Skip to content

P0 first-launch: genuine SOURCE NONE ChannelId mixer, zero-capture output-only FX workspace - #491

Merged
27 commits merged into
mainfrom
fix/p0-cold-start-silent-native-workspace-20261109
Nov 10, 2026
Merged

27 commits merged into
mainfrom
fix/p0-cold-start-silent-native-workspace-20261109

Conversation

@masarray

@masarray masarray commented Nov 9, 2026 •

Copy link
Copy Markdown
Owner

ASIO4ALL independent capture and playback — 5a923a7

The previous candidate treated a missing source or the loss of the saved MAIN pair as a reason to reject the whole native mixer. It also omitted pin labels when saving a session.

  • Keep the native ChannelId graph admitted with verified playback and zero inputs. Preserve a disconnected input's physical identity separately from callback ordinals. Reconnect a returning named pair, including when JUCE retained a zero input mask.
  • Recover a single recognizable physical speaker pair from the same registered ASIO4ALL driver when the old MAIN disappears. Keep the user's previous monitoring intent across that reset. Reuse existing FX leases when the negotiated sample rate permits.
  • Persist the input/output labels and waiting input patches in the v1 session format. Older sessions still load. Prevent a removed or deliberately repatched channel from reconnecting a stale source.

Native C++ regression verifies active PCM before and after missing capture, silence and working fader edits while disconnected, restoration of the stereo patch, and persistence. 303 Windows JS contracts and Chromium cold-start smoke pass locally. Exact Windows run.

Windows candidate 5a923a7d passed MSVC, 67/67 CTest, ZIP verification and Inno graceful upgrade. Installer ZIP · Portable ZIP.

Physical ASIO4ALL/Realtek listening, native calls that never return, and full editor operation without an available playback device are pending.


ASIO4ALL field follow-up — 29f29ac

The screenshot’s NEEDS REMAP was reproduced at the binding-policy level. The old policy rejected the entire ASIO pin list on an inventory change and the automatic retained-mixer recovery only handled WASAPI.

  • Reattach the original MAIN and capture pins by unique labels after a JUCE ASIO4ALL restart. Correct a stale ordinal mask once, verify the same registered driver and hardware readback, and resume prior MAIN intent after a valid graph bind. Reuse compatible native FX where the format and pins still match.
  • One deliberate Audio I/O Apply can select the sole recognized physical stereo pair when ASIO4ALL exposes a different speaker pair. Ambiguous or virtual playback does not qualify for this fallback. Audio I/O lists the resolved MAIN pin labels.
  • If a previously patched capture pin temporarily disappears, retain its saved binding for the next negotiated epoch. Report the precise missing MAIN/capture condition.

Local evidence: 190 JS contracts, native host compile preflight, Chromium 3-size smoke, and real native PCM fixture with sparse physical pins and reversed MAIN callback order passed. Physical Realtek/ASIO4ALL listening and protection against vendor calls that never return remain pending.

Windows candidate 6285f80: MSVC, CTest, ZIP integrity, Inno installer and graceful upgrade passed. Windows run · Installer · Portable. Physical ASIO4ALL/Realtek listening is pending.


2026-11-09 field recovery candidate

Exact HEAD c9563d14ed0ff91df81b077c8630610e986fd864 on main 381c140200d2f4ecb965183a2da1ba1e46d5d9c2.

  • Hide the reference mixer until a native roster is accepted. Verified playback with zero capture pins projects the actual ChannelId graph and keeps FX/WIRE controls available.
  • Audio I/O opens the selected backend directly, avoiding JUCE's intermediate default device open and fixed message-thread wait. Serialize re-entrant Apply and record close/open/verification/bind/rollback phases.
  • Audio I/O → Advanced → Apply two selected playback pins as MAIN is an explicit rebind to verified physical playback pins. A normal Apply preserves the saved MAIN pair. Restoring a saved session can replace a pristine startup graph after verified device recovery.
  • More → Test MAIN output renders a quiet tone through the selected MAIN pair without reopening the driver or replacing the session. Copy Audio Report includes the build SHA, post-gate JUCE output buffer peaks and the last test peak.

Local qualification: 190 fast contracts, approved bundle provenance, JUCE projection/oscillator/session recovery tests, native host syntax preflight, and Chromium cold-start/sparse MAIN/FX/WIRE/identity/meter checks passed. Physical ASIO4ALL/Realtek/VB-Cable FAT and actual hang stack evidence remain pending. Fully device-independent session editing and isolation of hung vendor driver/plugin calls remain open. Keep this draft until qualified.

Windows MSVC, CTest, portable ZIP verification and Inno install/upgrade/uninstall checks passed on the exact source above. Windows build · Installer ZIP · Portable FAT ZIP. Hardware listening remains pending.


Why

Fixes first-boot user trap documented in #490: mixer editing/FX incorrectly waits for a microphone patch. A real native SOURCE NONE graph must exist before any hardware IN mapping and output-only devices should be valid for internal Player and FX.

Source changes

  • LiveAudioEngine::startUnpatchedProductionSession: explicitly promote only the silent, same-epoch M4 prepared topology to the existing ChannelId PCM publisher at boot. No new callback, bridge, WebAudio or dummy input pin. Output remains DISARMED, physical inputs NONE.
  • LiveDryRoutingLab: constrained admitSilentUnpatched path checks all 10 SOURCE NONE channels; retains historical stand-alone unpatched M4 fail-silent semantics and output admission fence. 0 active capture channels supported with 2 genuine output pins.
  • Strict DIRECT route fingerprints permit canonical zero input mask only, reject missing output; JUCE port snapshot reports zero real capture pins rather than faking one.
  • web/approved-native-bridge.js: deviceReady no longer requires enabled capture inputs, so a genuine accepted graph may expose native FX/track/WIRE before any mic selection.
  • .smix codec and device rebind: preserve zero physical inputs with no source patch; invalid source/capture data still rejected.
  • Unit/regression tests: output-only mask attestation, DIRECT safety, persisted zero-input mixer.

Qualification still required

This is an intermediate vertical slice, NOT the full #490 acceptance. No-device purely offline session editing, first-paint reference model disposal, a nonblocking vendor VST3 prepare path, and real Windows first-install 0-IN/ASIO test remain. Do not merge until scoped Linux preflight + exact-SHA Windows CTests/FX/native GUI smoke are checked, and no regressions to #486 Player, compressor or #487 Harmony. Do not publish an installer or claim physical ASIO validation yet.

Closes part of #490 (do not close issue).

…puts

Output-only DIRECT fences and callback mask validation now permit zero capture without inventing input pins. A verified SOURCE NONE startup publishes the same authoritative native PCM graph used for FX, track and WIRE editing. Preserve output muting and driver identity, add zero-capture regression cases.
…ter, persist 0-input sessions

Only a verified ten-channel SOURCE NONE starter may enter native ChannelId without physical capture. Keep M4 fail-silent until admission, retain exact device fence and enable portable .smix roundtrip with zero capture. Extend tests.
@cloudflare-workers-and-pages

cloudflare-workers-and-pages Bot commented Nov 9, 2026 •

Copy link
Copy Markdown

Deploying arsonmix with  Cloudflare Pages  Cloudflare Pages

Latest commit: 3015960
Status: ✅  Deploy successful!
Preview URL: https://cbd86acf.arsonmix.pages.dev
Branch Preview URL: https://fix-p0-cold-start-silent-nat.arsonmix.pages.dev

View logs

…der true zero PCM

Adapt startup banner contract to zero-capture no longer blocking FX, add JUCE-backed 0-IN 2-OUT ChannelId bootstrap/track/PCM test; retain M4 unpatched refusal.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native 75e99c6

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native 178631b

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Exact-SHA qualification checkpoint (2026-11-09)

HEAD 178631b5b72750ae7a01f18487385c5ce18daaaa (main remains 381c140200d2f4ecb965183a2da1ba1e46d5d9c2). Fast Linux UI+native PR checks passed. Owner-triggered Windows MSVC/CTest run 42599706023 passed fast preflight and GUI provenance, configured CMake/JUCE, native executable and CTests compilation still in progress when inspected. Earlier exact-SHA Windows run at previous HEAD failed a stale JS test for DIRECT input/output mask guard; updated current test verifies the canonical input zero mask and still rejects output zero. No Windows/physical FAT result asserted yet.

Remaining release blockers: no-device true edit mode, first-launch fake reference-state removal, vendor VST3 scan worker indefinite shutdown join / message-thread plugin prepare, physical ASIO/out-only/Player FX field qualification. Keep Draft; do not merge, publish installer or close #490 based only on fast CI.

Allow explicit NONE capture for separate Windows audio endpoints and empty enabled capture masks for genuine ASIO devices, preserving exact verified stereo outputs, user choice, DIRECT fencing and readback. Match native bridge device Apply/ACK to accepted zero-capture graph; add regression contract.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native b0f485b

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui b0f485b

…fied device restore

Reject missing/stereo-incomplete playback but do not require a microphone when restoring the exact saved output-only endpoint. Retain DIRECT fingerprint and fail-silent guards. Repair advanced setup guidance and add regression test.
… VST3 scanner

Keep verified third-party catalog and Show/backup/handoff gates, but make native built-in FX available on initial SOURCE NONE workspace while background plugin scan is busy. Add guard contract regression.
… VST3 discovery

Built-in JUCE processors are already present in the executable: expose them in the approved Windows picker on first paint and after every external catalog refresh. Label their real native format, without fake vendor metadata or browser DSP.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native df15d63

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci ui df15d63

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

P0 vertical slice checkpoint — native output-only Audio I/O and instant built-in FX

Current HEAD df15d63b8b63ae8dfdd417e61b963aae0ebbbd2b (base main 381c140200d2f4ecb965183a2da1ba1e46d5d9c2).

Implemented on this draft branch:

  • 0 physical capture pins accepted by JUCE snapshot, DIRECT fingerprint, native silent ChannelId admission and .smix codec; no dummy microphone or second PCM renderer.
  • First-run and explicit Audio I/O accept genuinely output-only Windows backend, maintain exact output pin/epoch identity; GUI Apply/ACK/reopen preserve no capture source.
  • Verified same-endpoint WASAPI output-only recovery no longer rejects missing microphone.
  • Built-in EQ, Compressor and Harmony are no longer gated by an unrelated busy external VST3 scanner, and the approved native picker advertises these 3 actual compiled DSP processors from first startup, after vendor catalog refresh, never adds demo FX. Separate scanner still verifies third-party metadata.
  • Added C++ zero-IN PCM/track tests, Javascript contracts, compiled HTML Chromium test for 0-input Audio I/O and cold-start built-in FX; corrected stale vendor-only expectations rather than skipping tests.

Fast UI/native contracts PASS on this exact SHA. Previous exact SHA full Windows/MSVC CTest and Chromium success exist, but do not qualify latest SHA. New exact-SHA Windows qualification 42605359791 and full Chromium 42605389198 owner-triggered and currently in progress at last observation.

Release/merge fence remains: no-device offline logical editing (no playback endpoint), first-render reference UI while no native graph exists, third-party VST3 process isolation for unresponsive vendor code, physical ASIO/Player tests. Concurrent draft #486 native Player touches the same Main/LiveAudioEngine/LiveDryRoutingLab/session files: reconcile intentionally on integration, do not blindly merge. Keep #491 Draft, no installer claim until evidence.

…er on cold boot

The always-on ChannelId graph is real PCM, but on fresh startup must not make the saved Show unrecoverable. Permit replaceActive only for unmodified silent starter on exact verified device; do not overwrite authored FX/WIRE/track edits. Prove native 0-input saved Show restoration and regression safety.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci native 36a9bcc

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

P0 durable cold-start Show recovery — accepted implementation milestone

At exact SHA 36a9bccb7c1c9e90c2133d5574953eaf36f7694d:

  • Fixed real restart regression: boot now admits a genuine ChannelId SOURCE NONE graph before catalog readiness, which caused restoreSavedMixerSession() to reject every saved .smix as LIVE_MIXER_UNCHANGED. Restore now allows replacing ONLY canReplacePristineStartupMixer(): accepted ChannelId graph with the genuine unpatched starter source, no accepted FX, native control edits, BUS/WIRE structural edits or physical source, and exact verified device match. User edits are never overwritten by late catalog resolution.
  • Recovered previous checkpoint: a verified previous HEAD generation can also replace that untouched boot graph. usedPrevious still blocks automatic MAIN resume, preserving muted-safe recovery.
  • Real PCM regression: extended native_mixer_session_store_test.cpp::outputOnlyNativeStartup with a genuine 0-in/2-out initial ChannelId graph, native FX bank prepare and accepted replacement with a saved roster, then verifies that further native Add Track makes the graph nonreplaceable. One native publisher, no fake capture pin.
  • JS contract regression: asserts exact startup-only guarded restore path and proper wiring of replaceActive.

Fast native + UI PR checks: PASS at this SHA. Owner-authorized exact-SHA Windows native MSVC/CTest run 42606371346 was in progress at last inspection. Draft PR remains unmerged; no installer/FAT claims. Follow-ups #490 still include model editing with NO physical outputs, removal of fake reference first frame, bounded third-party VST scanning/prepare, real Windows hardware FAT and internal Player routing.

Show only the accepted native roster, open the requested backend directly under the existing output fence, and let the shipped UI test the verified MAIN pins without replacing the mixer. Report post-gate JUCE buffer peaks and source SHA for field diagnosis.
Keep normal device Apply anchored to saved physical MAIN pins. Accept deliberate assignment of exactly two driver-verified playback pins, and retry saved-session restore over a pristine startup graph after verified device recovery.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat c9563d1

Reconcile changed compact ordinals without losing the VB-Cable source, resume same-driver restarts with a bounded pin correction, and keep authored FX on compatible device epochs. Adopt a single verified physical speaker pair on explicit Apply when the old pair is gone.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat 6285f80

Recover playback independently, remember missing inputs by physical identity, and restore them automatically when the driver returns. Preserve compatible VST3 leases and persist pin inventories in existing mixer sessions.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat 5a923a7

masarray commented Nov 9, 2026 •

Copy link
Copy Markdown
Owner Author

ASIO4ALL field continuation — 6177068 (2026-11-09)

User reports playback silence/stuck after ASIO4ALL control-panel WDM changes, particularly with VB-Cable capture + Realtek speakers, despite earlier 5a923a7 candidate.

Source-level root causes corrected:

  1. The retained MAIN fallback matched only when exactly two ASIO4ALL playback pins were active. ASIO4ALL commonly exposes virtual outputs alongside the physical pair. The binding policy now finds one uniquely labeled physical stereo pair within any verified 2+ pin mask, maps physical names to actual JUCE callback ordinals (including 3/4), and refuses ambiguous or virtual-only targets.
  2. Explicit device Apply prematurely set asioRecoveryEpoch_ even before the native mixer rebind succeeded, suppressing the subsequent automatic retry. It now marks success only after mixer bind or verified rollback.
  3. Automatic restart now uses a bounded three-attempt retry with backoff per external driver epoch, maintaining its budget across the internal reopen and named-pin-correction epochs. Zero configured capture pins stay zero rather than enabling a mic on recovery. All driver operations remain on CONTROL, never realtime.

Regression coverage committed: native session binding tests for concurrent VB-Cable/Realtek outputs (MAIN at callback ordinals 2/3), source preservation and ambiguous two-speaker refusal; JS source-contract assertions for retry gating. Post-write source/contract presence checks PASS. Windows MSVC / CTest and physical Realtek listening are NOT YET RUN/PROVEN for this SHA. Earlier CI on 5a923a7 is not qualification of 6177068.

Field FAT after a verified new Windows build: select ASIO4ALL with VB-Cable capture + Realtek enabled; switch WDM pin availability while mixer is open; verify preserved ChannelId FX/WIRE and automatic PCM rebind (without manual remap), then play built-in MAIN test tone and capture Copy Audio Report during the tone. Verify both ASIO callback pin peaks and actual Realtek speaker audibility; no software-only fixture proves physical output. Repeated driver calls that never return and true no-playback offline editor still require separate architectural work. Keep PR draft and main unchanged.

Follow-up commit cbcc52d: serviceAutomaticMain() no longer records an ARM revision before the temporary native preflight passes; it retries CONTROL readiness rather than leaving MAIN muted indefinitely. audioDeviceAboutToStart() also captures an already-armed monitor intent before fail-muting the new ASIO generation, protecting the callback START-before-STOP ordering. Both behaviors have source-contract regression assertions. Candidate HEAD is cbcc52d; no exact-SHA Windows workflow or physical speaker test has yet been observed.

Follow-on recovery scheduler milestone — 2026-11-09

Production commit 0d2f1ba3 introduces the portable AsioRestartRecoveryGate, integrated into CONTROL serviceAsioMixerRestart. One 350 ms settling window per external device generation; only driver reopen/pin-correction consumes the maximum 3 vendor-transaction budget; internal JUCE-created epochs do not reset that budget; readiness checks retry at 750 ms without spending it; graph/FX rebind retries at exponential, capped 30 s delays. Automatic recovery is suppressed while the Audio Setup window is visible. Never use a new DSP callback, device default audio route, RT driver work or unverified speaker output.

Validation: local standalone g++ C++20 scheduling regression PASS. Exact-production Windows FAT requested via owner-accepted CI comment and run 42638912056, SHA 0d2f1ba; latest observed fast contracts PASS, MSVC still running. Separate test-only a5c737dd forces native scheduler assert checks to remain active with Release/NDEBUG; this test passed locally with -DNDEBUG. Runtime source is identical between the production FAT SHA and test-only HEAD.

Do not represent this as real ASIO4ALL/Realtek speaker proof. A driver call that never returns is still an in-process host boundary limitation, requiring process isolation for a hard hang guarantee. Keep PR draft; preserve main's separately merged Player work when reconciling later.

Follow-on milestone: missing JUCE device object recovery (2026-11-09)

01fcd097 fixes a nonrecovering path: the automatic ASIO4ALL service previously required getCurrentAudioDevice()!=nullptr before it could reach its own reopen branch. It now accepts a null JUCE device ONLY if the retained session is authored, the prior AUDIO SETUP remains exactly selected with ASIO backend, the saved playback/capture identity is unchanged, and the full registered ASIO driver class/CLSID still matches the previously pinned registration. All actual driver operations remain CONTROL-thread only, bounded by the previous retry gate. No new arbitrary endpoint and no silent fallback.

98330452 adds regression assertions to tests/mixer_io_operator_contract_test.mjs and runs focused ASIO restart contracts in the cheap Ubuntu quick gate before Windows MSVC. One exact-SHA FAT run 42649731559 was requested. The first fast native contract gate PASSED. The previous exact-source production FAT on 0d2f1ba PASSED 68/68 CTests, with ZIP and Inno installer artifacts (run 42638912056). This older FAT is not evidence that the new no-device branch compiles until 42649731559 completes.

Field acceptance sequence: ASIO4ALL + VB-Cable capture + enabled Realtek outputs, mixer ChannelId FX/WIRE audible to MAIN; unplug/disable/reorder WDM endpoints; confirm mixer document remains and no unapproved output; reopen the same ASIO4ALL, confirm CallbackCount increments, ASIO4ALL_DEVICE_MISSING is recorded for the null-device path, MAIN test tone routes to the verified physical stereo pair, and actual Realtek speaker is audible without manual re-MAP. Test an intentionally different ASIO driver: it must remain MUTED. Run repeated transitions to detect vendor hangs (currently not hard-isolated).

Coordination: Current PR is intentionally draft, behind main's separately merged ArSon Player update. Main changes overlap Main.cpp, LiveAudioEngine.h, CMake, roster and native PCM code; a blind auto-rebase would risk destroying active Player work. Reconcile with a tested merge before release. Do not claim full offline device-independent editing or physical Realtek FAT yet.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat 0d2f1ba

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat 9833045

…recovery #491

Integrate approved Player virtual sources, SOURCE ONLY and first-sound into retained native ChannelId/FX/WIRE engine without losing ASIO4ALL restart/readback recovery. Reconcile six overlapping files and preserve main's other player changes.
… restore

Preserve virtual MUSIC ChannelId, MAIN WIRE, and FX identity across all physical capture absence and return. Reject physical pin coordinates with zero input channels. Regression covers decode/encode, 0-to-2-to-0 ASIO recovery and malformed physical source.

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

/arsonmix-ci fat 3015960

masarray commented Nov 9, 2026

Copy link
Copy Markdown
Owner Author

Reconciliation checkpoint — ArSon Player + ASIO4ALL (2026-11-09)

Reconciled PR #491 with the current main commit d46aaa6b18ce94f5908c5db4f33ccba7d69947b4 using a true two-parent merge 681c4a16, preserving both histories and leaving main unchanged. Six intersecting native files were resolved explicitly: CMake, LiveAudioEngine, LiveDryRoutingLab, Main, NativeMixerSessionDocument and its tests. The other 13 ArSon Player files (PCM engine, worker decoder, approved GUI and native fixtures) came unchanged from main.

Integrated architectural contract: ArSon Player is a virtual DynamicSourceKind::playerDeck owned by the same native ChannelId PCM graph; it must never become an ASIO pin or enter the waitingAsioInputs queue. The ASIO4ALL hardware pin remapper now preserves PLAYER_DECK across 0-IN → 2-IN → 0-IN changes while waiting/rebinding physical microphones independently. This also preserves MAIN/WIRE/FX roster identity and SOURCE ONLY semantics.

Follow-up 3015960b fixes an additional cross-thread serialization bug: the old .smix decoder rejected any source at physicalInputs==0, including the legitimate device-independent music deck. It now admits virtual sources only, while retaining strict per-source rejection for physical input pins; native test covers roundtrip, malformed physical source and ASIO4ALL disappearance/return.

Repository evidence: PR now compares ahead, behind 0, and GitHub marks it mergeable (still DRAFT pending physical FAT). New exact-SHA UI and lightweight native contracts PASS on PR. Windows full MSVC, CTests, experimental installer/FAT package qualification requested exactly once for 3015960b via run 42652469196. Do not claim Windows build or speaker output PASS until confirmed from this run and the field machine respectively. No merge into main or release promotion has been done.

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant