Skip to content

P0 dynamic mixer — qualify add/delete/reorder, native FX/routing and Show persistence #192

Open

Description

@masarray

Milestone B verified triage — 2026-11-17: P0 / mixer / defect / field-fat. Classification ≠ completion of product feature or physical FAT.

Scope / release boundary: Dynamic track reliability with stable ChannelId; do not return to fixed ten-strip code.

Exit criteria:

  • Test add/delete/reorder/rename arbitrary channels while keeping native FX/WIRE, bus, fader and device-independent session identities.
  • Verify exact PCM/ACK/readback, .smix reopen and physical device restart; channel5 must work without channel1, no stale route or callback lifetime leak.

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


CURRENT CHECKPOINT — 2026-11-08 (main c57720ea25585c861af89802e1365ffef7d03d17)

Current main is beyond the original 10-strip baseline. Merged #418's stable ChannelId graph plus #445 operator Add Audio Track are present; do not copy old fixed-M4 branches. Remaining scope: prove dynamic add/remove/reorder/rename of channels across native PCM, VST3 lifetime, bus/WIRE routing, Show persistence and device restart for arbitrary reasonable counts. Address user-reported inability to add a track as a regression with exact-head tests. This issue remains the active dynamic-channel owner, not an invitation to fork another PCM engine.


Dynamic Channel Architecture — one ChannelStrip class for every channel, scalable logical mixer

Owner requirement (2026-10-30): all channels must have identical native FX, source/format, send, bus, meters, fader, mute/solo, edit and Show capabilities; CH1 is not special. User adds/removes/reorders channels at runtime without manually coding another CH number. NOT a fake UI that creates a strip but renders no audio.

Current source-grounded blockers

Target: one shared implementation, runtime-sized instances

  • Stable 64-bit ChannelId (not strip index), unified ChannelStrip state and processing chain reused for channel 1..N; physical sources are independent of logical channel count. Explicit role INPUT, FX_RETURN, GROUP, MONITOR, MAIN; role controls routing capability, not a separate FX engine.
  • Append/remove/duplicate/reorder channel session transaction off RT. Preserve stable ID and plugin vendor state on reorder; remove/duplicate must not copy/release a live VST on realtime callback. Clones get new unique IDs and separately instantiated plugins only when operator explicitly chooses plugin-copy.
  • One shared VST3 rack and editor/ACK across every channel, abolish processLegacyChannelOne via deliberate migration with regression proving CH1 Orbit/editor/vendor state retained. No fake 8 VST instances per idle channel.
  • channelCount is session data; per-session capacity based on measured CPU/latency/deadline and memory preflight, not product cap 10/12 or physical interface port count. Default maybe 10 for backward compatibility, but add/remove up to resources. No promise literal infinite channels/latency.
  • Control-thread graph compiler creates a compact immutable render schedule and preallocated buffers for audible/active channels; audio callback loops only compiled nodes and never allocates/locks/creates VSTs/scans device. Reuse existing RealtimeRoutePublisher and epoch/revision ACK; structural changes require output MUTE and explicit APPLY. Negative ACK leaves audible graph unchanged.
  • Direct/mono/stereo source, bus DAG, send tap, group PDC, physical output and Show V5 migration must use IDs not array indices. Version old 10-channel Show safely. Graph must be one authoritative path.
  • GUI virtualizes/paginates lanes + bank navigation; mounting 200 DOM faders and updating every meter per frame is prohibited. L2 geometry of visible strips unchanged for CH1..N+MASTER. Top MIXER/FX RACK/PATCH unchanged.
  • No arbitrary hardcoded channel number in native/JS route validation; routing capacity is resource-driven and accepted by native API.

Completion gates (meaningful vertical slices)

DYN0: robust control-only runtime channel roster/ID/budget/rollback with deterministic tests (NOT live channels yet), architectural migration seams, reusable contract.
DYN1: dynamic native prepared render graph with 2/10/32/64/128 logical channels, no allocation in callback, identical actual PCM and FX interface for CH1/CH11/CH64, no CH1 legacy switch; preserve full accepted route and PDC isolation.
DYN2: user + Channel / remove / duplicate with native MUTE→APPLY→ACK, visible GUI bank virtualization, mono/stereo source/format, FX, sends, meters and persistent Show roundtrip. Negative ACK and device stale tests.
DYN3: hardware + CPU/deadline/memory soak 2×2 and 8×8 ASIO/WASAPI 0FX/multiFX; prove audio continuity and low spec performance. #15 remains independent release blocker.

Dependencies: product epic #170, G5B #186, #174 MS2, #175 MS3, #95 FX/PDC, #85 Show, #176 GUI lock, #15 hardware 0FX. No premature 'all channels live' claim from DYN0.

Activity

  1. masarray commented on Oct 30, 2026

    @masarray
    OwnerAuthor

    DYN0 merged — 2026-10-30. PR #193 is in main at 8e53f1d0b06cf8aa7da2f2cbd15f7f087bf27ef8 after exact-head Windows CI 41044867738 PASS. The code review bug ignoring STEREO on add is repaired, tested (stereo at creation + bad-format rollback) and the review thread resolved. The shipped app still has the existing fixed native 10-input/4-AUX render graph: do NOT claim +CHANNEL LIVE. Required next vertical slice remains DYN1 real PCM for CH11/CH16/CH32 and one shared FX lifecycle, followed by DYN2 banked UI + MUTE→APPLY→ACK and Show migration. G5C #190 also merged at 593427d3 with verified Windows artifact run 41044034992. Preserve #174 typed sends and #15 0FX hardware FAT.

  2. masarray commented on Oct 30, 2026

    @masarray
    OwnerAuthor

    DYN1A implementation created (2026-10-30): PR #199, latest head ddf1567e14693eca61c4502f09c141f07f5901c5, exact-head Windows CI 41051705731 currently RUNNING (not merged). New engine/core/DynamicPreparedMixerBlock.h: offline replacement-ready single-kernel C++20 native PCM graph, ID-based channel/input patches and mono/stereo gain/PRE+POST sends/FX callback/MAIN, Kahn O(V+E) DAG compilation, channel reordering and fail-closed graph rollback, duplicate-send rejection. 128 logical channels with real PCM tests for CH1, CH11, CH32, CH64, CH127 and MAIN, 1024 disconnected logical lanes consuming only MAIN scratch, plus MUTE/SOLO/PRE and budget tests. Dynamic roster gains explicit MUTE/SOLO/PRE editor semantics. This is NOT LIVE CH11+: the current JUCE LiveDryRoutingLab still uses stripCount=10, and CH1 legacy FX / 4 AUX / Show/device epoch publisher are not migrated. DYN1B must replace the single live graph and connect VST3 lifecycle (no parallel audio callback); DYN2 must then expose native-ACKed Add/Remove/Duplicate with banked GUI and Show. Preserve L2 geometry and #15 hardware FAT. Do not call DYN1A production-ready until Windows MSVC/CTest/package passes.

  3. masarray commented on Oct 30, 2026

    @masarray
    OwnerAuthor

    2026-10-30 DYN1B handoff after substantial MS3 / FX integration

    Latest verified merged baseline: main@0801cc8 includes DYN0 #193, DYN1A #199 (offline dynamic PCM kernel, 128 active lanes / 1024 descriptors), G5B–G5E vendor VST3 #186/#190/#194/#197, MS2A #189, MS3A #198, MS3A2 #201 and #202 FAT docs. Independent MAIN+MONITOR true physical render bus MS3B PR #203 is still draft, exact-head Windows native CI outstanding; do not assume merged or hardware qualified.

    Non-naïve DYN1B next kernel replacement seam: DynamicPreparedMixerBlock must replace PreparedMixerBlock within the same existing LiveDryRoutingLab published graph, not run in parallel; preserve physical output-pair selection and optional separate monitor destination, all 14 input/AUX FX identities/state/instance leased lifetimes, mono/stereo conversions, PRE/POST send DAG, CH1 legacy Orbit/vendor path, native MUTE/epoch/revision ACK, real PCM meters, live fader interpolation and portable Show. Reuse full G5E native FX state capture/restore; don't recreate vendor processors on fader, channel selection, GUI bank scroll or graph reorder. A standard-channel ChannelId must produce the same engine path and feature set for CH1, CH11, CH32, CH64 etc.; no CH1 production special case in the final stage. Budgeted lazy scratch must remain; 1024 mostly-disconnected roster entries must not allocate 1024 fixed channel buffers. No unbounded per-callback DAG traversal, locks, JSON, pointer refcounting or UI timers.

    Stop/go gates: (a) prove identical CH1–10 PCM+FX & 4 AUX behavior and 2- and 8-output safe patch against old live renderer; (b) prove CH11+ PCM uses that same native callback with correct stable identity, physical patch and mute/ACK; (c) unify native VST3 instances and editors under per-ChannelId ownership, then DYN2 real +CHANNEL and banked GUI; (d) #15 real ASIO/WASAPI crackle FAT remains mandatory. Only run one full exact-head CI per completed vertical slice, not docs-only Windows builds.

  4. masarray commented on Oct 30, 2026

    @masarray
    OwnerAuthor

    DYN1B execution dependency lock — 2026-10-31

    Do not commence live JUCE dynamic graph migration on stale main. Canonical audited #170 status: main@9746e3ee, Windows Native/CTest + Chromium + Pages succeeded. DYN0 #193 and DYN1A #199 are merged but only control roster/offline PCM, while shipped LiveDryRoutingLab::Block = PreparedMixerBlock still has fixed 10 input strips + 4 AUX and a special CH1 vendor/Orbit seam. NO actual runtime Add Channel audio yet.

    Current shared-source MS3B PR #203 exact-head Windows + Chromium SUCCESS but is not merged and is 1 main commit behind L2 GUI #208. Finish its semantic GUI rebase and exact-head CI first; do not touch shared renderer/FX/IO files concurrently.

    DYN1B single milestone contract after MS3B: one immutable graph and existing JUCE callback, stable ChannelId on CH1–N, actual real PCM CH11/32/64, full native VST3 FX ownership/state + CH1 Orbit compatibility, 4 AUX and new MONITOR outputs, sparse physical pins, PRE/POST, mono/stereo, live ramp/peaks, MUTE→APPLY→epoch/revision ACK+rollback, bounded lazy scratch and zero RT alloc/lock; parity regression at 10CH and live 11+CH required. Only DYN2, after native acceptance, may add UI +CHANNEL/banked virtualized strips and Show V5. Real device 0FX crackle FAT #15 is independent.

  5. masarray commented on Oct 30, 2026

    @masarray
    OwnerAuthor

    DYN1B-1 source milestone — PR #212, commit 55380c82: extended the existing runtime-sized ChannelId PCM graph from MAIN-only to one-pass MAIN + independent MONITOR. Control-thread compiler now validates singleton MONITOR, refuses monitor feedback, builds reachability from both sinks (monitor-only channels remain audible), allocates scratch only to scheduled lanes, and retains transactional compile rollback. The existing ARMixMixerRoutingTests now runs 64 dynamic input lanes + MAIN/MONITOR with distinct PCM, CH32 MON-only, CH64 MAIN-only, stable reorder IDs, output-buffer alias rejection and duplicate-monitor / feedback rollback. Exact-head Windows CI: https://github.com/masarray/arsonmix/actions/runs/41134955586 (pending at checkpoint). This does not yet authorize GUI +CHANNEL: DYN1B-2 must migrate the ONE current live JUCE graph, full 10CH+4 AUX, PRE/POST, real VST3 host/editor/vendor Show, CH1 Orbit state, native epoch/ACK, and monitor/output semantics before DYN2. Avoid duplicate live callbacks, fixed 10-channel UI masquerading as dynamic, or per-channel DSP copies. Reference: #212.

  6. masarray commented on Oct 30, 2026

    @masarray
    OwnerAuthor

    2026-10-31 DYN reconciliation (source-verified): PR #215 was merged to main as ff173964; the real 10CH+4 AUX legacy-to-ChannelId migration has offline 64-frame PCM parity and fail-closed validation. It is NOT live DYN +CHANNEL, and the single native JUCE/VST3 callback remains authoritative. New convergence ledger merged via #217 at 71bcd715 to preserve progress without overwriting RAIL3/OUTPUT1/Show. Next executable milestone: DYN1B-2B one-graph atomic live cutover with existing plugin leases, CH1 Orbit/editor, AUX, MAIN+MONITOR, output secondary routing, smoothing and device epoch/revision MUTE→APPLY→ACK. No second callback or false low-latency claim. Issue #15 physical 0FX crackle remains independent P0 release gate.

  7. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    2026-10-31 CURRENT-MAIN AUDIT — CH1 SPECIAL CASE IS A PRODUCT-BLOCKING ARCHITECTURAL DEFECT

    Audited exact main@d3293af72eef029032a9138f26e6c71dc69967f7 (merged M1 #231, Windows/Chromium CI green). Source and UI still violate this issue's original requirement that CH1..N and audio bus/FX returns share one native strip + FX engine. NO claim of code fix or physical FAT.

    Verified source defects

    1. Two FX presenter/command paths. web/bridge.js:202–305 bifurcates FX RACK by selected!==0 / state.chain: CH1 legacy bespoke Live FX Chain vs routed CH2..10. web/bridge.js:691–785 similarly forks every strip's six visible FX rows; MASTER renders disabled FX slots (:696–714). web/bridge.js:185 hardcodes NATIVE CH1 · ORBIT. Not just CSS.
    2. Two FX instance ownership paths inside ONE callback. engine/core/LiveChannelFxGraph.h:264–287 routes strip==0 to externalFx / legacy CH1 lease while other lanes use processors_[strip]. src/LiveAudioEngine.h:1659–1680 constructs LegacyChannelOneLease; LiveDryRoutingLab.h:163–171 forbids simultaneous legacy and new CH1 inserts. No parallel output callback, but special lifecycle/API and feature parity remain.
    3. Distinct VST commands/state src/Main.cpp:1271,1407,1429 routed FX vs :2058,2118,2139,2159 legacy ADD/REMOVE/MOVE/BYPASS; LiveAudioEngine.h:698–753 addVst3 and :756–809 CH1-specific state/preset. web/monitor.js:1008–1017 refuses routed FX for CH1 if legacy rack exists. The UI cannot truthfully promise identical operations until the native paths are reconciled.
    4. Runtime adds/bus capability not shipped. web/monitor.js:278,398,588,733,822,912,1012 remains hardcoded 10 inputs, plus 4 specific AUX returns. engine/core/DynamicChannelRoster.h:13–45 already models stable 64-bit ChannelId, roles INPUT/FX_RETURN/GROUP/MONITOR/MAIN, uniform eight-insert definitions, but explicitly says it is a CONTROL-THREAD session model, not live. DynamicPreparedMixerBlock.h:88–210 has a generic topological bus/send PCM compiler and cycle prevention, not yet wired as the sole native processing graph. LegacyLiveGraphMigration.h:13–17 is offline only.
    5. Master and user-defined audio bus not FX peers. web/bridge.js:696–714,250 explicitly says MASTER FX unavailable. Current MAIN/MONITOR and fixed AUX are supported in routing, but user cannot freely create a vocal/drum/FX bus as a live native channel with identical multi-insert and Show lifecycle.
    6. Safety and latency remain gaps. LiveAudioEngine.h:735–738 rejects latency-bearing CH1 plugins with sends because PDC is absent. P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15 zero-VST physical crackle not qualified. No fake 0ms end-to-end or fake DAW parity.

    Architectural decision — ONE unified ChannelStrip (source ≠ role ≠ output)

    Adopt one stable-ID runtime node for all audio-processing strips (input, receiver, subgroup, FX return, main); shared UI component and native FxRack interface (same actual slots, insert/bypass/reorder/editor/vendor-state, PRE/POST, fader, pan, mute/solo, meters) for any audio lane. Route role is configuration/default wiring/capability, not a separate FX engine or a duplicated CH1/CHN presenter. Main is a special final output destination, monitor can expose role-specific controls; DCA/VCA is control-only and should not pretend to own an audio processor.

    • Each channel can have source=None, physical mono/stereo, or incoming sends/receives, and independently chosen output edges to other channels/buses, MAIN, MONITOR or no output (intentionally silent). No mandatory to-MAIN. Named bus templates Drum Bus, Vocal Bus, FX Reverb, Submix initialize roles/routing/presets but share one DSP processing class.
    • Support replacement/main assignment and parallel PRE/POST sends as distinct edge types with stable ChannelIds. Validate bus summing, level and pan, no duplicate sends, no self feedback or multi-hop cycles. None/no outputs is valid intentional silence; show clear disconnected status. Expose only real physically enumerated output pins, no second callback.
    • One immutable resource-budgeted audio graph built off callback; channel IDs stable across order changes; preallocated scratch, topological schedule, bounded SPSC telemetry. Prepared VST instance leases are per ChannelId/insertionId; control-thread creation and destruction, callback immutable processing leases/epoch ACK, safe fade on publish, rollback old graph on failure. Do not move shared_ptr refcounts or heap allocation into RT.
    • Migration is nonoptional: import CH1 existing 8-insert state/UID/order/automation/vendor chunks/Orbit into unified per-strip rack exactly once, preserve history/Show and no double processing; CH2+ uses same entrypoints afterward. Delete/retire legacy CH1 presenter and externalFx fallback only after A/B PCM tests on same VST3 and Show. Native MASTER FX then becomes a real processing node, with PDC/latency contract and live-safe gate.

    ONE meaningful integrated milestone with guarded phases (not cosmetic PRs)

    A. Unified live DSP+FX runtime cutover: on same native callback make all existing CH1..10, AUX1..4 and MAIN use one per-node FX lifecycle and one UI FX Rack without CH1 special branch. Keep old Shows valid, mono/stereo buses, independent MAIN/MONITOR/OUTPUT1. Tests: CH1 vs CH8 same source+chain = matching PCM, same automation/editor/bypass/move/state, no duplicated processing, 0FX and latency test; no callback file/plugin operations.
    B. Runtime channel & bus creation: ADD CHANNEL / ADD BUS / ADD FX BUS produces actual native ACK-installed signal nodes using same class, user may name/route CH12, Drum/Vocal/FX BUS. No UX-only ghost channels. Sends to arbitrary channels/MAIN/None, PRE/POST and topological cycle rejection, valid mono/stereo adaptation, ASIO/WASAPI real physical device independent of logical channel number. Resource budget used instead of 10CH hard cap.
    C. Full live qualification: arbitrary chain state/Show recall + rollback, route switch clicks/fades, plugin-induced delay compensation/explicit rejection in zero-latency live mode, device hotplug/buffer changes and #15 hardware 0FX FAT; 32/64/128/256 frames, 44.1/48/96 kHz, normal iGPU GUI, 30-minute zero-FX wired soak; measured CPU and xrun.

    Dependencies/ownership: product epic #170, all-channel VST/PDC #95, Show #180, hardware #15. #232/#233 currently modify DEVICE/SOURCE and must not be overwritten. Never merge 18 old graph/UI branches to 'solve' this. Avoid new standalone audio engine / CH1 special migration exceptions beyond the temporary adapter.

    Acceptance gate: a NEW channel (e.g. CH11) and CH1, both configured from same inputs and same FX, show identical controls and audible result; create Drum Bus and FX Bus and send CH1+CH11 to them, route one Bus→MAIN and another→None, save/reopen Show with exact VST vendor state and no unexpected output. No 'finished' claim until this passes on Windows native and actual hardware.

  8. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    2026-10-31 M2 implementation underway — PR #235. Referenced official Fender Studio Pro/Studio One Console and Effects Signal Routing. Actual code: one shared FX Rack presenter (fxLaneSnapshot+fxSlotMarkup) paints 8-slot workspace/6 compact rows for all existing CH1–10; two truthful native command adapters preserve legacy CH1 vendor Orbit/Show vs routed CH/AUX VST leases. + BUS creates a real subgroup by borrowing a genuinely unused native strip, source MAIN off + unity POST to BUS → MAIN; + FX BUS creates real parallel .25 send keeping source MAIN. Both use same native DAG/preflight/MUTE→APPLY→revisioned ACK; rejects full 10CH capacity/dirty drafts/armed output. Critical fix preserves stage BUS fader unity through syncNativeLevelsToRouteDraft, including structural APPLY. Not complete #192: no CH11+/unlimited buses, no single per-ChannelId VST owner yet, MASTER VST/PDC absent, no hardware #15 FAT. Do not falsely mark all-channel parity done. PR CI still being validated; must merge only exact-head green and preserve #232 DEVICE owner.

  9. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    M2B MILESTONE MERGED 2026-10-31 — PR #235 → main@54e4229bd3e0d1038aad39b5839116bedaf18515. Reconciled after successful M1 #236. Exact head 1097fc8eb4d78f842726230ff26e4ed4302295bf Windows MSVC/native RT tests/ZIP verification #41195235215 PASS and full Chromium including multi-source BUS after native ACK #41195235235 PASS. Live native 10CH+4AUX supports Studio-style +BUS +FX BUS, multiple CH→one accepted BUS, exclusive group vs parallel POST, legal BUS nesting through DAG preflight and fail-closed MUTE/APPLY/native ACK. Same FX Rack presenter for CH1–10; legacy CH1 vendor/Orbit lease remains distinct under adapter. Latest main Windows distribution run #41195859441 building physical FAT candidate. #192 still requires stable dynamic ChannelId owner for CH11+, true flexible named BUS/FX channel roles, unified all-lane VST3 lease, native MASTER FX/PDC and 0FX crackle #15 hardware evidence. Do not reimplement M2B or merge stale legacy P1C/P1D branches over current main.

  10. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    2026-10-31 P0 PRODUCT RESET — Studio One Console workflow parity, ArSonMix live architecture

    Audited exact main: 91043355b8762112f10e980a6997c5306bd3a387; Windows CI #41198065189 SUCCESS and Chromium #41198065205 SUCCESS. Inspected actual Chromium preview screenshots in GUI artifact #13041431976: mixer-shell-fx-1680.png, mixer-shell-route-1680.png, mixer-shell-sends-1680.png, mixer-shell-compact-1366.png, native-input-patch-review-1680.png. A screenshot rendered without JUCE is a GUI preview, not an audio/hardware FAT. Native x64 build success is not audible 0FX evidence (#15).

    A. User decision / non-negotiable product target

    ArSonMix is a real-time digital audio MIXER, not a timeline editor, recording/arranger clone, OBS-specific output target or zero-millisecond hardware claim. Product functional workflow must closely match Studio One / current Fender Studio Pro Console while retaining existing ArSonMix premium gunmetal visual language, compact 11-fader geometry, physical PNG faders and responsive low-spec WebView2. Do not copy proprietary assets, branding or code.

    Official console references:

    B. Source-verified current defects / why M2B is NOT Studio One parity

    1. Fixed visual 10CH+MASTER with demo metadata: web/app.js:3–17 seeds precisely ten raw channel descriptors. web/app.js:225–229,449 still renders Groups as six static name rows; clicking data-group literally reports Group selection — interface placeholder. web/app.js:398–415 manages only ONE currentChannel and rejects channels beyond channels.length. No multi-select transaction or bulk output.
    2. Audio graph is fixed-width: web/monitor.js:279–304 constructs 10 draft CH and 4 AUX, 14 monitor send taps. web/mixer-io.js:111–112,684–695 enforces 0..9/for 10 in channel selector. engine/core/ChannelRoutingContract.h:62–68, ChannelSendPlan.h:24–67 define std::array<...,stripCount> and max 32 sends. src/LiveAudioEngine.h:1682–1698 runs actual dryRoutingLab_.process on the sole callback. Native BUS creation in web/monitor.js:1060–1110 consumes an UNUSED 10CH lane; this CANNOT scale.
    3. Only FX appearance partially unified: web/bridge.js:203–266,698–740 one FX strip renderer; but engine/core/LiveChannelFxGraph.h:264–287 still uses special strip==0 && externalFx for legacy VST3, src/LiveAudioEngine.h:1693 still constructs LegacyChannelOneLease. CH1 vs CHN host lifecycle, vendor GUI and Show cannot be declared the same DSP class. Native MASTER inserts/PDC not complete (ARD: All-channel native FX Rack + safe drag reorder + yellow-green toolbar + recessed fader #95).
    4. Console routing discoverability is weak: GUI screenshot uses 10 tightly packed colored columns; FX/ROUTE/SENDS GLOBAL bank swaps entire upper row. web/mixer-io.js:89–147 shows a good selected CH SOURCE→FX→SEND→OUT workbench but label OUT may reduce arbitrary group sends to SEND ONLY. web/mixer-io.js:601–630 strip send chips enumerate just four AUX targets, not every accepted internal BUS and output edge. In FX Rack, web/bridge.js:275–300 places + BUS/+ FX BUS in a separate workspace, and attaching a second source requires another route draft; no fast multiselect action or always-visible primary output selector on each strip. Operators cannot reliably inspect group membership, BUS fan-in, destinations or direct-vs-parallel routing at a glance.
    5. Dynamic building blocks exist but are NOT live: engine/core/DynamicChannelRoster.h:13–94 explicitly states control-thread-only, stable uint64 ChannelId, identical roles input/fxReturn/group/monitor/main, logical add/duplicate/remove/reorder. engine/core/DynamicPreparedMixerBlock.h:14–85 has preallocated compiled DAG with budget/cycle safety, not wired to JUCE callback. engine/core/LegacyLiveGraphMigration.h + docs/DYN1B_LEGACY_PARITY_BRIDGE_20261031.md provide offline PCM parity against current 10CH/4AUX, not live activation. Use these foundations, do NOT write a parallel renderer/second graph.
    6. Live physical qualification missing: P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15 reports crackle at 0VST and needs fresh hardware evidence. Windows CI != stable ASIO/WASAPI sound. The successful source screenshot is a browser preview without native audio.

    C. Single canonical design: unified SessionAudioGraph and Console UI

    • Definitions: immutable 64-bit ChannelId independent of GUI order and physical port; one AudioChannel DSP/FX/meter/fader/pan/mute/solo/sends ownership API for INPUT, GROUP BUS, FX RETURN, AUX, MONITOR and MAIN where role-appropriate. A VCA/DCA is a control group, not a fake audio summing bus.
    • Separate SOURCE, ROLE, PRIMARY OUTPUT, SENDS: physical input/verified software source/none is selected independently of audio processing format. Primary OutputDestination can be BUS, MAIN, MONITOR/sub-out (when physical config permits) or NONE; PRE/POST sends are independent parallel edges, each with stable ID, gain, pan, mute. Some buses can cascade, but DAG rejects cycles, duplicate output paths, accidental doubled dry sound. NONE is legitimate silent output even when channel has an input; if sends remain live, show SEND-ONLY explicitly.
    • Bus templates rather than separate DSP engines: Add Audio Channel, Add Bus, Add Bus for Selected Channels, Add FX Channel, Duplicate, Delete. Group creates a new independent ChannelId WITHOUT consuming a 10CH input/hardware pin and atomically assigns selected sources' primary outputs in the SAME user operation (one guarded control-thread batch). FX Channel can be created empty, its default 100% wet plugin setting is a sensible recommendation but not enforced as a fixed engine property. 0 input channels may still receive BUS sends. Default new BUS→MAIN; user may change.
    • Live-safe graph: derive one immutable prepared topological schedule with resource admission, scratch preallocation, dynamic stereo/mono mapping, clock-specific physical routing and ASIO sparse-port support; no heap/locks/plugin creation/device scan/memory reclamation in callback. ChannelId-keyed plugin instance leases prepared off RT, accept/rollback with epoch/revision/ACK and bounded fade at graph publish. Realtime fader/pan/send gain/mute remain independent atomic/smoothed controls; structural changes are async-control transactions and must never silently unmute. Preserve vendor plugin state/Show/Orbit/sidechain/latency and one AudioDeviceManager/callback. No claims of 0ms device latency or universal zero plugin delay.
    • UI read model: one native-ACKed session snapshot supplies actual channel descriptors, role, primary output, sends, fan-in list, and metering to all views; a separate DRAFT overlay shows edits. No fabricated browser-only channels, no duplicated channels[] 10CH vs native ChannelId truth, no full strip reconstruction for meter events. Reorder/hide/channel bank preserve stable DOM identity and meter subscriptions.
    • Two-tier routing UX: (a) permanent compact per-strip IN, OUT: Vocal Bus / MAIN / NONE, chips for active SENDs, BUS has IN: 3 channels popover; grouped sources highlighted when BUS selected, BUS highlighted when channel selected; shows ACK vs pending. (b) optional Signal Flow selected-context graph of primary solid vs parallel dashed paths, device pin sink, cycle refusal, source list; never draw canvas wires permanently across all faders. Fast console right-click/multiselect -> Add Bus for Selected Channels (one operator gesture); source/output/send selector each ≤2 actions when safe.
    • Visual: preserve compact premium brushed gunmetal ArSonMix, established L2 GUI BASELINE LOCK — aligned CH1–10/MASTER faders, left-rail layers, three-tab workspace #176 L2 equal-height faders/meter, existing BTN-180 and color IDs, pinned MASTER far right, light low-cost outline/border highlight, no blur/glow/per-pointer large DOM rebuilds. At 1366px show 8–10 strips with horizontal banking/virtualized lanes as needed; do not cram 30 channels in a single static row. BUS/Fx channel glyph + label and source-count indicator, natural Show Sources / Hide Sources, not oversized cards, extra toolbars or a new app shell.

    D. Three TRUE acceptance milestones (no trivial split PRs)

    M3 — LIVE Dynamic Channel/BUS Graph cutover (P0, big source milestone): integrate DYN0/DYN1B compiler into sole active callback and native Show, migrate current 10CH+4AUX losslessly including CH1 actual vendor FX; create CH11 + independent DRUM/VOCAL BUS/FX RETURN in same device. Route 2 physical Mics→new Vocal Bus→MAIN with real PCM, fader/mono/stereo and FX; no borrowed input strip, no second callback, old Show import/rollback and no 0FX regression. Stop if plugin state/Orbit cannot migrate safely. Use fixed audio resource budget (measured) rather than arbitrary 128-audible promise.
    M4 — Studio Console operational parity (P0, single integrated UI milestone): dynamic ChannelId strips and multi-select, right-click Add Bus for Selected Channels, Add empty Bus/FX Channel, rename/reorder/remove/duplicate, one shared FX Rack with real editor, channel routing selector, bus fan-in popup, selected-source highlight, source hide/show, fast Send assignment, topological nested buses, ACK/DRAFT distinction; 1366/1440/1680 geometry stable, MAIN pinned right and low-spec pointer responsive. Accepted graph is source of truth; no fake Groups.
    M5 — Full audio parity + qualification: native MASTER/VST3 + PDC policy (including low-latency live modes), scene/Show portable round trip for dynamically created BUS and vendor chunks, safe hot switching/guarded rollback, detailed PCM/golden graph tests + CPU/xrun budget, ASIO/WASAPI physical #15 0FX baseline at 44.1/48/96k 64/128/256 frames, 30-min 0FX and FX soak, device loss/hotplug, no dropped/duplicated samples or surprise output. Post-merges must have green exact-SHA Windows native CTest/verified ZIP + Chromium, then genuine human physical FAT.

    E. Definition of DONE in plain user terms

    On one verified MAONO/ASIO or WASAPI device, create or import 8 physical audio channels, multi-select 3 vocal channels and choose Add Bus for Selected Channels; a brand-new Vocal Bus appears (8 original inputs remain; now 9+ logical audio strips), receives exactly those 3 primary outputs, has the SAME VST3 rack and controls as CH1, and routes to MAIN by default. Create an FX Reverb return and send one Vocal channel in parallel while preserving dry-to-Vocal-Bus. Click Vocal Bus: all 3 sources immediately visibly highlighted + available in input popover. Route BUS→Drum Bus / MAIN / NONE via a compact direct output selector, rejects cycles and never doubles the dry path. Save Show, close/reopen, verify complete native graph and plugin vendor state; adjust gain while live with no structural restart. No UI-only behavior passes acceptance.

    Merge/release discipline: Existing #235 M2B is MERGED, do not regress it. #192 owns audio/UI cutover; #95 VST3 engine, #170 product, #174 routing and #15 hardware qualify. Quarantine stale PR #117/#115 (fixed 10CH offline); selectively port only unique tests. DO NOT launch a cosmetic +TRACK PR before live ChannelId native owner is ready. Current source snapshot and screenshots are reproducible from links above.

  11. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    M3 Studio Console v3 execution started — PR #242. Adds verified-code-level control-thread atomic createBusForSelected(expectedRevision, selectedChannelIds, name, group|fxReturn) with independently allocated Bus ChannelId (does NOT borrow CH10), tagged primaryOutput versus parallel SEND, OUT NONE retaining sends, revision/duplicate/unknown-ID and budget rollback. New deterministic 2-Mic PCM test through Vocal BUS and parallel FX RETURN with an independent group FX seam; same existing DynamicPreparedMixerBlock runtime-sized offline renderer. Updates three authoritative PRD/ARD/HANDOFF v3 files, AGENTS and old v2 pointer headers; official Fender Studio Pro Console/Effects references. No native JUCE live cutover/GUI +CHANNEL claim and no real #15 hardware result. Exact-head Windows test #41203847458 is IN PROGRESS; keep PR draft until green. Preserve DEVICE4 named software I/O and ongoing #241 P0 physical capture diagnostics. Next meaningful M3 deliverable is a single ChannelId-based live graph and CH1 VST lease migration, NOT another GUI-only BUS.

  12. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    2026-10-31 M3A + M3B merged, source-verified, no fake live claims. M3A PR #242 → d638fa5e8fed165cb7b2eb1c1bf849a25aa774f2, exact-head Windows CTest #41204492501 PASS: independent ChannelId Bus for selected sources, primary OUT vs parallel POST sends, group and FX PCM, legacy-to-dynamic MAIN double-dry prevention. M3B PR #244 → main@6022a4fc0218decac694568d73c772a634eba37a, exact-head Windows CTest #41209496821 PASS: reusable single DynamicNativeSessionPublisher with precompiled ChannelId DAG, epoch/revision/mute/retirement gates, no RT allocation and 16-node 2-input PCM proof incl independent monitor and FX-return; never second JUCE callback. FAST1 PR #243 also merged, explicit one-click native SOURCE patch ACK → guarded START AUDIO and MUTE. Important NOT DELIVERED: actual dynamic +CHANNEL live button, replacing LiveDryRoutingLab fixed 10CH render, unified CH1/CHN vendor VST3 leases, Show data migration, realtime fader/send controls for dynamic node, Master PDC, hardware #15 acoustic FAT. The real main Windows ZIP candidate must come from successful main@6022a4fc CI #41210427151; main Chromium #41210427286. DO NOT claim main production FAT until those pass; PR exact-head only proves native tests. Next M3C milestone requires CH1→ChannelId owner migration and single native graph cutover, then M4 Studio Console fast UI with real accepted graph IDs. No old branch bulk merges or cosmetics while CI runs.

  13. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    M3C engine progress is now in draft PR #246 based on main@150842ef: native ChannelId fader/pan and existing PRE/POST parallel send gains with <=64 sample callback ramp, no graph rebuild; epoch/revision gated control, and accepted Console snapshot (primary OUT vs sends, fan-in source lists) attached to the exact PCM revision. New PCM/safety tests included. Not yet active in production JUCE callback, no independent +BUS GUI or physical FAT; M3D must migrate VST3 CH1/Show then cut over the incumbent single callback without dual rendering. Windows CI for #246 is running, do not report as merged/production until verified.

  14. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    M3C #246 has merged to main (Windows native CI SUCCESS, merge commit 2b581a37). M3D candidate PR #250 adds DynamicAcceptedSession transactional native control: clone roster → batch Add Channel/independent empty Bus/Add Bus for Selected/FX Return/Remove Bus → compile PCM + native epoch/revision/outputs-disarmed publish → only then accept roster. Rejections leave previous accepted routing/PCM intact; realtime ChannelId fader/PAN/parallel SEND ACK is persisted for next graph compile. New native CTest includes 2 mics, nested Bus, FX Return, MAIN/MONITOR, OUT NONE, armed-output/stale/feedback/source rollback. #250 Windows CI still running; production JUCE callback cutover, dynamic FX/Show leases and physical FAT remain NOT done. Do not activate fake dynamic WebView +BUS or claim 0FX issue #15 resolved.

  15. masarray commented on Oct 31, 2026

    @masarray
    OwnerAuthor

    2026-11-01 M3E safety seam PR #260: audit found dynamic live ChannelId fader/PAN published left and right targets as two distinct atomic stores; a simultaneous callback can consume mismatched L/R coefficients (partial PAN). PR packs both into one lock-free atomic<uint64_t>, reads exactly once per callback block, preserves existing 64-frame dezipper, no allocations or new callback, and adds extreme-left/right/center real PCM tests through FX Return. Native Windows CI in progress; NOT merged or claimed live yet. Other owners #259 FAST2 first sound and #258 Console meters remain separate and should not be overwritten. Full M3E 10CH→ChannelId production callback + FX/Show cutover remains required for true Studio One-class dynamic Bussing.

  16. 71 remaining items

  17. masarray commented on Nov 3, 2026

    @masarray
    OwnerAuthor

    E2 installed native ChannelId PCM cutover checkpoint — 2026-11-03

    Actual code: #266 commit c3ea53bb. Inside the existing one JUCE LiveAudioEngine callback/verified LiveDryRoutingLab::process, accepted 10CH/4AUX physical routes with no occupied CH1/Orbit can now run DynamicNativeSessionPublisher PCM instead of M4 DSP, via UnifiedLegacyProductionAdmission. Reuses original JUCE VST3 leases, sparse pins, independent MAIN/MONITOR, output gates/fades/secondary, mirrors actually ACKed GUI controls (fader/pan/mute/solo, physical SOURCE, AUX/monitor), and projects genuine ChannelId L/R meters; M4 retained only as fail-closed compatibility for untransferred original CH1 Orbit or unsafe graph. No second host callback or browser DSP. This is conditional 10CH/4AUX cutover, not full CH11/Group/FX Return operator workflow; universal LIVE flags remain false.

    CI EXACT SHA VERIFIED: Windows Native run 41615144204 PASS 54/54 native CTests, Chromium UI 41615144188 PASS. Experimental FAT ZIP artifact id 13221003367 (4,502,003 bytes), EXE SHA256 45bd7f9810afe618347e3ccfd5d5ce8ab6c2a7e2cd4fdc61abb1d05888ab7ae3, ZIP SHA256 ab4bdce9a6128404d0df85bda4a39767ec4a85d7afd9c5a22bbf0aa2589759e5.

    Open release blockers: actual hardware 0FX 30min / 1FX ASIO/WASAPI and crackle #15 NOT TESTED; original CH1 Orbit state/editor/Show transfer NOT DONE; full dynamic CH11/serial GROUP/parallel wet FX RETURN via #271 native WebView accepted snapshot/ACK NOT DONE; single joined GUI+engine+FAT #270 SHA NOT DONE. Preserve approved GUI #278/#285; no release promotion or customer UAT claim.

  18. masarray commented on Nov 3, 2026

    @masarray
    OwnerAuthor

    Integration checkpoint (3 Nov 2026): Joined ENGINE+GUI+FAT draft PR #270 now heads 6d0ae6ad7d70. It inherits current ENGINE #266 conditional one-callback CH1/Orbit + 10CH/4AUX ChannelId ownership; no native PCM files were overwritten here. Native GUI #271 approved appearance is untouched by these commits. A Playwright source gate was corrected to target the true dynamically labelled SVG hardware selection controller (not stale inner_text/aria), keeping canonical ChannelId selection and operator FX/group intent assertions. Chromium #41639643857 PASS exact SHA.

    The packaged Windows SHA-bound FAT collector is extended to record matching real wired 0FX vs one ACTIVE native VST3 (1FX) on identical driver/rate/buffer/pins, each with a witnessed 30-minute and SHA256 full diagnostic record; early 1FX Start rejects missing/incomparable 0FX. Tool cannot measure speaker sound, read VST3 count automatically, or ARM hardware, and always says hardwareCertified=false. Windows #41639643865 is PENDING exact SHA at this handoff. The original CH1 Show file-persistence gap, universal GUI CH11/serial Group/parallel FX Return, crackle #15 and physical FAT are NOT COMPLETE; do not promote universal/stable claims. Preserve branch ownership; test a single joined SHA before release.

  19. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    2026-11-04 WIRE-FIRST NATIVE GRAPH PRIORITY — docs PR #365

    Read ARD and GUI/ENGINE/FAT handoff. Single universal ChannelId graph: default source->MAIN real edge, unlimited legal fan-in/fan-out within admitted budgets, 0 extra samples for pure internal routing, no extra per-WIRE buffer/thread, realtime callbacks allocation/lock-free, event-driven Guardian (cycles/duplicate/double-dry warn). Proposed atomic APPLY_WIRE_TRANSACTION is NOT IMPLEMENTED: coordinate additive v8 schemaVersion=6 op + native C++ fixture with ENGINE #266, native GUI #271, FAT #270; keep old producer/GUI/CI contract until same-fixture cutover. Dedicated BUS/FX RETURN new creation postponed; preserve existing native AUX/Show/VST3 data. Physical 0FX crackle #15 remains a release blocker. One exact integrated SHA and measurable native PCM, not isolated green PRs.

  20. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    NATIVE EXECUTION FAST START — merged PR #369, main@1e5a835f. Use https://github.com/masarray/arsonmix/blob/main/docs/workstreams/START_HERE_WIRE_FIRST_20261104.md. PACKET B: #266 single ChannelId PCM graph and native atomic WIRE transaction/default real CH->MAIN; one topo callback, 0 added sample delay for pure WIRE, native fixture. PACKET C: #271 accepted ACK/snapshot GUI binding. PACKET D: #270 joined Windows FAT, #15 0FX crackle. v8 contract remains; no second engine/protocol. Do NOT treat GUI screenshot as PCM or require new BUS/FX Return feature to ship.

  21. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    WIRE v8 consumer handoff, 2026-11-04: Reference GUI Packet A is merged via #376 at main 7ddf79bb3f26bea32bc43fc4995319ec7f04225c. Native GUI #271 head 6b3588b8467cceb53571b33fab42e789cc960694 now consumes (capability-gated) additive APPLY_WIRE_TRANSACTION using the existing v8 Intent/ACK/snapshot request IDs and native accepted graph.mainChannelId, graph.edges[] typed as decimal uint64 edgeId/from/to, tap:POST_FADER|PRE_FADER|PRE_FX, finite nonnegative gainLinear, enabled, signalKind:AUDIO. Real #266 source currently does not advertise/implement this operation (checked current engine branch 96d2d30), so no LIVE audio claim and no consumer operation enabled by default. ENGINE #266 must publish the installed callback and real C++ snapshot/ACK fixture; FAT #270 must validate same exact SHA and real routing audio impulse; #15 hardware crackle still stop-ship. Do not add a second JS graph or fork native routing protocols.

  22. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    ENGINE Packet B2 — native WIRE transaction checkpoint (2026-11-04)

    Actual implementation: PR #382, latest exact SHA 0481e1db54abf2d568b948b5f177d0b143efbecf based on ENGINE #266@96d2d306a832d1dd16908da344ebfc928bb9b155. PR #375 duplicate PRE/POST suppression is already merged into #266 (previous Windows CI PASS).

    Code now adds stable ChannelSendRef.edgeId, atomic CONNECT/REMOVE/RETARGET/ROUTE_ONLY control batches with full candidate graph compile + previous-audio rollback, accepted edge inventory in DynamicConsoleRoutingView and nativeBusStatus().edges, and a real LiveAudioEngine → LiveDryRoutingLab → nativePublisher_ WIRE edit path reusing original VST lease bank. Pure wire topology may change while OUTPUT armed ONLY on verified same-epoch device/physical config; all other structural edits stay safety gated. Original M4 input/AUX sources remain expressly rejected until occupied CH1/Show migration; no false universal supportedOps.

    C++ regressions cover 7th-sample impulse direct vs parallel two-hop, exact main disconnect, stable retarget/route-only, duplicate/cycle/stale rejection, rollback, armed topology vs altered hardware patch. CI: Windows run 41801595686 in progress at time of note (NOT PASS YET).

    Not integrated/product-qualified: v8 APPLY_WIRE_TRANSACTION WebView intent + ACK + same-revision snapshot; canonical GUI consumer #271, M4-source universal cutover/VST+Show parity, joined #270 Windows artifact and physical crackle #15. Do not promote productionCallbackLive or supportedOps until the installed native transport/GUI/PCM contract is exercised on same SHA.

  23. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    Packet C contract gap found against current ENGINE #266 head e6a92d1744744d47bea0beecaffce17ad76ef270: src/ConsoleLiveBridgeV6.h presently handles NATIVE_APPLY_WIRE_TRANSACTION on the native/legacy command path, requiring canonical change.from even for REMOVE/RETARGET, whereas GUI #271 v8 consumer emits APPLY_WIRE_TRANSACTION via arMixConsoleIntent and REMOVE carries edgeId only. These are NOT yet the same installed ABI; do not treat existing native M5 transaction as v8 ConsoleIntent support. ENGINE owner must explicitly publish one installed APPLY_WIRE_TRANSACTION op in graph.capabilities.supportedOps, canonical graph.edges[] and mainChannelId, plus exact native ACK/same-revision snapshot fixture; agree whether REMOVE/RETARGET carry from in shared schema. GUI remains fail-closed until then. PR #271 has additional safety commit de063388a49f21587ad3f6760ce27d26e35313a6 (same-revision accepted WIRE ACK rejected) and latest standalone parity commit b321fef52c28322ea25912434ff421ec0b39fc97; full CI is running.

  24. masarray commented on Nov 4, 2026

    @masarray
    OwnerAuthor

    CI SMART LANES — PR #386 (2026-11-04). Native team: ENGINE src/**, engine/**, CMake, C++ tests and true native WebView bridge/monitor still trigger the canonical Windows CMake + full CTest. Pure GUI site/** / browser CSS/static HTML no longer triggers Windows on main. Ordinary engine PRs no longer waste time packaging/staging/verifying release ZIP; main, manual workflow_dispatch or integration/product-* release gate still packages. Native PCM safety, 0FX crackle #15 and integrated exact-SHA FAT are not waived. Keep audio/driver validation as the high-priority CI; avoid needless pushing of unrelated GUI/css files from ENGINE branches. PR #386 tested GUI fast path 1.6 min, native full gate still running at latest check.

  25. changed the title [-][DYN P0] Runtime-addable identical native mixer channels: stable IDs, unified FX graph, zero 10CH cap[/-] [+]P0 dynamic mixer — qualify add/delete/reorder, native FX/routing and Show persistence[/+] on Nov 8, 2026
  26. masarray commented on Nov 8, 2026

    @masarray
    OwnerAuthor

    Reconciled 2026-11-08 (main b0cea49). Stable ChannelId work in #418 and operator Add Audio Track/device recovery in #445 are already merged; #458 also merged DAW interactions. Original 10-channel limitations described in the historical body are not a verified description of current main. This issue now tracks remaining native parity for add/delete/reorder/rename, non-CH1 PCM/FX/WIRE/BUS operation, channel growth, safe VST lifetime, Show persistence and restart. Do not reopen fixed-M4 implementation threads or claim complete end-to-end audio purely from a rendered Add Track control. Cross-reference #412.

  27. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    2026-11-10 — GUI acceptance ownership after consolidation of #389

    The one approved /preview/ renderer is already integrated; this issue exclusively retains native dynamic ChannelId lifecycle: create/delete/rename/reorder with stable IDs, arbitrary reasonable count beyond CH01, real FX/WIRE/BUS/persistence and negative-ACK rollback across device changes. Cross-check ChannelId/EdgeId epoch/revision ACK and later snapshot on both Pages-assembled native GUI and real PCM. Real signal/routing/endurance FAT belongs #412; first-launch no-device authoring belongs #490. Archival record: PR #524. No functional-completion claim.

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