Skip to content

P0 native audio FAT — verify WIRE, SOURCE ONLY, BUS/FX and device recovery #412

Open

Description

@masarray

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

Scope / release boundary: Authoritative physical routing/SOURCE ONLY acceptance; no synthetic demo substitutions.

Exit criteria:

  • On current Windows hardware, verify IN→CH05→MAIN without CH01/wizard/ARM, SOURCE ONLY retains channel meter but silences MAIN, cable reconnection restores PCM.
  • Verify BUS FX→MAIN and VST3 rack bypass, retarget/delete/mono/stereo/device restart with exact graph ACK and no stale sound/routing.

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


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

Integrated source authority: main has merged native #418 and later #445 (Add Audio Track/device recovery) and #458 (DAW right-click and FX editor). Preserve the beginner workflow: choose a real device/IN once → channel/FX/WIRE → physical MAIN sound, no hidden mandatory monitor/arm/start toggles. Still open: independent user field evidence of SOURCE ONLY keeping channel signal while MAIN is silent, cable route-to-channel/BUS and reconnection, input none→mono→stereo, safe device restart, actual VST3/bus FX, and optional STREAM destination. UI ACK/Chrome simulation alone does not close these operator-facing engine gaps. No stale M4 wizard or second GUI.


P0 — ArSonMix live workflow recovery (2026-11-05)

User evidence: real Windows screenshots show generic AR MIX — ASIO / WASAPI Setup with Windows Audio, Input = CABLE Output (VB-Audio Virtual Cable), Output = CABLE Input (VB-Audio Virtual Cable), 48 kHz / 480 samples (10 ms buffer); mixer says CH1 IN None and SOURCE ONLY while its meter and MAIN meter appear active and bottom status says Output armed / REAL NATIVE PCM · FAT PENDING. Native First Sound operations live in More > Audio Setup > physical input > Patch CH1 > Arm output. The visible result is intimidating and does not make audio destination or signal authority intelligible.

P0 decisive source finding — two competing device workflows, not missing customization

web/mixer-io.js:456,760,1174-1325,1740-1749,1983-2012 ALREADY implements a custom Audio Setup · Windows Devices & Routing WebView overlay with inventory, cached driver selection, draft vs active readback, explicit APPLY DEVICE, safe mute and rollback/readback, and listener intercept for [data-native-action="audioSetup"]. Meanwhile web/approved-native-bridge.js:216-220 dynamically creates a different More > Audio Setup button and directly emits arMixAction{action:"audioSetup"}, bypassing the existing WebView overlay and opening the oversized native AudioDeviceSelectorComponent in src/Main.cpp. That explains the submitted screenshot. DO NOT create a third setup UI. First low-risk integration patch should mark/dispatch the existing More button through the mixer-io.js setup action and keep the stock JUCE selector as an explicitly labeled Advanced / Driver Recovery fallback. Test that clicking More > Audio Setup opens only the existing custom WebView overlay; no JUCE popup, no duplicate device APPLY, no duplicate listener/realtime callback. Improve the existing overlay rather than rebuild or restyle the approved mixer.

Root cause verified against main at 7ad33ef

  • src/Main.cpp:141-165: 740×590 native AudioSetupWindow embeds stock juce::AudioDeviceSelectorComponent with selector size 720×550; huge unused bottom area and no cohesive Apply/verified output journey.
  • src/Main.cpp:1093-1142: entering setup mutes via beginManualAudioSetup; finishManualAudioSetup and route re-admission happen on dialog close, not a visible Apply button.
  • web/approved-native-bridge.js:148-264: critical CH1 input routing, hardware ARM/MUTE hidden under More as an M4 recovery widget, not the main mixer IN control.
  • web/approved-native-bridge.js:266-296 plus src/ApprovedConsoleNativeTransportV6.h: G0 ChannelId controls are unavailable until genuinely admitted; never misrepresent visible faders/meter demo as controllable native PCM.
  • CABLE Input as playback device is a virtual receiving endpoint, not headphones. An additional physical monitor route is needed to hear actual speakers if streaming via VB-Cable. A virtual-to-virtual path may also risk routing feedback if a source returns its output to the same cable.
  • site/preview-v2/index.html contains UI-only demo meter implementation; native must not allow that as a real-audio substitute. Retain sole approved /preview/ skin and provenance.

Grounded benchmark (use interaction conventions, not a visual clone)

Required workflow (first-run + persistent live sessions)

  1. One visible AUDIO I/O location, preferably compact native WebView2 setup panel/docked overlay within the approved /preview/ window. Device controls: DRIVER, CAPTURE INPUT device, PHYSICAL MONITOR output device, optional STREAM/virtual output (labeled separately), sample rate/buffer in expandable Advanced. Enumerate only actual backend-supported pins. Show Virtual output — not audible in physical speakers for CABLE Input and equivalent virtual endpoints; do not silently reroute, infer hardware from names alone, or claim physical signal.
  2. Explicit APPLY DEVICE for a complete proposed AudioDeviceManager::AudioDeviceSetup transaction; show opening/verifying/accepted/rejected with real device name, rate, buffer, input and output masks; rollback and stay muted on failure. Retain advanced raw JUCE selector as recovery/diagnostic, not first-run journey. No output unmute just because a device opened.
  3. Per-strip IN [real active input dropdown] in the already visible IN control. Choosing e.g. Mono In 1 or Stereo In 1–2 triggers existing epoch-guarded native route transaction; only accepted ACK + later source readback confirms selection. Add no duplicate PCM engine. For normal strips the default output should be MAIN after verified routing if not explicitly overridden; advanced SOURCE ONLY remains accessible.
  4. Prominent MONITOR OFF / ON (speaker icon) on MAIN/output strip, and optionally channel [I] to indicate input monitoring; leave Record Arm [R] a separate future control. This replaces hidden Arm output jargon. On device change, hotplug, route fault and mismatched epoch: fail closed to MUTED; deliberate user enable after safe readback. Persistent accessible PANIC MUTE always works.
  5. Visible truth under the mixer: Input — Received / Silent / Unknown → CH routed / Unpatched → MAIN PCM / Muted → Output: <actual name> (physical / virtual); distinguish callback PCM from hardware audibility. When demo meters are active in web-only preview label them DEMO and never project them to native as real metering; native meter needs accepted callback provenance.
  6. Fast live operation: saved qualified hardware profile/route recalled on startup with safe mute; operator performs pick input → monitor on, not 4 hidden steps. Add FX from channel insert, bypass/editor visible (stage 2 after dry 0FX path verified). Advanced Matrix/Routing/Driver diagnostics remain optional.
  7. Windows window chrome can stay native titlebar initially; custom the setup content. If changing JUCE chrome later, use setUsingNativeTitleBar(false) + custom LookAndFeel/Component or WebView2; preserve OS drag/resize/close, HiDPI and keyboard behavior. Vendor VST3 editors remain native.
  8. Keep native audio RT callback ONE, no JS WebAudio routing, no allocation/locks in callback, device switching asynchronous on message/control thread, preserve plugin lease and route epoch. Never advertise literal zero physical latency: show negotiated buffer/driver report and measure round-trip where possible.

Acceptance / release gates

  • On first open output destination is always visible and distinguishes CABLE Input from speakers/headphones before ARM; no silent redirects.
  • An operator can configure driver + physical capture + playback using ONE compact pane; then explicitly select CH1 input and enable MONITOR from mixer without visiting More or a second generic dialog.
  • A single-CH dry 0FX source→CH→MAIN route has real callback ACK/readback; fader/mute/MAIN mute change PCM on the same native route; real hardware listening must be verified separately.
  • The IN None and animated real meter contradiction cannot appear without explicit PREVIEW/DEMO labeling; stale G0 states never show NATIVE LIVE.
  • Test physical-versus-virtual output warning, callback ordinal sparse ASIO mapping, driver change + hotplug mute, rejected source switch, double clicks / feedback routes, 44.1/48 kHz and 1366×768/1680×900/1920×1080; maintain exact approved skin, CSS ownership, and WebView2 bundle provenance.
  • Hardware FAT still mandatory: actual wired headphones or amp, real source and destination, 0FX ≥30min 44.1/48k if driver supports, monitor response, crackle P0 — Crackling with 0 VST3: diagnose WASAPI clock/input/output buffering before engine expansion #15 and VST3 editor/chain. CI green alone is not product ready.

Execution sequencing

P0-A device + destination comprehension and compact Apply; P0-B native per-strip IN + MAIN monitor workflow with ACK/readback, no fallback to fake G0 controls; P0-C readout/meters and First Sound FAT; P1 FX insert/rack/streaming monitor refinements. Preserve existing GUI appearance; solve the human workflow, not cosmetic themes. This is a product integration epic, not permission to merge unfinished code. Reference merged #410 and pending #411.

Activity

  1. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    Critical correction after tracing the actual Windows WebView2 bundle: web/mixer-io.js and web/monitor.js are standalone-only, not included in the shipping /preview/ Windows app. The Windows-native adapter is web/approved-native-bridge.js, injected by scripts/build_approved_native_gui.py into site/preview-v2/index.html. The visible IN field uses .strip[data-id], not legacy .strip[data-strip]. PR #416 now targets the real shipped adapter with 10-channel native routing, true M4 controls, compact device setup and MAIN monitor. Earlier assertion that the existing standalone I/O overlay could simply be promoted without moving code was wrong; this correction is a mandatory architectural source-of-truth for subsequent work.

  2. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    Implementation correction, 2026-11-05: the shipped Windows app embeds the approved site/preview-v2/index.html plus web/approved-native-bridge.js, not the legacy web/mixer-io.js/web/monitor.js I/O workflow. PR #417 now modifies the actual native bridge and ApprovedNativeRuntimeSnapshot.h to connect the existing per-strip IN, M4 fader/pan/mute/solo, and compact Audio I/O to real JUCE callbacks. No alternate HTML, CSS, audio callback, or WebAudio engine. Chromium PASS; Windows MSVC/CTest and physical hardware FAT remain separate gates. Ignore prior wording that assumed the legacy mixer-io.js panel was shipped in the approved EXE.

  3. masarray commented on Nov 5, 2026

    @masarray
    OwnerAuthor

    2026-11-05 — reconciled /preview/ ↔ JUCE First Sound baseline (source-verified, not hardware-qualified)

    One operator target: public /preview/ as generated by scripts/build_pages.py. site/preview-v2 + button material are internal inputs to that build, NOT a second application route. Native packaging stages the generated preview/index.html and appends one JUCE/WebView2 bridge without separate JS DSP.

    Current main 591444d342492120b5e848d9a0dbea0d33f15efc: approved Pages GUI remains web DEMO, not live hardware audio. PR #418 is the single proposed native DAW First Sound candidate, NOT merged.

    User journey / feature Native evidence Release gate
    Select one verified ASIO/WASAPI device in compact Audio I/O PR #418 WebView pane -> JUCE audioDeviceInventory/audioDeviceApply; regular JUCE selector under Advanced Needs Windows UI + actual hardware Apply/readback
    Choose mono IN1 / mono IN2 / enabled adjacent stereo pair on CH01–CH10 channelOptions is native-overridden only for compiled /preview/; selection -> routingSourceSwitch -> switchDryRoutingLabSource -> repatchPhysicalSource -> epoch/revision ACK + readback Verify CH01/CH05/CH10 with noncontiguous enabled pins and same source on two channels, reject invalid/disabled pin
    New physical session starts silent, without Patch/Arm/Start wizard Engine startUnpatchedProductionSession() prepares real 10-channel graph; operator route is one input gesture. AUTO monitor requested ONLY after accepted physical route to MAIN, admitted ChannelId PCM, matching device epoch, never known VB self-loop Verify actual first click produces capture→CH→MAIN; all Fader/MUTE must be active; if G0 admission fails, remain MONITOR OFF and diagnose
    Fader, pan, mute, solo, main fader Native ChannelId SET_FADER/SET_PAN/SET_MUTE/SET_SOLO + ACK/readback in admitted graph G0 physical M4 fallback controls are NOT integrated in #418; can't claim fully reliable normal mixing if ChannelId admission is absent
    User-visible per-strip OUT dropdown GUI exists; M4 repatchDryRoutingLabChannel(... SET_PRIMARY_OUT ...) engine API exists NOT WIRED in #418 UI/native dispatcher; not Adobe/FL output parity yet. Do not pretend demo select changes PCM
    MAIN hardware and monitor Active playback device shown; virtual CABLE output tagged. PR #418 supplies safety panic mute and first-user-input automatic monitor; C++ blocks known same VB-Cable self-loop independently of browser MAIN direct Monitor ON/OFF not yet properly exposed as single normal mixer control; test physical vs virtual. Optional streaming monitor is separate
    FX rack / PFL / buses Native ChannelId FX/graph exists with separate pending PR #421 for master rack power. Public preview may simulate controls. FX rack master power #421 and PFL are NOT part of First Sound candidate #418; verify dry 0FX before FX; do not merge conflicting PRs raw
    Hardware FAT ZIP New #418 branch-only packaging condition in existing p1-windows.yml, labelled ArSonMix-P0-DAW-First-Sound-UNQUALIFIED-Windows-x64 Artifact must be produced and independently verified at EXACT head; CI green ≠ hardware tested

    Tests: a real Windows headset/ASIO interface, volume initially LOW, select same physical input on CH01/CH05 (or independent two input pins), ensure CH meters and MAIN native PCM, fader/mute/solo modify actual PCM, MAIN MONITOR OFF/ON explicit operator action eventually; test no-feedback VB virtual pair, changed device/hotplug, no fake LIVE, sparse ASIO ordinal mapping, 44.1k/48k if driver supports, 0FX ≥30 min (crackle and latency), then VST3. Preserve safety mute and single callback. PR #417 contains historical G0 fader/control work not fully reconciled; avoid blind cherry-picks. PR #421 touches same Main.cpp + native bridge + workflow; needs precise diff integration after First Sound.

    CI provenance: 00fea7df Product Qualification SUCCESS run #41998987321; later experimental packaging/feedback safety commits require their own passing checks; no physical Windows audio has been heard in CI. Keep #412 OPEN until hardware FAT and required operator controls are proven.

  4. changed the title [-]P0 UX: DAW-style First Sound workflow — visible I/O, per-channel IN, physical monitor & truthful PCM[/-] [+]P0 native audio FAT — verify WIRE, SOURCE ONLY, BUS/FX and device recovery[/+] on Nov 8, 2026
  5. masarray commented on Nov 8, 2026

    @masarray
    OwnerAuthor

    Reconciled 2026-11-08 (main b0cea49). #418 is merged, plus #445 (Add Audio Track/device recovery) and #458 (DAW interactions). This issue is not a request to rebuild or re-merge the old M4/First Sound architecture. Remaining testable acceptance: physical WIRE to channel/BUS, SOURCE ONLY keeping channel signal while MAIN stays silent, rewire to MAIN, FX processing/bypass, input mode transitions, and automatic device recovery without MONITOR/ARM surprises. Confirm by current-source native PCM/hardware evidence; GUI ACK alone is not a physical sound test. Dynamic lifecycle belongs #192. Retained only for these unverified outcomes.

  6. masarray commented on Nov 10, 2026

    @masarray
    OwnerAuthor

    2026-11-10 — Authoritative native GUI/PCM acceptance inherited from #389 / integration safeguards from #510

    This P0 issue now owns native GUI-to-real-PCM end-to-end qualification, not a second GUI implementation.

    • Public Pages and JUCE WebView must display the same approved /preview/ without alternative live console. Verify accepted native intent→ACK→later matching graph-revision readback for WIRE, MAIN, BUS, SOURCE ONLY, FX picker/slot/rack bypass, fader/pan/mute/solo; reject and roll back stale, duplicate, cycle or missing-plugin commands clearly. No fake LIVE meters / G0 demonstration state.
    • Test actual physical hardware input → any ChannelId → FX rack → WIRE or BUS → MAIN → physical output, independent of CH1. SOURCE ONLY retains channel source/meters while MAIN remains silent; reconnect restores sound without Monitor/ARM/START. Confirm dynamic input NONE/MONO/STEREO, device/buffer restart, ASIO recovery and no stale WIRE routing.
    • Preserve one native JUCE PCM callback, off-thread prepared graph/FX leases, stable IDs/epochs and honest separately identified logical ACK, PCM-installed ACK and audible output. Maintain published Harmonice ID and warm sample-rate-correct 8ms bypass/dropout regression when integrating newer work.
    • Native screenshot/interaction parity at representative widths and owner hardware FAT (0FX ≥30min 44.1/48k if supported, FX, XRUN/clicks) remain required. No CI-only hardware certification.

    First-launch editable/offline model belongs #490; add/delete/reorder/Show persistence to #192; zero-FX crackle root-cause and endurance evidence to #15. Integration rules archived via merged #524 and AGENTS.md. #412 remains OPEN until these physical end-to-end gates pass.

  7. masarray commented on Nov 17, 2026

    @masarray
    OwnerAuthor

    Milestone D1 merged: PR #753 #753, commit 07cbb52. Windows exact-head 100/100 CTest: https://github.com/masarray/arsonmix/actions/runs/44075656959 ; Chromium PASS: https://github.com/masarray/arsonmix/actions/runs/44053006557.

    Milestone D physical candidate: draft PR #756 #756, exact head ee7b81e, with hardware acceptance matrix for real IN->CH05->MAIN, SOURCE ONLY metering, primary+parallel SEND, BUS FX, device recovery and VST3 startup.

    Owner-authorized FAT build dispatched: https://github.com/masarray/arsonmix/actions/runs/44081695177. FAT software build is not physical audio proof; #412 remains OPEN pending real ASIO/WASAPI evidence. No public alpha.26 change.

  8. masarray commented on Nov 17, 2026

    @masarray
    OwnerAuthor

    Milestone D software FAT candidate from exact source ee7b81e444c054623c0b2c00f102516631b9c704 is now BUILT & VERIFIED: FAT Actions #44081695177 succeeded with MSVC, 100/100 CTest, installer update and genuine historical upgrade/rollback PASS. Portable and Installer preview ZIP artifacts were uploaded with validated hashes. Draft workpack PR #756 now carries exact artifacts/hashes and an operator FAT matrix. BUILD_PROVENANCE.fatStatus=SOFTWARE_VERIFIED_PREVIEW_CANDIDATE, not physical hardware certified. Please keep this issue OPEN until observed real ASIO/WASAPI CH05→MAIN, SOURCE ONLY silent MAIN with live meter, BUS/FX/SEND, dropout and device restart records exist.

  9. masarray commented on Nov 17, 2026

    @masarray
    OwnerAuthor

    Correction to earlier Milestone D FAT qualification (18 November 2026): Owner installed the Windows FAT #756 and observed physical audio working, but ImmerSIB is missing. This is a confirmed distribution/reconciliation regression, not a missing source plugin. Inspection of the previous uploaded EXE found the GUI-facing name but no native ImmerSIB C++ descriptor literals. The generic CMake flag was OFF, so the previous 100/100 CTest did not include the two ImmerSIB tests. REVOKE #756 as a product-complete candidate (PR closed unmerged). Fix #759 merged to main with compiled-catalog/scene binary gates, and corrected exact-head FAT #760, Actions #44157301318, is running. Device audio success is accepted as a useful operator observation, not full routing/BUS/FX field FAT. Keep #412 open until the newly corrected package and full physical routing are verified.

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