897f41d (2026-11-11 Asia/Jakarta). This issue is a source-grounded engineering debt and extensibility epic, not permission for a bulk rewrite, a competing audio engine, a GUI redesign, or a release claim. Keep existing open product/field-test ownership in #490, #412, #192, #15 and protected release policy in AGENTS.md.

Verified source inventory / maintainability hotspots

Area Current baseline Architectural concern
src/Main.cpp 6,499 lines ConsoleWindow hosts UI/browser events, device flows, FX/vendor editors, sessions, Player, telemetry and operations; high regression blast radius
src/LiveAudioEngine.h 3,564 lines Device lifecycle, graph/VST3 ownership, output gating, callback and multiple processing policies interact in one class
src/LiveDryRoutingLab.h 3,261 lines Historical M4/Lab semantics coexist with production ChannelId graph, buses, FX and session transitions
web/approved-native-bridge.js 3,646 lines Native adapter also handles numerous feature-specific handlers and DOM projection, increasing event/ACK coupling
web/monitor.js 2,118 lines Device, routing, monitoring and diagnostic presentation are coupled; confirm exact assembled-GUI inclusion before moving code
src/ApprovedNativeV8Projection.h and integration/ApprovedNativeV8Projection.h 179 and 196 lines Identical nativeV8Decimal helper duplicated; projection variants must be given explicit, single ownership and retain installed-vs-retained authority safety
docs/ 238 top-level entries Historical design, handoff and newer implementation truth need discoverable CURRENT / PARTIAL / HISTORICAL status; e.g. 2026-11-07 device-independent ADR still says “NOT YET IMPLEMENTED” although logical Milestone A PR #513 has merged, while hardware binding remains outstanding

Important qualifications: web/styles.css is large mostly because it embeds a font as base64; raw byte size alone is not evidence of bloated CSS rules. The repository has extensive tests/ coverage and selective CI, so test existence is a strength; test counts are not a substitute for real hardware FAT. Existing engine/core, engine/device, engine/builtin_fx, engine/player, integration, contracts and .smix storage are a foundation to consolidate, not discard.

Architecture target — preserve ONE authority end-to-end

ONE approved /preview/ GUI (native WebView uses exact assembled renderer)
    -> typed Intent (requestId, expected doc/device/graph revision)
    -> control-thread command handler / use-case service
    -> single logical MixerSessionDocument + ChannelId/EdgeId model
    -> validated, preprepared native routing/FX graph
    -> atomic immutable publisher -> ONE JUCE audio callback -> real devices
    -> physical readback + explicit accepted ACK/snapshot -> same GUI

Device discovery/restart, plugin scanning/state/editor lifetime, player decode, session persistence and WebView DOM must be independent adapters/services around this path; never become a second PCM graph or bypass ACK/revision. ASIO primary monitoring must never wait for WASAPI secondary output, disk, JavaScript or vendor UI. SOURCE ONLY must preserve native channel processing while disconnecting output only; device loss must never erase the logical session. No locks, allocations, disk, blocking calls or unsafe ownership reclamation in callback.

Coherent implementation milestones (avoid micro-PR churn)

  • M1 — Contract and ownership consolidation. Document a concise current architecture/dependency map from actual code with one owner per domain: logical session, device identity, graph compiler, PCM publisher, FX lease, player, .smix storage, GUI adapter, diagnostics. Identify C++/JS control event names and same-revision ACK/readback semantics. Resolve the duplicated projection helper through one canonical implementation with installed/retained regression tests; do not accidentally admit offline/retained PCM as LIVE. Archive or mark superseded historical docs instead of adding more competing handoffs. Scope: no audible behavior change.
  • M2 — Decouple ConsoleWindow incrementally. First isolate native Device/Session/FX/Player control use cases behind narrow interfaces; keep src/Main.cpp as composition/startup shell and WebView wiring. Move existing code rather than create redundant wrapper layers; each slice removes real coupling and retains one renderer. Test each migrated intent, negative ACK, readback and saved state before moving the next.
  • M3 — Consolidate production graph ownership. Trace actual live route from LiveAudioEngine / LiveDryRoutingLab / DynamicNativeSessionPublisher to render callback and plugin lease. Remove obsolete M4/Lab-only compatibility only after a regression fixture proves complete ChannelId routing, SOURCE ONLY, BUS FX, fader/PAN/MUTE/SOLO, VST editor identity across move/reorder, device restart and rollback. Avoid change-of-behavior plus mass rename in one patch.
  • M4 — Make new FX an extension, not a host rewrite. A future built-in FX should use common instance lifecycle, parameter validation, serialization, bypass, host metering and editor/ACK protocol. Prove by adding a test fixture processor/parameter through normal registry without new per-plugin switch cases scattered across Main/Engine/GUI. Preserve existing Harmony, Compressor and Vocal Throw latency/PDC and real PCM semantics.
  • M5 — Lightweight architecture fitness checks. Add cheap source-boundary/contract compatibility checks on relevant PR paths, deterministic routing and session migration fixtures, and targeted static analysis/formatting for touched C++ and JS. Do NOT impose arbitrary file-size gates, mask tests, or run duplicate Windows/Chromium jobs. Use already established exact-SHA quick → UI and/or FAT dispatch. Record callback deadline, allocations and retention behavior using real PCM fixtures.
  • M6 — Maintainable onboarding and docs lifecycle. One short CONTRIBUTING / CURRENT_ARCHITECTURE navigation entry should identify the supported build, extension points, dangerous realtime rules, relevant exact tests, diagnostics and owners; make stale ADRs visibly historical. A new agent/contributor must find the correct source of truth without interpreting hundreds of dated plans.

Required regression/acceptance contract for each relevant milestone

  1. User product stability: first launch with no physical audio still shows genuine editable native model; cold start/device swap preserves .smix, ChannelId, routing, VST slots and accepted state; no fake meter or forced monitor ritual. Preserve the one approved /preview/.
  2. Real audio truth: physical IN -> channel FX -> optional BUS FX -> MAIN; CH -> SOURCE ONLY silences MAIN while retaining CH meter; disconnect/reconnect WIRE; device/buffer restart; no unintended duplicated send. Hardware-specific FAT remains PENDING unless actually performed and witnessed (see P0 native audio FAT — verify WIRE, SOURCE ONLY, BUS/FX and device recovery #412).
  3. Plugin lifecycle: native and external VST3 instance identity, bypass, state, editor windows, slot reorder/remove and controlled disposal survive structural edits without callback destruction or unintended resets.
  4. Performance / safety: no new callback, no blocking/allocations/locks in RT; unchanged measured latency/PDC under the same render graph; rejected and stale intents leave the last accepted graph intact.
  5. Engineering agility: for each new feature, record modules touched, why, tests and any coupling introduced; plugin/channel features must not require editing unrelated device or release code. Demonstrate reduced hotspot responsibility, not merely relocated line counts.
  6. Qualify honestly: cheap tests on each in-scope change; one milestone ui and/or fat lane only when appropriate, no duplicate jobs; report exact commit, code PASS/FAIL, hardware FAT PENDING/PASS separately. New documentation alone must not trigger audio binaries.

Coordination

Deliverable for this audit: engineering baseline, credible risk ranking and gated follow-through. No implementation, test run, merge, or physical-FAT success is claimed by creating this tracking issue.

Activity

  1. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    2026-11-11 — Maintainability recovery / M2a takeover checkpoint

    A new agent has taken over after M1 PR #530 merged on main@68e74ee332c3258900c2bbce2a8da635a3d0b62c.

    M2a code: PR #534, exact head 02b7c2d25af47b874e5734984f0edf6917397b31 (7 files). This is real source work, not a documentation-only plan.

    • New src/NativeDeviceApplyCoordinator.h owns one JUCE control-thread Apply/restart lease, native request parsing, 64-bit-safe physical pin selection, direct driver opening, and evidence-based installed/rollback readback. No new AudioDeviceManager or callback.
    • Main.cpp delegates these duties while retaining device/session binding and output safety. The last accepted .smix document, VST3 processor leases and DSP remain untouched.
    • New hardware-independent JUCE CTest checks reentrant leases, output-only with 0 inputs, forged/duplicate/missing pins, exact accepted readback and ASIO rollback identity.
    • Source-path-only contract assertions were moved to the new owner without weakening the underlying safety checks. Fast native, fast UI and Harmony DSP CI have PASS at current head.
    • One exact-SHA Windows MSVC/CTest qualification is in progress. DO NOT merge or claim full PASS before final status. Physical ASIO FAT still PENDING. No fat/installer lane was requested.

    M2b after M2a verifies: extract remaining applyCustomAudioDevice transaction states and device rebind/rollback orchestration from Main.cpp, align modal Audio Setup with non-destructive session lifecycle, and verify real same-driver ASIO recovery, changed pin maps, rate/buffer restart, SOURCE ONLY, bus/FX, and protected MAIN output. Reconcile against active draft #512, not a bulk merge. Maintain #529 Harmonice independently.

    Issue #527 stays open until M2+ and field safety acceptance are proven.

  2. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    M2a MERGED (2026-11-11): PR #534 squash commit 4eb541af90a8d9c1876fb49a014b7d7b34ebf62f. Exact SHA 02b7c2d25af47b874e5734984f0edf6917397b31 passed Windows MSVC and 81/81 CTest in run #42897246430; fast native/UI/Harmony checks also passed. No physical ASIO FAT. Next M2b: serialize native JUCE modal Audio Setup with explicit Apply and ASIO restart through one existing NativeDeviceApplyCoordinator lease, preventing nested driver opens and repeated setup mute; preserve .smix graph/FX and validated physical pins. Do not close #527.

  3. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    M2b continuation (coding delivered, verification underway): PR #538, head 2d426a91f17cb51ce9ef7609ba733e48c9d6f496, builds directly on merged M2a main@4eb541af. All three physical device mutation paths (explicit Apply, native JUCE Setup dialog, automatic ASIO recovery) now share one tagged control-thread lease; the modal selector holds it for its entire visible lifecycle through close/rebind, repeated setup clicks do not re-mute, and shutdown releases the lease before window destruction. Added JUCE no-device lease lifecycle tests and native UI-source regression assertions. Fast native PASS. One exact-SHA Windows MSVC/full CTest in progress. No physical ASIO FAT yet. Do not merge before final qualification; do not close this epic; #512 remains separate and potentially conflicting.

  4. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    2026-11-11 continuation checkpoint — M2b merged; M2c code delivered

    • M2b merged: PR #538, main@29fef13e285c7be11f6541ffb0679747dc670ec2. Exact candidate 2d426a91f17cb51ce9ef7609ba733e48c9d6f496 passed Windows MSVC 81/81 CTest, run #42902171866. One JUCE control-thread lease now covers native modal Setup, explicit Apply and ASIO restart, preserving session state and preventing reentrant device mutations.
    • M2c under review: PR #540, head cf2ecdbc9313d3cceb1df9fc62821b2368020593, rebased via force-with-lease onto merged M2b content (one-source-owner, five-file diff). Production readback uses actual JUCE device object and verifies ASIO driver/name/masks/rate/buffer; accepted mixer session rebind selection is now a pure tested control-plane decision. Rejects missing/zero/nonfinite live audio format. Hardware-independent C++ tests and source-contract tests added.
    • Current PR SHA fast native PASS: run #42906822527. One exact-SHA MSVC/full CTest dispatched: run #42906988016, IN PROGRESS, not yet merged. No expensive FAT duplicate.
    • Physical ASIO/ASIO4ALL hardware FAT PENDING; software CTest does not prove glitch-free monitoring. Draft audio reconciliation P0 reconcile: preserve native mixer, recover verified ASIO and clickless Harmony without hiding GUI #512 remains separate.
    • Issue [P1][ARCHITECTURE] Long-term maintainability: consolidate mixer/FX host ownership and reduce high-coupling modules #527 was already closed by repository owner. This checkpoint does not change that state; the PR remains the live engineering evidence.
  5. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    2026-11-11 maintainability checkpoint: M2c merged; M3a production instrumentation

    M2c merged into main: PR #540, squash c48bdce0d680c020cf6236cf1882331a29c13305, exact candidate cf2ecdbc9313d3cceb1df9fc62821b2368020593 qualified on Windows run #42906988016, 81/81 CTest PASS.

    M3a coding complete, pending Windows qualification: PR #542 head c1591090bc71aa5cb18e04e262e0d83e8a603b8e. A pure realtime atomic ProductionRenderAudit is now called by the single JUCE lab renderer and read from control status to distinguish real native ChannelId vs M4 PCM render attempts/successes/failures, overlap denial, muted and unpatched exits. Adds one C++ CTest, source call graph and architecture map; does NOT change PCM/DSP, plugin lifetime, GUI or remove M4. Seven files in PR.

    Quick native + Harmony DSP PASS on exact M3a HEAD. Initial Windows qualification stopped at a single legacy JS source-format regex (323/324 tests), fixed by requiring the same pre-DSP mute/silence safety sequence plus audit hook; revised SHA fast checks PASS. One exact-SHA Windows MSVC/CTest run #42912789221 is IN_PROGRESS, result PENDING. Do not merge #542 without CTest PASS and live main reconciliation. Real hardware ASIO/ASIO4ALL FAT PENDING.

    Roadmap issue #527 was closed by repo owner; status untouched. Draft #512 still overlaps LiveDryRoutingLab and must be reconciled selectively.

  6. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    Maintainer checkpoint — 11 Nov 2026 (live source and CI)

    Completed and merged into main

    M4 native FX host extension (not yet merged)

    Still open (NO false-completion claims):

    Source constraints: one approved /preview/, one JUCE audio callback, no RT allocation/blocking, .smix and Hotkey H2 preserved. Run selective exact-SHA CI without repeated Windows/FAT, keep maintainability epic #527 OPEN until M3–M6 acceptance really completes.

    2026-11-11 M3b continuation: PR #560 source head 8f8b4503e374507907f509e237403d39f3f4a7 introduces a conservative snapshot/epoch-qualified native render evidence classifier and C++ regression. It never deletes M4, authorizes output, or claims physical FAT. Fast native PASS, one native exact-SHA qualification requested via PR comment. M4 #553 reconciled 3-file scope; source-equivalent older 84/84 CTest PASS, new head 13121450eef46488ce51d1cfabfeb5b137c39b67 native Windows #42932151152 IN PROGRESS, not merge-qualified at this checkpoint. Keep issue open.

    2026-11-11 qualified milestone update

    2026-11-11 additional milestones (current thread)

    • M3b render-evidence gate MERGED: PR #560 squash main@c62ccfa41c9b6f805db1647dc2f40c47edf1abc1. Exact candidate 8f8b4503e374507907f509e237403d39f3f4a7 passed MSVC/full CTest 87/87. Classifier only assesses cumulative native/M4 render counters as bounded same-epoch windows, never authorizes or removes M4. Physical FAT still PENDING.
    • M5 safety follow-on M5: reject malformed native FX lease snapshots without corrupting accepted GUI state #561 PR, candidate b143e59d2e09e395bdadeffc72dab7a54343d7a3: native FX instance token and catalog validation on accepted UI snapshots; reject malformed/duplicate cross-channel tokens and >8 slots; preserve previously accepted state and pending ACK on invalid snapshot. Reconciled historical sidechain test to real public schema (id/catalogId), and old Move/Remove test now distinguishes invalid transport versus valid-but-wrong FX readback. Exact source quick native + quick UI PASS; one full Chromium qualification #42941894526 RUNNING. Not yet merged at this checkpoint; no redundant Windows CTest dispatched (JS-only).

    M6 native producer safety PR #562 (2026-11-11): code and tests HEAD a3bde75b096f5fe6e8bd878528f36dfbd79bd685, 2 files only. The canonical JUCE CONTROL projection rejects duplicate VST3 instance leases across channel/rack positions and >8 slots; JUCE regression asserts rejection leaves existing native snapshot unchanged. Fast native PASS; one exact-SHA Windows MSVC/full CTest #42942394921 running. NOT merged and hardware FAT NOT proven. Independent JS fail-closed M5 #561 is awaiting full Chromium #42941894526. Do not close issue #527.

    2026-11-11 latest verified merges

  7. masarray commented on Nov 11, 2026

    @masarray
    OwnerAuthor

    2026-11-11 maintainability implementation continuation (GitHub connector)

    Verified against current main@ff651726c0226f3532da5288c1bd99909d7de14a. This is code integration, not a closing claim for epic #527.

    Still incomplete: M2 real device restart / ASIO pin identity physical FAT; M3 real native window PCM and safe M4 compatibility retirement; full M4 parameter/editor/host lifecycle convergence; further M5 bridge domain separation; broader M6 architecture guardrails. Physical ASIO / actual vendor VST3 listening PENDING. Preserve one approved /preview/, one JUCE callback, one installed ChannelId graph, exact .smix state/FX identity, real ACK+later readback. Do not close #527 or #412 based on CI.

  8. masarray commented on Nov 11, 2026

    @masarray
    OwnerAuthor

    Maintainability execution update — 2026-11-11, follow-up

    Current verified main@01e4a97f97efbdf5a1d7e4a3daf3d9e75c6960f4 (no new physical-audio FAT claimed).

    Remaining: further ConsoleWindow control-use-case separation; full M4 editor/parameter/lease shared lifecycle; M3 compatibility retirement only after installed callback PCM proof; M2 device/ASIO restart and vendor editor physical FAT; M6 broad, selective architecture guardrails. Separate Harmony #529/#521, physical multi-output #563, native recovery #512 remain untouched. One approved /preview/, one JUCE callback, no RT allocations/blocking and one accepted ChannelId PCM authority retained. Keep epic #527 OPEN.

  9. masarray commented on Nov 11, 2026

    @masarray
    OwnerAuthor

    Final completion of 2026-11-11 maintainability follow-up

    The remaining M4 PR #567 has now MERGED to main, squash b5b03669c616ac55a5d935023e0402ec7a99f7be. Live main verified at exactly that SHA. Original exact PR head 556eb2cb8ebf03fcd64fb0c25ba0915d62f1ee11 passed fast native and Windows MSVC / 88/88 realtime CTest PASS in https://github.com/masarray/arsonmix/actions/runs/42953943679 (run completed SUCCESS). Shared NativeFxControlRequest is now used by all four built-in FX handlers and the native registry-extension C++ test covers exact identities/decimal mailbox sequences. No audio callback/FX DSP changes.

    This completes the following three actual merged PRs in this conversation:

    Epic #527 remains OPEN: remaining actual coupling reduction in ConsoleWindow / FX editor lifecycle; M3 legacy M4 render retirement requires installed native PCM and safe fallback proof; M2 real physical ASIO restart/rebind/VST3 continuity FAT remains PENDING; M6 selective architecture fitness checks remain ongoing. Concurrent Harmony/physical multichannel work must be reconciled from live main before touching their files. No physical FAT, acoustic performance or release certification is claimed.

  10. masarray commented on Nov 11, 2026

    @masarray
    OwnerAuthor

    Live cross-thread compatibility checkpoint — M4 typed FX envelope (#569)

    Verified main: b5b03669c616ac55a5d935023e0402ec7a99f7be.
    New scoped PR: #569 head 798f3151ad728fded6af77e31a15396ad82cc95e; only src/NativeFxControlRequest.h and tests/native_fx_registry_extension_test.cpp changed. Fast native PASS; one exact-SHA MSVC/full CTest https://github.com/masarray/arsonmix/actions/runs/42968789458 IN PROGRESS, not yet qualified/merged.

    Semantics preserved: valid int slot, string requestId and optional string workspaceToken; logical FX on deviceEpoch "0" remains separate from installed PCM and retains the mailbox sequence. Reject newly-proven ambiguous input types: fractional/bool/string/int64 slots, numeric requestId, numeric workspaceToken. One JUCE callback and one approved /preview/; no Harmony DSP or physical-output changes. When #569 passes Windows and merges, active Harmony and O integrators must re-check their final combined HEAD before release/merge. No physical ASIO FAT proven here.

  11. masarray commented on Nov 11, 2026

    @masarray
    OwnerAuthor

    M6 cross-lane source-of-truth integration (2026-11-11)

    Merged: #570, squash 5374ffc0d55e547ee96bdefd43bff4c7b4b55046; docs-only CONTRIBUTING.md, architecture README and feature extension playbook now describe current merged #565/#567 FX host, #564/#568 GUI readback, and unmerged Harmony #529/multi-output #563 contracts. As per smart CI policy, no heavy CI was dispatched for doc-only changes.

    Code candidate: #569 head 798f3151ad728fded6af77e31a15396ad82cc95e: typed native FX control envelope for all built-in editors and Harmonice preset, with regression that LOGICAL_ONLY epoch 0 remains editable and preset hpres- cannot be confused with DSP mailbox harm-. Fast native PASS. Windows MSVC/full CTest exact-SHA run 42968789458 was IN PROGRESS at this comment; do not merge/claim PASS prematurely.

    Source compatibility details delivered on Harmony #529 and output O #563. Their current PR HEADs must still be tested as final combined source, with physical ASIO FAT pending. Epic #527 remains OPEN.

  12. masarray commented on Nov 11, 2026

    @masarray
    OwnerAuthor

    Final execution checkpoint — 2026-11-11, concurrent Harmony/O1 compatible delivery

    Current verified main@fd2dd0c4d9cb78abd34c6f12e28f33f0e8a55b74. Two PRs completed and MERGED this continuation:

    1. #569 M4 typed native FX control envelope, merge fd2dd0c4d9cb78abd34c6f12e28f33f0e8a55b74, exact qualified head 798f3151ad728fded6af77e31a15396ad82cc95e. FX target validation is now fail-closed for non-integer slots, numeric/bool request IDs and non-string logical workspace tokens across EQ, Compressor, Harmonice and Vocal Throw. Regression covers logical-only epoch "0" with mailbox persistence, stable 64-bit IDs, Harmonice hpres- preset namespace vs harm- DSP. Fast native PASS; Windows MSVC 88/88 CTest PASS and zero failures: https://github.com/masarray/arsonmix/actions/runs/42968789458. No DSP, RT callback, device, WebView or PCM graph edited.
    2. #570 M6 source-of-truth and cross-thread extension guidance, merge 5374ffc0d55e547ee96bdefd43bff4c7b4b55046. CONTRIBUTING.md, feature-extension playbook and current architecture map now reflect merged M4/M5 owners, logical-vs-live FX request identity, Harmonice preset commands, O1/O2 branch boundaries and physical FAT caveat. Docs-only, no paid CI invoked.

    Reconciled live concurrent work AFTER these merges: Harmonice #529 currently head 6548800cf708e5d7263e6e7986e69680432f6e34 and its existing NativeFxControlRequest still equals the historical main blob; its changed files do not include either #569 changed path. Native UI payload type survey supports new stricter policy. Standalone Harmony CI on its own head is not combined HEAD proof. Physical multi-output #563 currently head 50b0be1058be2e947a4c733e564057a82aaf6623 and GitHub reports mergeable_state=dirty / behind main by 4; this predates current M4 Main.cpp and M5 pure presentation approved-native-bridge.js changes, so O1 branch owner must reconcile semantic overlaps. No competing branch was overwritten. Updated contextual compatibility comments on both PRs (Harmony and O1). No fake claims of O1/Harmony audio performance.

    Still open in epic #527: M2 physical ASIO/restart/vendor editor field FAT, M3 legacy-M4 PCM retirement after witnessed native ChannelId renders, M4 broader vendor editor/parameter host lifecycle consolidation, M5 further native GUI adapter decoupling after O1/Harmonice integration, and incremental M6 fitness checks. Keep #527 OPEN, physical sound/ASIO FAT PENDING, approved /preview/ and one JUCE PCM callback untouched. No owner actions/manual CI needed for this checkpoint.

  13. masarray commented on Nov 11, 2026

    @masarray
    OwnerAuthor

    Maintainability continuation — live implementation checkpoint (2026-11-11)

    Latest checked baseline main@fd2dd0c4d9cb78abd34c6f12e28f33f0e8a55b74; DO NOT close #527 or infer physical FAT.

    Two new real source-change PRs are now open and have fast native PASS:

    • #571, HEAD e1d541bded2a4ba371019a2d395e5d02f45f9654, M2/M3 native CONTROL safety: strict JUCE device name types (reject numeric/boolean/array driver or hardware endpoint IDs), and ProductionRenderAudit::assessSeries requiring 3+ consecutive non-silent native-only render windows with independently witnessed same-process/device-epoch/admission plus a nonzero callback count before classifying sustained ChannelId ownership. Negative fixtures cover M4, mid-callback, epoch/restart, reset, mute and insufficient native activity. 5 files (including tests/docs), no JUCE callback, graph or UI changes. Fast native PASS, one exact-SHA Windows native job 42974020854 running; no merge/PASS claim yet.
    • #572, HEAD 4497119e2fd03fa5c1b2f3f30eb8ba8a240cc75e, M4 host-wide built-in prepare guard: common EQ/Compressor/Harmonice/Throw factory rejects NaN, Inf, invalid sample rates, excessive or impossible hardware quantum and broken state buffers before constructing any processor. Fixture constructor counter proves early rejection, keeps valid 16-frame host boundary. 2 files. Fast native PASS, one exact-SHA Windows native job 42975094052 running; no merge/PASS claim yet.

    Concurrent product work: O1/O2 #563 independently reconciled the prior src/Main.cpp / bridge conflict on cdc73973… (behind main 0, GitHub clean at checkpoint, full Chromium PASS). Its branch has advanced since that qualification; final updated HEAD still requires appropriate CI. Harmonice #529 remains actively changing DSP/FX editor. Neither #571 nor #572 touches those active files. Examined O1 combined source and confirmed it retains shared NativeFxControlRequest::parse, pure protocol.operatorMessage delegation and external output handling. No other branch force-updated.

    Still unresolved: real physical ASIO multi-output/callback continuity, production Harmony crackle/perceptual improvements, M3 legacy M4 retirement (requires field evidence), and full M4 ConsoleWindow-specific FX lease/editor use-case extraction after Harmony/O1 stabilization. No hardware-listening or release claim here; no redundant FAT ZIP or Chromium requested for the two CONTROL-only PRs.

  14. masarray commented on Nov 17, 2026

    @masarray
    OwnerAuthor

    CURRENT MAIN HANDOFF — 2026-11-18 (supersedes older draft notes)

    • Canonical modular MVVM and Multi-Device IO architecture: #763 merged at 1cbe62779319bb6505e38984111ade648ebdf42d. Technical authority: docs/architecture/ADR_MODULAR_MVVM_FEASIBILITY_20261118.md, docs/PRD_ARSONMIX_MULTI_DEVICE_IO_20261118.md, docs/ARD_ARSONMIX_MULTI_DEVICE_IO_V2_20261118.md. These are design documents, not installed dual-ASIO functionality.
    • Owner-friendly completion + nonblocking parallel development: #764 merged at current main 3c8dfd2ffeb31568bd34c5085c8972df9c16c8bd. New root AGENTS.md requires every working ChatGPT agent to close its own task (including failure/partial completion) with HASIL, STATUS, AMAN DIPAKAI?, TINDAKAN ANDA, BUKTI, LANGKAH BERIKUTNYA in simple Indonesian. Same completion note belongs in PR on significant milestones, not in a shared lock ledger. Detailed guide: docs/architecture/OWNER_FRIENDLY_TASK_HANDOFF_AND_MULTI_IO_20261118.md, with IO-0 through IO-5 modular feature slices. The old four-lane owner/collision policy is explicitly archived and must NOT be revived.
    • Verified repo safety: after docs: mandatory owner-friendly task reports + nonblocking multi-device handoff #764 merge, AGENTS.md still contains docs: adopt ArSonMix multi-device I/O PRD/ARD + AGENTS and incremental MVVM rules #763 Multi-Device IO contracts; no native renderer, VST3, PCM, GUI assets, CMake, installer, CI workflows, rulesets or output permissions changed. ArSonMix / Required CI Merge Gate passed on PR head 296a1525be34239d9659010f1a1217c35b883cf9 before exact-head squash merge. Hardware FAT is NOT APPLICABLE to docs: mandatory owner-friendly task reports + nonblocking multi-device handoff #764, and still PENDING for new multi-device functionality.
    • Nonblocking rule: no exclusive file/domain owner, no waiting for other threads/old draft PRs, no mandatory overlap status check. Agents can code on independent branches immediately and reconcile tested changes at the merge boundary; normal product safety CI remains.
    • Next real implementation: first a proven Device Setup ViewModel vertical or isolated secondary capture-only fixture with synthetic drift/underrun and physical device feasibility. Maintain one native mixer render graph with potentially multiple I/O-only callbacks, existing primary ASIO zero-extra-bridge path, stable ChannelId/EdgeId and .smix, and one approved /preview/ UI.
    • Operator action now: TIDAK PERLU MELAKUKAN APA-APA; no new multi-device binary was produced by the docs PRs. Do not ask the owner to install an old build to test unimplemented dual-device functionality.

    Important: The new instruction cannot automatically notify separate past ChatGPT conversations. Each future agent must read main AGENTS.md and report in its own conversation + PR. Keep issue #527 open until actual modular source implementation and regression evidence, not merely architecture documents.

  15. masarray commented on Nov 17, 2026

    @masarray
    OwnerAuthor

    2026-11-18 — Chat/context-cost audit (nonblocking, no new governance)

    The owner reports ChatGPT sessions frequently reach a limit before a coding milestone is complete. Do not conflate platform usage/rate limits (unobservable from GitHub) with context-window pressure. We can reduce the latter with targeted retrieval and modular code, not new ownership locks or excessive documentation.

    Measured on exact main@3c8dfd2ffeb31568bd34c5085c8972df9c16c8bd:

    • AGENTS.md: 26,137 characters / 198 lines.
    • Nine selected architecture/runbook/PRD docs combined: 159,272 characters, not counting code.
    • src/Main.cpp: 489,266 characters / 8,873 lines.
    • web/approved-native-bridge.js: 219,878 / 4,283 lines.
    • src/LiveAudioEngine.h: 205,034 / 3,637 lines.
    • web/mixer-io.js: 124,198 / 2,082 lines.
    • CMakeLists.txt: 107,970 / 1,608 lines.

    These values indicate high source-reading and context cost, not a measured model quota or actual token count. Do not pull all these files into every thread. Earlier ChatGPT reports and broad PR diff/log dumps compound the cost.

    Recommended engineering response, without a new CI gate or PR dependency:

    1. Keep a concise front-of-entrypoint navigation in existing AGENTS.md so new agents read the short current product invariants first, then only relevant feature docs/source ranges. Consolidate duplicated/stale guidance incrementally without removing requirements. Do not create another lengthy ADR to describe this.
    2. Prefer GitHub code search + fetch_file(start_line,end_line), individual PR patches, and filtered CI failure excerpts over entire 8k-line files/100k-character logs. Return small task-relevant excerpts, not full tool payloads.
    3. One meaningful vertical product milestone per conversation, preserving exact main/PR HEAD and user-facing HASIL/STATUS/AMAN/TINDAKAN/BUKTI/BERIKUTNYA in the PR as durable handoff. At context pressure, commit/test/report partial actual state rather than claim background continuation.
    4. Implement the already merged modular MVVM plan (docs: adopt ArSonMix multi-device I/O PRD/ARD + AGENTS and incremental MVVM rules #763) incrementally: pure Device Setup VM, remove migrated state from mixer-io.js, extract one real native application service out of Main.cpp while preserving current command/ACK/schema and single audio render owner. Measure unrelated files touched per feature, reading footprint, and regression failures—not a speculative token savings percentage.
    5. Multi-device I/O: start small with independent capture-only fixture/clock-domain bridge and safe native source mapping from the merged PRD/ARD. Additional I/O callbacks must not become a second mixer/VST3 renderer. Owner hardware FAT still pending.
    6. This is advisory task optimization, NOT exclusive ownership, a new waiting queue, a context-budget CI rejection or permission gate. Never block another active feature PR because of this issue.

    User action: NONE. This comment performs no runtime code change and does not fix ChatGPT platform quota.

  16. masarray commented on Nov 17, 2026

    @masarray
    OwnerAuthor

    Independent audit of the two new parallel threads — 2026-11-18 (nonblocking)

    Current verified main: 42841dc9b451d84a3ee0332008b3af0e21ff51d2 (new #760 ImmerSIB/SonKu installer/FAT work merged while the two threads proceeded). It does NOT include the new generic dual-ASIO/multiple independent capture-device feature. No new current implementation PR or newly dated Multi-Device capture branch was discovered in the GitHub PR/branch search at this checkpoint. This cannot rule out unpushed local work; do not confuse prior docs #763 (merged) or older input-only research #757 (draft) with shipped multi-device runtime.

    Thread A: MVVM native GUI — CORRECT DIRECTION, NOT YET FULLY QUALIFIED

    • PR #765, checked HEAD c0a56a508e0279b5dabb07340bbad93f215260a7: actual shipped native /preview/ Device I/O presentation VM and linked setupBegin/ACK/readback/timeout path; native coordinator reused, no DSP renderer duplication. Exact-head Full Chromium #44189797846 SUCCESS, fast UI/native SUCCESS; PR is NOT merged yet and current required merge gate was PENDING on audit.
    • The pure ViewModel is a genuine test boundary, but the JS bridge only lost ~55 net lines and the main 8,873-line src/Main.cpp native device application logic is not modularized by this PR. Do not claim full MVVM or source maintainability solved yet.
    • Demonstrated correctness gap: setupReadback compares routing device.epoch to JUCE deviceGeneration (driver start counter). At baseline routing epoch 40, old start generation 3, new ACCEPTED ACK generation 4, an unchanged old schema-v6 snapshot with epoch 40 and identical endpoint/format/pins returns phase=ACCEPTED even without proof of a newly installed generation. This was reproduced by executing the exact HEAD pure VM in isolation, not an actual vendor-ASIO field bug. Existing tests use near-equal epoch and generation. A detailed nonblocking audit with regression request is on PR refactor(mvvm): shipped native Device Setup VM, Apply ACK and readback lifecycle #765 comment 6963008417. Verify accepted-readback provenance under divergent counters and replay before merge; don't hide by weakening tests.

    Thread B: Multi-Device native I/O — NO NEW IMPLEMENTATION PROOF YET

    • The existing code already contains engine/device/AdaptiveClockBridge.h (preallocated SPSC/FIR drift correction), ClockBridgeSizing.h, JuceIndependentClockDevice.h (two independently clocked WASAPI JUCE managers: capture-only producer, render owner), LiveSecondaryOutputDevice.h (secondary WASAPI MAIN-copy consumer), and deterministic existing tests adaptive_clock_bridge_test.cpp, adaptive_clock_virtual_soak_test.cpp, independent_clock_adapter_contract_test.mjs.
    • These are NOT a finished generic topology of multiple independent ASIO captures, persisted DeviceKey/pin→ChannelId source binding, and independent BUS output sinks. They should be audited and extended/reused, not reimplemented naively and not mistaken for multi-ASIO certification.
    • The next meaningful code/proof milestone is IO-0/IO-1: exact JUCE ASIO vendor-A + vendor-B secondary input-only simultaneous-open feasibility (including fail-closed driver collision, primary direct callback unaffected) and a bounded capture-only producer feeding the one existing native graph through independently-clocked bridge. Start with deterministic mismatched-rate, overflow/underrun, reconnect/failure-isolation tests; physical vendor FAT PENDING until an actual Windows interface pair is observed.
    • Keep one accepted mixer DSP/VST3 renderer and primary ASIO zero-extra-buffer fastpath. Additional capture callback must NEVER run FX/VST3 or replace/arm a new playback device. Source mapping requires DeviceKey+pin, not two unrelated drivers' IN 1 ordinals. No custom signed driver, duplicate JS accepted state, cosmetic refactor, or new owner/coordination gate.

    Coordination: Both threads may continue on separate branches without file locks. A real shared contract change should be tested at integration; no permission token. One Indonesian completion report per milestone with actual PR SHA, tests, pending FAT and owner action. The immediate operator action is TIDAK PERLU MELAKUKAN APA-APA, with no new multi-device installer to test.

  17. masarray commented on Nov 18, 2026

    @masarray
    OwnerAuthor

    2026-11-18 MVVM maintainability checkpoint — source proven

    GUI-1 Channel Strip, GUI-2 FX Rack, GUI-3 WIRE, MV-2 Device Setup and MV-3 native read-model are already merged. Additional genuine ownership extractions merged today: #779 MV-4 WIRE native gesture→transaction (commit 4c2299208e3f8d531792054aeb50e19c8bebfb33) and #780 GUI-4 Channel Strip PRIMARY OUT/SOURCE ONLY intent (commit 1bd5d49fbd6b5a5f27682226b5b4ec5ca8fe9a11). Existing native dispatcher/ACK/readback and one PCM graph preserved. Exact-head fast UI/native, full Chromium, required merge gate passed for both PRs. Main post-merge CI is independently evaluated; no Windows 2-vendor ASIO FAT or new installer was created by these refactors.

    Remaining evidence blocker (not disguised as green): Main-push UI screenshot parity test Stage C failed on main@3307c4bb run 44236089782 and main@4c229920 run 44239675104, although exact-head full Chromium PR qualifying runs succeeded. Both fail at 1366x768 pixel screenshot hash comparison, not a proven deliberate GUI change. Diagnose deterministic image decode, font readiness, animation/frame timing with captured diff evidence; do not disable the assert or update visual baseline blindly. Next maintainability acceptance should demonstrate an isolated source/OUT feature edit through existing VMs with stale readback/epoch and SOURCE ONLY vs parallel SEND regression.

  18. masarray commented on Nov 18, 2026

    @masarray
    OwnerAuthor

    2026-11-18 MV-5 milestone complete — external output source ownership

    PR #781 merged at main@7f3f4d8f16ccc9a1b4aa53c4f3185f7667f2b3e8. Existing pure Channel Strip ViewModel now prepares EXTERNAL OUT per-ChannelId ASIO physical pair and shared-output confirmation; the one native bridge still owns transport, ACK/timeout and installed PCM readback. Validation rejects stale epochs, invalid/double labels and MAIN/MONITOR pins without inventing another routing graph. Two new negative/positive deterministic unit contracts; bridge removed its inline preflight. Stage C CSS original-vs-compiled full screenshot gate now requires stable decoded assets and exact identical PNGs (no pixel tolerance). PR fast UI/native PASS, exact-source 109-step full Chromium PASS, required merge gate PASS. Postmerge main UI/native CI still running at time of note. Hardware ASIO FAT and new installer NOT RUN; source milestone only. No change to C++, JUCE, DSP, plugin leases, /preview/ HTML/CSS/PNG or .smix. Historical main push Stage C failures were documented in issue thread; a PR pass cannot itself prove main-push flake eliminated.

  19. masarray commented on Nov 18, 2026

    @masarray
    OwnerAuthor

    2026-11-18 MV-6 merged — SOURCE Queue epoch/ACK ownership

    PR #782 MERGED MAIN at 1e5df9a3cc91c31b4fd457fb9ce360725afcf2d6 (PR HEAD 061a7f39389a91ac5c27cdda622fe1b823421512). One existing pure Channel Strip ViewModel now owns queued SOURCE input choice, device-epoch pin admission and native-vs-M4 ACK correlations, removing old duplicated bridge policy. Real bug prevented: a queued old-epoch input selection silently reused with a new device/revision. Preserved native uint64 string epochs >2^53, M4 safe numeric counters, first-sound G0 without an extra Monitor step, one native source dispatcher and installed readback. Source-integrity tests+Fast UI/native passed; Full Chromium 109/109 + Stage C passed 44247970033; Required Merge Gate passed. Main-push UI/native were still running at note time. No code change to C++/DSP/ASIO/device/GUI visual, no installer, physical ASIO FAT not performed. Next bounded milestone should prove an isolated pending readback transition on one admitted source without a second UI or graph store.

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

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions