Repository navigation
Description
R6 CURRENT EXECUTION / L2 PRESERVATION — 2026-10-30: Code-audited architecture and priorities · Operator GUI contract · Merged PR #181 · UX implementation #182 · L2 immutable GUI #176. Source baseline main@8251bd6 (re-fetch HEAD before editing): G3 and L2 merged; channelFormat MS1 #173, typed bus MS2 #174, independent hardware outputs MS3 #175 and uniform FX/PDC #95 are NOT completed. #15 real zero-FX audio hardware blocker remains unresolved. Protect ONE seven-row grid and fader alignment for CH1–10+MASTER, FX/ROUTE/SENDS left rail, top MIXER/FX RACK/PATCH, CH1 VST+Orbit, native epoch/revision ACK/MUTE. No old PR bulk-merge, fake Master FX/monitor pins or GUI-only LIVE.
MONO/STEREO REFORM — CANONICAL EXECUTION (2026-10-30)
ADR R5 native MONO/STEREO | R5 handoff | merged docs PR #172.
User product requirement: processing channel MONO/STEREO must be independently editable from physical input source format; bus processing formats and hardware outputs independently typed. Existing ChannelRoutingContract exposes channel format and mono fold, but NativeDryRouteCommand couples format and source and forces BUS stereo. Current G3 send editor merged, not format-aware bus processing. Do not create GUI-only stereo buttons or a second graph.
Order / ownership: MS1 #173 native channel format + GUI → MS2 #174 format-aware sends/FX AUX/GROUP → MS3 #175 physical output patch (historical #143 is CLOSED) → MS4 #95 FX/PDC/Show expansion. #15 real hardware crackle is a parallel release blocker. One vertical native+GUI+test slice per PR, latest main rebase and single integrator. No code change declared done just because docs/issues exist.
CANONICAL REQUIREMENTS & EXECUTION RECONCILIATION — 2026-10-30
This issue is now the integrated product routing / I/O / AUX bus implementation tracker. PRD v2.0 · ARD v2.0 · next-thread handoff. The older 2026-10-28 claims below are preserved as historical feature intent; never treat them as source status at current main.
2026-10-30 executable milestone sequence (one coherent vertical slice per major PR)
- G1 MIX/FX/ROUTE UI: mode bank FX/ROUTE/SENDS on every strip and independent FX RACK workspace; CH1 8 native FX slots, editor, bypass, reorder and Orbit remain accessible and unchanged in 10CH routing; truthful one-insert CH2–10/AUX.
- G2 Native custom hardware I/O: original ArSonMix Device / Inputs / Outputs M/L/R port matrix; real ASIO/WASAPI masks, physical port labels, direct main-input selection in MIX, truthful MAIN L/R output readback. No fake external output selection or raw JUCE dialog as daily frontend.
- G3 AUX bus usable workflow: 4 actual stereo FX return strips, visible PRE/POST per-channel sends, selected wire trace, 10x4 native matrix, level/pan/mute; actual ACK + physical listening FAT.
- G4 Capability-gated physical outputs: native output patch plan first, then main/monitor/AUX to distinct hardware pins on supported multi-output cards. Stereo/mute/readback and 2-output negative case tests; do not claim AUX return = HW out.
- G5 FX/PDC/Shows: unify CH2–10 multi-insert lifecycle, protected realtime ownership, delay compensation and stateful vendor VST3 session recall.
- G6 Hardware qualification: issue P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15 0-VST crackle, real 48/44.1 kHz device FAT, low-end PC interactive performance; release only after verified evidence.
Merge gate for G1/G2: operator can see INPUT → native FX → MAIN output from MIX, change input with MUTE + native revisioned ACK; FX Rack still operates after toggling; screenshots 1366/1440/1680, same-head Chromium and Windows native CI (if C++ touched), downloadable Windows artifact. UI feature must not suggest unsupported driver pins. Preserve current native 10CH+4AUX, 32 sends, existing CH1 vendor FX, pre/post, no new RT callback. Latest source status observed around main@3b89c86; re-fetch latest main before work. Issue #15 remains OPEN.
Historical original issue (2026-10-28)
ArSonMix — per-strip input assignment, mono/stereo hardware mapping, channel-to-channel sends & real bus routing
Status: principal architecture / implementation contract only. No DSP code shipped or hardware acceptance claimed by this issue.
Source baseline audited: main@20222c6b72122d3adc92c92d6bee310b96aa7299 (2026-10-28). Recheck main and open PRs before making changes.
Dependencies: #15 0-FX physical crackle and ASIO/WASAPI integrity; #95 true all-channel FX/graph and transactional edits; #85 portable performance and device remapping; #56 GUI performance. Avoid overriding active #102 FX order, #101 CI or concurrent routing changes.
Product intent
Each ArSonMix strip must be an independent native DSP channel that can:
- Pick a specific active hardware input port (mono single input) or any two individually selected ports as L/R (stereo including nonadjacent ports), regardless of strip number.
- Receive from a named internal source or selected upstream channel/bus tap, with explicit mono/stereo conversion policy.
- Have an independently configured main output destination and optional auxiliary sends to buses/returns/monitor destinations, with pre-/post-fader tap, send gain/pan, mute/solo policy.
- Reuse one physical input on multiple strips when explicitly selected, without consuming an additional hardware device, double-processing the same plugin instance or fabricating meter readings.
- Route to selected output hardware port(s), within driver capabilities and explicit user-selected active masks; remember mapping in performance/workspace. Unsupported/disconnected ports remain UNASSIGNED/MISSING and silent, never silently remap by ordinal.
- Support routing without any VST loaded; 10 visible strips must correspond to 10 real routable DSP channels, not preview clones.
Adobe Audition reference (functionality, not copied UI/code): track input/output / sends & buses, hardware channel mapping.
Why this is an architectural correction (source-confirmed)
src/LiveAudioEngine.h: oneGraphwith a single FX slots array, globalchannelGain_,channelPan_,channelMute_, onlyscratch_[0,1].start()callsdeviceManager_.initialise(1,2,...); callback copies first two callback inputs (or duplicates first) to stereo scratch, executes one insert chain, maps to output, not independent input buses.src/Main.cpp::onParameteraccepts onlychannel == 1for per-strip fader, pan and mute; there is no route assignment protocol.web/bridge.js::patchMixermarks CH2–10 preview/non-wired, master FX preview. Never enable those controls cosmetically before real DSP.- Existing
LiveGraphHandoffsupports prepared snapshot swap but currentaddVst3/removeAt/moveSlotcan detach callback. Concurrent PR Fix P0: true DAW FX insert drag, atomic realtime order, flat fader recess & single-instance #102 works on logical insert order; routing design must preserve and integrate its final approved semantics. - This is separate from P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15: a large routing graph does not fix 0-FX hardware crackles.
Architectural boundaries
DeviceHost / ClockDomain (pinned ASIO DIRECT or verified WASAPI DIRECT)
-> CapturePortRegistry + active-mask snapshot + stable endpoint identity
-> InputPatchMatrix [physicalPortId -> strip-local mono/stereo L/R]
-> PreparedMixerGraph [topological routing order, indexed preallocated buffers]
-> ChannelRuntime [trim, inserts, true meters, pan, gain, mute/solo, sends]
-> BusRuntime [Aux / return / subgroup / monitor; optional FX]
-> MasterRuntime [summation, FX, gain, meters]
-> OutputPatchMatrix [master/buses -> selected physical output ports]
-> Hardware render
Control plane: RoutingModel + Validate/Compile -> immutable GraphPlan with revision
-> safe publication at audio buffer boundary -> callback ACK -> deferred retirement
- Route references use stable
sourceId, portId, channelId, busId, not UI strip indices, display strings, or raw transient callback indices. - Important JUCE nuance: map physical device port names/IDs, enabled channel bitsets, and callback pointer indices explicitly; never assume noncontiguous enabled ports share ordinal indexes. Rebuild maps after a legitimate device-format change.
InputPatchMatrix: sparse precompiled copies/mix coefficients, no runtime menu lookup/string hash. Mono and stereo buses have explicitly negotiated widths; no accidental gain doubling when input is reused or stereo collapses.- Each strip gets its own processor instance and stable insert IDs, independent parameters, state and meter provenance. Plugin mono/stereo bus negotiation must be validated off RT with an explicit adapter/rejection policy; no blind arbitrary VST3 mono/stereo invocation.
RoutingCompiler: validate port existence, dimensions/layout, clock domain, duplicates, legal tap positions, cycle-free directed graph, feedback exclusion, plugin bus support, resource bound, output path. Compile deterministic topological order + buffer liveness/reuse plan, meter slots and PDC requirements. No channel-per-thread design.- Send semantics: upstream tap selection (pre-FX/post-FX/pre-fader/post-fader as supported), independent send level/pan, pre/post mute/solo defined in contract. Source -> channel and source -> bus both allowed; self-routing/cycles denied unless a separate explicitly engineered delayed feedback node is added in future. Disable/diagnose acoustic monitoring loops where possible.
- One full-duplex ASIO device is preferred for real-time: single clock, no extra FIFO/resampler. Different physical ASIO/WASAPI devices must not be aggregated or auto-opened; cross-clock source support requires explicit supported independent-clock adapter, measured added latency and drift handling. Preserve P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15/direct-route fences.
- PDC for parallel buses/send merges uses path latency and bounded delay lines with clearly labelled LIVE monitoring limitation. Smooth gain/mute/route transitions (ramp/crossfade) without blindly applying a discontinuous source switch.
- Graph topology published as immutable snapshot; plugin DSP states are not cloned; maintain single processor ownership, non-RT preparation, explicit life-cycle affinity, graph generation+device generation fences, callback ACK and non-RT deferred reclamation. No lock, allocation, I/O, scan, device-open, WebView or VST instantiate/destroy in RT callback. Avoid cross-thread ownership of active processors.
- Preserve existing telemetry and distinction between DSP duration, capture discontinuities and unavailable render-xrun metrics. No fake meters / fake latency.
Minimal-visual UI contract
Keep user-approved compact 10-channel + MASTER layout and low-cost static CSS. Each strip exposes a readable SOURCE chip with assigned label (e.g. IN 1 · MONO, IN 5/6 · ST, CH 1 · POST FX, UNASSIGNED). Click to select Hardware Mono, Hardware Stereo (L/R independent), From Channel, From Bus, None. Keep the full routing matrix in one compact Patchbay panel; advanced SEND/OUTPUT editing lives there or the selected strip inspector. Don't create bulky cards, hover blurs, 10 parallel animations, or cosmetic channels pretending to work. Native host supplies channel/port catalogs with revision, authoritative ACK/error; preview uses a clearly marked demo fixture.
Execution and gates (one focused PR per proof, no naive bulk rewrite)
R0 Contract tests & inventory — derive stable physical port/callback-index mapping, fake 1/2/8/16-input device registry; channel metadata protocol; no real device changes. Reconcile #102 and #15 before code.
R1 Real input mapper + strip DSP — start with 2 true channels, mono+stereo; prove CH1 can use physical port 3 and CH2 stereo L=1/R=4 when backend supports it. Expand to 10 real strips using one preallocated callback graph; every fader/mute/pan/meter independent. Preserve working CH1 session and no-VST audio; device setup may need explicit user-approved input mask change.
R2 True buses / sends / returns — source -> destination internal edges, multiple fanouts, pre/post-fader sends, shared reverb auxiliary bus, route DAG compiler, graph transactions. Strict no-feedback default, clear error and no interruption on invalid route.
R3 Safe FX + PDC + performance integration — channel-specific plugin instances/state, mono/stereo bus policy, master/bus processing, click-managed changes, plugin latency and live monitoring warnings, save/restore route IDs with #85 portable workspace. Never use localStorage as authority.
R4 CI & physical FAT — deterministic synthetic impulses/tones per physical port, random legal DAGs, cycle rejection, channel isolation/bleed, gain/pan/mute/solo, silent ports, no clipping/doubled volume, L/R nonadjacent mapping, mono/stereo plugins, independent M/S state, volume changes during VST processing, device disconnect/renumber, sample-rate/period changes, 1k+ concurrent edit attempts, memory/callback deadline telemetry. Windows MSVC+CTest + Chromium UI; actual ASIO 4+ input hardware plus WASAPI/virtual device acceptance and 0-VST acoustic tests. Require measured absence of live-edit crackle and consistent latency before calling feature complete. No duplicate CI runs or broad unrelated changes.
Explicit exclusions
Not a claim of arbitrary cross-device aggregation, simultaneous unlimited input pins, recording timeline/clip features, supported CLAP/VST2/MIDI, or guaranteed portability of physical device IDs. Don't silently activate unavailable hardware channels. Document supported matrix limits from measured device/runtime capabilities and keep future adapter extension seams.
Definition of done
Given one 8-input interface, user selects CH1 mono IN3, CH2 stereo IN5+IN6, CH3 mono IN1, CH4 from CH1 post-FX, sends CH1 and CH2 to a shared FX AUX, directs one monitor bus to explicitly selected output ports, hears only intended routes, sees true meters, safely saves/reloads mapping, and can modify valid routing during operation without a graph detach, avoidable click, cycle, false UI status or loss of stable CH1 legacy behavior.
Activity
masarray commentedon Oct 27, 2026
APPROVED SCOPE ADDENDUM — independent MONO/STEREO per mixer strip + compact live-routing UX (2026-10-28)
Owner decision: Every CH1–CH10 must have an independent, user-switchable Mono/Stereo format; SOURCE selects any available device input(s) or internal channel/bus tap. Main OUT and SEND are independently routable. Target DAW-level routing convenience (Audition / Studio One / FL Studio) in a lightweight live-performance mixer. This is a requirements/design update, not yet DSP implementation or hardware qualification.
Essential distinction (don't design naive stereo-only strips)
Model three independent layout dimensions:
sourceLayout: selected hardware mono input, selected stereo (two explicitly named L/R physical input ports, not necessarily adjacent), or compatible internal source/tap;channelFormat: user-facing MONO / STEREO channel behavior (default inferred from source on initial assign, explicit override thereafter);processingBusLayoutandoutputLayout: negotiated plugin in/out bus width, including mono-input to stereo-output reverb and stereo buses/output; cannot assume processor bus width == physical input width == channel strip width.Keep
sourceId,channelId, stable physical port identifiers,lPortId,rPortId, format, conversion policy, DSP plugin graph, fader state, sends, and output as different typed fields. A user may fan out an input to multiple independent channel instances without consuming another physical port or silently summing/doubling gain.Conversion / pan semantics:
SOLOsignal-flow semantics and pre/post sends deterministic: distinguish default post-fader sends from pre-fader (pre-fader ignores fader but mute behavior must be documented) and explicit pre-/post-FX taps. Avoid accidental double-main signal on bus send.Real-time graph transaction for format switches
requestId,channelId,sourceKind,sourcePortIds,channelFormat,inputConversion,outputId,expectedGraphRevision,expectedDeviceGeneration; statuspending/applied/rejectedand actionable reason. A slow/stale ACK must never repaint saved route incorrectly.Minimal GUI — preserve approved ArSonMix v0.8 layout
On every strip, include small always-visible control label(s):
Mmono /STstereo MUST represent channel mode, not channel mute or master mono. Distinct shape/label/icon from MUTE and CLIP; no misleading visual indicator.None / Hardware mono / Hardware stereo / From Channel / From Bus; show actual currently enabled ports, coherent L/R combo,Swap L↔R,Custom L/R, device offline reason. Fast search when port list is large. No modal covering entire performance console.Concrete LIVE workflow
Required tests added to R0–R4
Inspirations (product behavior only)
Acceptance: CH1–10 format independently selectable, real native hardware mapping and bus routing, required native mono/stereo VST negotiation, no fake GUI, safe graph transitions and checked physical-FAT. Don't merge a surface-only source selector PR as if engine capability were done.
masarray commentedon Oct 27, 2026
Implementation checkpoint (2026-10-28): PR #110 — P1B Mono/Stereo + sparse hardware-port contract opened as a stacked PR on P1A #108. Code and tests now cover independent channel-format selection, mono↔stereo conversion, noncontiguous hardware port→callback index mapping, duplicate/missing-port rejection, input fanout, rollback and Release-mode assertion execution. New CTest target included in existing Windows workflow. Local GCC optimized ASan/UBSan test PASS; exact-head Windows/Chromium CI is required and must be checked. Neither PR activates CH2–10 JUCE DSP, send buses or GUI controls yet; issue #15 hardware 0-VST crackle remains a live release blocker. Keep FX #102, portable #85 and approved GUI unmodified.
masarray commentedon Oct 27, 2026
Cross-thread integration checkpoint — 28 Oct 2026
Authoritative main:
c0d7c794a27fcdc7adbe77c6166a2e97b848b82c.Windows Native Release + CTest and Chromium passed on that exact merged commit (runs #40594052242 and #40594052249).
Complete on main: #103 Orbit Near/Far mapping and contextual GUI; #105 dearVR-only Orbit/POLAR controls + capture-discontinuity diagnostic; #106 fast-fall/ease-out meter; #107 sample-level FX ON/OFF de-click. Preserve these accepted features and do not replace with older thread source.
Workstreams OPEN: #102 DAW reorder (conflicting draft; reconcile physical FX bypass envelope, Orbit target and atomic order); #101 lean CI (separate, not merged); #108 P1A ten-strip routing pure math (draft) → #110 P1B mono/stereo/sparse physical port map → #111 P1C pre/post channel sends DAG (draft). The P1A/B/C work is verified at pure-core/test level only, NOT wired into live JUCE callback or GUI; do not claim active CH2-10 native DSP. Integrate stacked PRs in parent order, rebase on current main and rerun existing Windows CTest without dropping FX bypass tests. No duplicate CI workflow producers.
Hardware unresolved: #15 audible crackling with zero VST on WASAPI capture/render remains open, distinct from #107 hard-bypass clicking. #107 bypass smoothing still needs real plugin A/B FAT. Do not certify stable production audio until ASIO/WASAPI tests prove it.
Sync actions performed: latest dependency/integration notes are posted to PRs #108, #110, #111, #102 and #101. Keep next engineering thread anchored to this issue and read current main before new edits.
masarray commentedon Oct 27, 2026
2026-10-28 integrated milestone prepared: PR #114 — P1D end-to-end 10-channel DSP consolidates P1A #108 (routing), P1B #110 (independent mono/stereo + sparse port mapping), P1C #111 (acyclic pre/post sends) and a new
PreparedMixerBlocksample-processing executor. Validated audio fixture covers physical stereo/mono → channel FX seam → fader → sends → return without physical input → stereo master, solo graph closure, fader-zero pre-send, bad-route rollback, and 1000 no-allocation audio blocks. Branch starts from current main and preserves GUI/Orbit/FX bypass. Chromium PASS; Windows native CI is running. Not live hardware output: callbacks, VST3 bus negotiation and GUI SOURCE remain gated behind #15 physical crackle diagnosis, safe plugin ownership, epoch/ACK and FAT. Do not mark issue done or claim all 10 live FX yet.masarray commentedon Oct 27, 2026
ArSonMix final thread-scope handoff — 28 October 2026
Authoritative integrated
main:66989fa46f004aa852207f1159bf3e1b8e9506c1.The user-approved dearVR Orbit/POLAR, contextual Near/Far, main GUI and meter ballistics remain intact (#103, #105, #106). FX ON/OFF de-click was merged in #107 (
c0d7c79): sample-accurate bypass crossfade for zero-latency inserts, fade-through-zero for latency-bearing inserts, no new realtime allocations or engine buffering. Windows Release + CTest and Chromium passed, but real-hardware click-free playback still requires a separate FAT.The thread-owned CI work is now also complete: PR #101 merged at
66989fa. It was re-authored from newestmainso that all current Windows targets remained in the one canonical build, includingARMixFxBypassEnvelopeTests, dearVR/POLAR tests and WASAPI regression tests. PR-head Chromium and Windows MSVC/CTest passed. PR Windows artifact upload was deliberately skipped; only the main Windows run (or explicitly dispatched branch FAT) uploads verified ZIP. UI PR skips Firefox, and Pages publishes only a successful, non-obsolete main GUI revision. The tested PR and merged commit have identical source tree SHA0bb6d2701aeb92404f0d386bd6a2587fee3747a1.Other threads own remaining open work: PR #102 FX drag/reorder is still draft/conflicting and MUST preserve the de-click per-physical-insert envelopes. P1A #108 → P1B #110 → P1C #111 are stacked pure routing/mono-stereo/sends engine plans, not activated multichannel JUCE paths. Integrate in dependency order and keep Windows CTest targets current. Issue #15 zero-plugin WASAPI crackle is a separate open physical-A/B diagnosis (capture discontinuities are not proof of hardware render underrun). Do not merge old GUI/engine snapshots or declare production audio universally crackle-free.
This comment is the cross-thread completion and transfer record. The current main's post-merge Windows artifact and UI publishing jobs are scheduled; their outcomes are visible in GitHub Actions.
masarray commentedon Oct 27, 2026
P1C implementation checkpoint (2026-10-28): PR #117 adds a runnable device-independent block DSP processor for 10 mixer strips + 2 AUX buses. Includes channel-to-channel input, physical mono/stereo routing, send level, AUX return, MAIN routing control, independent gain/mute, optional prepared insert hook, deterministic DAG ordering and cycle rejection, transactional compiler, preallocated buffers, and sample-exact CTest vectors. It is stacked after P1A #108 and P1B #110 and intentionally does not modify JUCE live callback, product GUI, Orbit, FX #102 or device policies. Its exact-head Windows/Chromium CI is running; hardware audio and graph-swap lifecycle remain mandatory before native live activation. Do not represent this offline DSP milestone as a real-device-ready mixer.
masarray commentedon Oct 27, 2026
Implementation progress P1D 2026-10-28: stacked PR #115 over #112 composes real physical input maps, 10-strip format conversions, FX hook, per-strip pan/mute/solo, true pre/post-fader sends and internal bus returns, to MASTER. Includes deterministic 10-source mix, nonadjacent hardware input, input-loss fail closed, zero-fader PRE sends, mono-to-stereo unity and graph rollback tests. Only dry prepared processing; native live CH2–10 FX remain disabled, and #15 hardware 0-FX crackle is still a blocker. Windows exact-head CI currently running.
Merge P1D: validated integrated ten-channel offline DSP execution
masarray commentedon Oct 28, 2026
Cross-thread authoritative baseline (2026-10-28):
main@4951f3balready contains P1D integrated 10-strip DSP from merged #114, Studio Backup P3 from #116, and pre-fader/DAG corrections from #118. New P1E PR #119 advances to opt-in real JUCE callback DRY 10-input routing with device epoch/revision, native GUI ACK, fail-closed stale device and real Windows FAT operator flow. #102 FX drag remains pending/isolated. Please do not bulk merge older stacked P1A/B/C/D routing code over main or treat isolated offline FX hooks as live per-strip VST3. Reconcile any improvements as small tested deltas on current main. P1E Windows native CI is being verified; #15 zero-VST crackle still requires physical FAT, not inferred from CI.masarray commentedon Oct 28, 2026
2026-10-28 reconciliation follow-on: main now
15450e6(P1D #114 MERGED, P1E port-epoch publisher integrated, FX drag+atomic reorder integrated with bypass/Orbit/meter). PR #115/#117 remain overlapping offline graph variants and are not being blindly merged. New PR #123 P1F adds actual JUCE active input/output physical mask snapshot, binds it to preallocated 10-strip DSP + safe revision/epoch publisher, and includes native end-to-end audio-vector JUCE/CTest tests. Exact-head Windows CI: https://github.com/masarray/arsonmix/actions/runs/40630237793. Native live callback, independent per-channel VST3 instances and #15 zero-VST hardware qualification remain gated; no GUI fake activation.masarray commentedon Oct 28, 2026
Cross-thread engineering handoff (28 Oct 2026): #125 is the verified current-main handoff. Audited
main@3b327bd5with exact-main Windows/Chromium/Pages CI PASS; #121 P1E DRY ROUTING LAB already merged, #123 P1F current head is being actively updated by another thread. Please read #125 before P1F/CH2–10 work to avoid duplicate/old PR merges. P0 #15 zero-FX crackle FAT remains unresolved; prepared/native dry lab != production ten-channel VST3.masarray commentedon Oct 28, 2026
P1F–P1G MERGED to main (2026-10-28): PR #123 squash-merged as 240b901. Base reconciled against latest approved GUI main before merge, no loss of CH1 VST3/Orbit/bypass, atomic FX reorder, 10-strip layout or backup. Functional native P1G Routing LAB: JUCE real active physical IN/OUT masks + fail-closed epoch/channel count, 10 physical MONO/STEREO or internal BUS channels, up to 32 PRE/POST-fader internal sends, MAIN/mute/solo/gain, true per-channel post-fader peaks, transactional revision ACK and 256-sample de-click; operated via the existing SIGNAL CHECK Routing LAB without inventing another engine. Exact-head Windows MSVC/CTest PASS, Chromium PASS. Post-merge main release Windows CI now running. Still DRY zero VST3 LAB; per-CH VST3, hardware AUX outputs, portable routing and #15 zero-plugin hardware crackle FAT remain incomplete; never claim DAW production/live FX qualified from CI.
masarray commentedon Oct 28, 2026
2026-10-28 — P1H SIGNIFICANT MILESTONE MERGED to main: PR #126 was squash-merged into main@758ec23e959305115d7fc4e23f7c9f73ddad43c8 AFTER reconciliation with the full cross-thread GUI baseline (main@240b901).
Actual native functionality: the already-live P1G Routing LAB now permits one independent stereo-compatible VST3 per channel, including BUS return effects, with the existing
PreparedMixerBlockexecuting actualjuce::AudioProcessor::processBlockBEFORE PRE-fader sends/fader; no competing mixer engine. Up to 10 isolated vendor plugin instances per published graph. UI picks actual IDs from native VST3 catalog; no JS-supplied file paths. Off-RT creation/prepare/layout validation and off-RT, reader-lease protected retirement; no callback malloc/locks from host; callback chunks at device quantum. Zero-latency-only until PDC, rejected plugin and route fail closed. GUI approved CH/MASTER/master mono/button palettes unchanged; normal production CH1 FX rack, Orbit, drag reorder, FX bypass, portable backup retained.Exact PR-head validation: Windows MSVC application+package PASS, 43/43 CTests PASS including
P1HRealNativePerChannelFxAndSafeRetirement, static ZIP verification PASS; Chromium GUI PASS. Post-merge main CI running; P1H hardware artifact available only after success/upload, do NOT invent an artifact. Full FAT guide:docs/P1H_NATIVE_MULTICHANNEL_VST3_20261028.md.Next gates, not done: LAB still operator-opt-in rather than production normal mixer; 1 insert/strip and stereo main only; no vendor FX parameter editors in LAB, plugin state persistence/restore between APPLY, multi-FX racks on CH2–10, mono-only/mono-in-stereo-out bus layouts, PDC, physical AUX outputs and #15 zero-VST hardware crackle FAT. P0 live DIRECT/independent clocks must be tested. Existing offline duplicate PR #115/#117 must NOT be bulk merged. New thread: start from main@758ec23, audit fresh main SHA/CI first, then plan native plugin persistence + multi-insert rack lifecycle, validate hardware after #15.
masarray commentedon Oct 28, 2026
ARD REVIEW / AUDIT ROUTING — 2026-10-28 17:07 WIB (read-only source review)
Exact baseline: main
758ec23e959305115d7fc4e23f7c9f73ddad43c8; Windows JUCE/MSVC, CTest, Chromium and GUI publication all SUCCESS. Source/contract audit only, not visual runtime screenshot comparison or real interface hardware FAT. Prioritise this review over older issue text that predates P1G/H. No code edited in this audit.Verdict
Architectural core is advancing correctly (preallocated 10-strip DSP, validated sparse physical mapping, cycle-free PRE/POST sends, real BUS return, owned per-channel VST3 processors, revisions/epochs and realtime leases). Not yet equivalent to Adobe Audition or Studio One's routable console in everyday operation. Main GUI still has CH1-only real FX and automation; CH2–10 FX rack/strip events are explicitly labelled preview even though opt-in Routing LAB has native VST3. This is a broken UX/SSOT seam, not a request for cosmetic redesign.
Confirmed technical debts (from latest main)
web/bridge.js::renderRack/patchMixer: CH2–10 and MASTER FX controls still disabled/preview, whileweb/monitor.jsimplements an entirely separate nested Routing LAB with VST3 selection and APPLY.src/Main.cpp::onParameterdispatches fader/pan/mute only when channel==1. Main console is not the real 10-channel control surface.engine/core/MixerRoutingPlan.h::stripCount=10is shared for physical sources and BUS returns: creating a BUS consumes one of ten strip slots. Goal must remain 10 independent physical SOURCE strips plus distinct AUX/FX/subgroup/output buses.ChannelSendPlan::toMasterandLiveDryRoutingLab::processroute to stereo callback output[0]/[1] only; a recorded active output mask is not equivalent to selectable arbitrary hardware outputs, headphones/cue/stream independently.LiveChannelFxGraph.hhas one stereo-compatible FX slot per channel; no channel insert racks, plugin editor, mono-only/in-out format negotiation, optional sidechain or master inserts in multichannel LAB.VstBusLayout.hturns off auxiliary buses.LiveDryRoutingLab::applybuilds a new graph and recreates all selected native plugin instances, not just changed slots. Hence APPLY route/send edits reset vendor plugin state/tails, perform file/instantiation on message thread, and require MUTE. This is not acceptable for a production live DAW console. Retire old objects off RT is correctly designed but insufficient to preserve state.NativePresetArchive.hserializes a CH1-only rack, not the multichannel routing graph and all plugin states;web/app.jsstill stores cosmetic scene state separately in localStorage. No canonical persistent state for input, bus, output and VST3.src/LiveAudioEngine.h::startasks JUCE initialise(1,2): ten physical inputs depend on user-activated device masks and hardware compatibility. Currently one confirmed direct pair for output. P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15 physical crackle with zero VST3 remains unresolved despite software CTest PASS.RealtimeRoutePublisherhas a bounded no-RT-alloc graph swap lease; positive. P1H calls third-partyprocessBlockon callback (vendor code may block/allocate/throw), so hard deterministic RT and low-end hardware performance are not proved.Decided long-term architecture (not another parallel engine)
ONE SSOT
MixSession(stable ChannelId/BusId/PortId, saved schema, device identity, user patching, FX chains, parameter state, scenes) ->RoutingValidator/Compiler(cycle, bus width, endpoint mask, graph reachability, bus role, PDC, resource budget) -> immutableRenderPlan(preallocated buffers, topological execution, independent source/bus/master output nodes, device-bound revision) ->RT ParameterLanefor fader/pan/mute/send knobs with per-block smoothing and no graph rebuild ->FxInstanceRegistrywith stable plugin instance IDs/ownership and off-RT lifecycle, immutable chain order edits/FX de-click, vendor state persistence and retire-after-ACK ->OutputPatchMatrixto explicitly validated hardware MAIN/CUE/STREAM outputs -> async low-cost typed UI status/ACK. UI overlays only EDIT the SAME native MixSession; no separate LAB truth.Ordered meaningful gates
P0 FIRST: #15 0-FX hardware crackle/root-cause test on ASIO DIRECT, WASAPI DIRECT, independent clock as explicitly separated. Do not interpret CPU/callback gap as confirmed hardware xruns.
M1: One semantic
MixSessionSSOT and complete native per-strip control UI; 10 real fixed inputs, separate buses. Existing GUI and plugin functionality must not regress.M2: stable multi-FX instance chain and plugin state persistence; 1000 knob/send edits must not recreate an unchanged processor, lose tails/state or detach callback.
M3: proper patchbay incl named hardware outputs/cue and full bus/sidechain/mono-stereo layout negotiation. Reject unavailable/identity-changed port.
M4: DAG PDC and phase-aligned buses, plus reliable live low latency/mixing modes.
M5: real hardware FAT low-end/onboard GPU + ASIO/WASAPI, live silence/dropouts, hotplug, restore, USB portability, soak, load and CI gates.
Do NOT merge overlapping stale #115/#117 wholesale or replace latest GUI. Avoid phase-by-phase mock features; one milestone must include real native routing + observable GUI + deterministic tests + source-level cross-thread reconciliation.
masarray commentedon Oct 28, 2026
P1J CROSS-THREAD CHECKPOINT — 2026-10-29 WIB. User screenshot confirms non-LAB Windows GUI still shows NATIVE CH1, 10CH routing only activates after explicit DIRECT+MUTED LAB APPLY. Work on PR #130, head
05d584552ef9ef0180ac70ee0addaf80f1ea1945, based onmain@abcdf2b9(latest P0 GUI + P1I native controls; new main7c48cfb1only handoff docs). Do not overwrite any newer GUI or FX Orbit state. Native P1J concrete code: shared stable per-channel JUCE VST3 ownership across routing graph revision if same (strip, catalog ID, device epoch, quantum); publisher control-only accessor, RT atomic try-enter and C++ exception protection, late retirement; native acceptedRoute stored only on ACK and control thread, not mutated in deviceStopped (critical race fix); GUI Routing LAB rehydrates pins FX sends + live fader PAN controls from authoritative native state; SIGNAL FLOW tab now opens actual routing editor; main CH1 preview normal path unchanged. C++ real AudioProcessor test extended for same pointer, no reprepare, old reader safe, RT allocation; JS contract updated. Chromium exact-head CI PASS; Windows x64 exact-head 40770500881 still building as of this comment; DO NOT MERGE without Windows x64 CTest and package success. Earlier failures were test contract staleness and are corrected in latest head. Limits: structural APPLY still requires MUTE and LAB opt-in, 1 VST3 per channel, no native portable full-mixer snapshots/PDC/AUX hardware outputs, and zero-FX #15 crackle hardware qualification outstanding. Do not merge stale PR #115/#117. Full FAT docsdocs/P1J_STABLE_FX_ROUTING_SESSION_20261029.md.masarray commentedon Oct 28, 2026
P1J COMPLETE / SOURCE OF TRUTH (2026-10-29 WIB). PR #130 SQUASH MERGED at main@bc11edf0633fadbdf3e886f1bb55c108d7009425, after reconciling the entire P1I 10CH native UI/P0 latest approved GUI/Show snapshot progress and #129 handoff docs. Exact PR-head
05d584552ef9ef0180ac70ee0addaf80f1ea1945validations: Windows MSVC build, 44/44 native CTest, ZIP integrity SUCCESS, Chromium UI SUCCESS. Postmerge main CI running: Windows 40772310172, Chromium 40772310160. DO NOT assert post-merge is green until it completes.\n\nSubstantial functionality: unchanged per-channel VST3s retain original AudioProcessor instance/vendor parameters/tail memory across accepted native routing graph updates; control-only shared ownership, zero-lock/audio-alloc RT try-enter, deferred release; changed plugins instantiated/retired separately with identity/epoch/revision checks. Native acceptedRoute is updated only AFTER graph publication and exposed in status; routing GUI hydrates physical pins, FX catalog IDs, PRE/POST sends from native state and overlays live fader/PAN/mute/solo. Carefully fixed deviceStopped race: JUCE var native session can only be changed on control thread, epoch atomics invalidate status on device thread. Top-level SIGNAL FLOW page of Windows app now offers OPEN ROUTING MATRIX, using the same existing P1G/H/I LAB engine without another DSP/GUI branch. Prior standalone HTML demo retained in parity, latest BUTTON-180/snapshot GUI and Orbit/CH1 full VST3 untouched. Full FAT guide:docs/P1J_STABLE_FX_ROUTING_SESSION_20261029.md.\n\nStill NOT completed: routing is opt-in until user mutes OUTPUT, selects real DIRECT ASIO/WASAPI and APPLY; screenshot badge NATIVE CH1 is correct for ordinary CH1 mode. 10 fixed physical INPUT strips + separate AUX/FX buses is NEXT architecture change; true multiple VST3 inserts/channel with vendor GUI/state and native Shows is not shipped; no hardware AUX/CUE, PDC or field validated #15 0-FX crackle. Avoid merging ancient overlapping #115/#117 or claiming Adobe/Studio One parity. Next thread MUST audit main SHA and post-merge CI first and preserve all parallel GUI snapshot and Orbit work.masarray commentedon Oct 28, 2026
P1K CROSS-THREAD COORDINATION — separate AUX buses native DSP milestone. PR #136 is currently testing, based on exact
main@faee66f(P1I, P1J, P4 ShowV2). Do NOT merge until exact-head Windows CTest and Chromium succeed. User's next routing priority is 10 permanently available input strips + FOUR ADDITIONAL AUX stereo buses without using CH9/CH10 as FX returns. P1K extends the existingPreparedRenderGraph/PreparedMixerBlock/LiveChannelFxGraph/LiveDryRoutingLabandNativeDryRouteCommand, with real 10-source samples, AUX PRE/POST tap, independent AUX FX+gain+MAIN, off-RT preallocation, catalog validation, revision/epoch, P1J plugin instance retention, and native accepted-route GUI ACK. Native C++ and JUCE processor tests and WebView UI contract added.web/monitor.jsincludes AUX1..4 controls and CH/AUX send destination; no premium GUI layout changes. Parallel PR #133 changessrc/Main.cppand GUI layout, so reconcile P1K's onlyMain.cppAUX VST3 catalog lookup with PR #133 if that PR merges first. Do not overwrite PR #133 visual and startup fixes, Orbit, P4 ShowV2, original CH1 production FX. Zero FX #15 hardware FAT remains open. CTest Windows run: https://github.com/masarray/arsonmix/actions/runs/40778420059; Chromium current head: https://github.com/masarray/arsonmix/actions/runs/40778419698. Full docsdocs/P1K_TEN_INPUTS_FOUR_DEDICATED_AUX_20261029.md.masarray commentedon Oct 28, 2026
P1K SIGNIFICANT ROUTING MILESTONE MERGED (2026-10-29 WIB). PR #136 squash-merged to main@662b060f7455abdc76e642cafd9b99d913852e91, based on global P1J FX lifecycle, P1I 10-channel console controls, P4 native ShowV2 and latest P0 GUI. Do NOT regress these baseline milestones or overwrite ongoing premium GUI PR #133.
What actually works: Ten input strips CH1–10 all remain independently available even when 4 dedicated stereo AUX/FX return buses are enabled. Native
AuxBusPlan.hcompiles 32 bounded source PRE/POST sends to AUX1–4, rejects invalid/disabled destinations, duplicate edges, gain/pan errors and feedback-capable AUX→INPUT attempts (impossible by type/indices). ExistingPreparedRenderGraph,PreparedMixerBlock,LiveChannelFxGraph,LiveDryRoutingLabperform actual signal render from physical input→VST3→PRE/POST tap→AUX stereo buffer→independent zero-latency AUX VST3→AUX gain/mute/Main→stereo master in same verified JUCE callback. AUX scratch allocated off RT only when needed (428192 floats), no new callback locks, heap allocations, threads or device reopening. AUX FX uses P1J shared processor identity/instance reuse across graph edits. Native command validatesauxes[4]andto=10..13;Main.cppresolves AUX plugin identity through trusted native catalog; status/ACK shows auxCount/auxSendCount/auxFxNames/auxPeaks. User can edit this in existing SIGNAL FLOW → OPEN ROUTING MATRIX, with native-ACK state hydration. Old P1G internal BUS representation remains for backward compatibility.Exact PR-head verification: Windows/MSVC full app build, 46/46 CTest PASS, verified ZIP package and Chromium UI PASS at exact
f9a6a3bd67b0de4deb2b17e7dd417bbef54ca3b7. CTest includesP1KTenIndependentInputsAndFourDedicatedAuxReturnstesting actual 10 input PCM, two independent AUX signal paths, PRE/POST, FX hook sample math, mute, compiler rollback and no allocations; updated JUCE processor test executes realAudioProcessor::processBlockon AUX and verifies stable instance reuse; P4 ShowV2 test still green. Post-merge new main Windows run 40780687293 and Chromium run 40780687265 were still in progress at last check; do not claim post-merge green or final FAT artifact before they pass.Reality/gates: LAB is still opt-in and requires operator OUTPUT MUTE + verified single-clock DIRECT device and real physical active input channels. Four AUX currently RETURN to the same stereo master (no selectable independent physical AUX/CUE outputs). One stereo-compatible VST3 per strip/AUX, no native PDC, multi-FX rack, native Full Show state restore for new AUX topology, and #15 real hardware crackle qualification unresolved. The separate GUI layout PR #133 must be reconciled if modifying
src/Main.cppor finalizing premium GUI; P1K did not touch itsweb/app.js/CSS/standalone files. Do NOT merge stale #115/#117 wholesale. Next genuinely significant milestone: production multi-FX racks with preserved stable vendor state and native full session backup/recall, then output patchbay + CUE and phase-aligned PDC. Detailed hardware FAT indocs/P1K_TEN_INPUTS_FOUR_DEDICATED_AUX_20261029.md.masarray commentedon Oct 28, 2026
CANONICAL HANDOFF UPDATE (2026-10-29): Future ArSonMix routing/FX/Show/GUI work must start at Issue #137 — full cross-thread engineering handoff. P1I/P1J/P1K, 10 INPUT + 4 dedicated AUX VST3 buses and latest #133 UI are already MERGED into
main@6077513. Do not repeat PR #136. At this checkpoint, main UI smoke is RED attests/ui_smoke.py:951(MASTER peak-hold reset after clip click), while main Windows CI still builds. P4 #134 is active with changing head; avoid merging stale mixed diff into main. #137 contains exact architecture, control/audio thread invariants, file map, merge and physical FAT gates, and next-thread prompt.masarray commentedon Oct 31, 2026
ISSUE RECONCILIATION — historical/overlap closed, NOT declared solved. Original channel input/mono/stereo and sends/bus routing program is now represented by universal ChannelId and typed edges in #192 / ARD v5; older fixed-channel and patch sequencing text is not current acceptance.
Current authority: PRD v5, ARD v5, native epic #192, GUI L2 #176, and hardware #15. Existing original comments and links remain.