Repository navigation
refactor(mv6): epoch-fenced Channel Strip SOURCE queue and accepted native/M4 ACK projection - #782
Conversation
Deploying arsonmix with
|
| Latest commit: |
061a7f3
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://d78d2598.arsonmix.pages.dev |
| Branch Preview URL: | https://refactor-mv6-source-queue-na.arsonmix.pages.dev |
|
/arsonmix-ci ui 061a7f3 |
Codex Review Summary: ✅ CompletedThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
OWNER DELIVERY — MV-6 SOURCE Queue & Readback (2026-11-18)HASIL pengguna: Real SOURCE IN picker on the ONE approved /preview/ now rejects stale physical-device choices after a device epoch switch instead of rebasing them onto another interface. A queued channel selection can be cancelled by selecting its installed input; when that same channel already has an in-flight request, a compensating choice is preserved. An invalid native source queue head no longer strands independent channel edits. The legacy first-sound G0/M4 input route still uses its direct JUCE command (no Start/Patch/Arm wizard, no surprise auto-arm). Same one PCM/VST3 native graph; ACK never paints source as installed without later authoritative readback. Implementation: Pure STATUS: MERGED MAIN — PR #782 squash commit Bukti: Fast UI PASS, Fast Native PASS (213 native/source tests), exact-HEAD Full Chromium /preview/ 109/109 PASS run 44247970033 including Stage C pixel parity; Required Merge Gate SUCCESS after Ready on same SHA. Initial fast tests found a real 64-bit epoch overrestriction, repaired to preserve string uint64 for ChannelId and safe numeric only for M4; headless first-sound M4 rejection preserved without a new DOM toast. Neither guard test was disabled. Postmerge main UI run 44248991762 and native run 44248991761 started and were IN PROGRESS at handoff: do not claim PASS or Pages publication until checked. AMAN DIPAKAI? YA for merged/tested GUI source integrity; BELUM for newly built Windows installer or actual physical 2-vendor ASIO hardware FAT. Tindakan Anda: TIDAK PERLU MELAKUKAN APA-APA. Next: watch main-push CI evidence, then continue one bounded input-source/readback projection or reduce bridge device control coupling; no duplicated accepted PCM owner. |
MV-6 — SOURCE Queue & Readback reliability, not cosmetic split
Base:
main@7f3f4d8f16ccc9a1b4aa53c4f3185f7667f2b3e8(MV-5 #781 already merged). Distinct from IO-1 dual-device draft #771; no overlapping active GUI PR.Verified defect being fixed
Existing native
pendingNativeSourcesand M4pendingSourceQueuestored only the selected input pin. After a device restart/swap the UI could dispatch that old selection with a fresh epoch/revision, potentially pointing at a different physical pin on another interface. Also a subsequent selection of the already-installed source could leave a stale queued edit; an invalid native queue head could strand later channels.Real ownership extraction
web/approved-channel-strip-view-model.jsnow projects:sourceQueueSelection: validates exact choice from current DeviceVM pins, captures device epoch as an immutable draft, allows cancellation to installed PCM, but queues a compensating choice if that channel has an active change.sourceDispatchPlan: validates exact same physical epoch before dispatch, unique pins, native exact decimal-string uint64ChannelId/installed graph revisions vs M4's 0–9 channel + refreshed routing revision; cannot dispatch M4 when ChannelId mixer owns PCM.sourceAckResult: correlates requestId; M4 checks native accepted-route row and epoch/revision, native ChannelId only ACKs request, awaiting independent PCM readback.web/approved-native-bridge.jsremoves duplicated native/M4 source field validation and command assembly, but retains the single two existing queue Maps, nativearMixActiondispatcher, the 8s timeout, ACK subscriptions and installed PCM readback. Stale queued head is dropped safely and does not starve other channels. No browser-local audio owner or timer-based polling.tests/approved_native_m2_contract.test.cjs: deterministic positive/negative tests for 64-bit IDs, epoch collision/swap, invalid pins, duplicate labels, M4 stale identity and owner upgrade, in-flight revert, request correlation, M4 delayed ACK, preservation of one dispatcher; legacy source-location assertions updated to the actual new ViewModel boundary.Invariants and risk
ONE approved
/preview/GUI; ONE ChannelId/EdgeId/VST3 native render graph; source-only changes NEVER auto-arm hardware or silently patch a new physical device. No changes to JUCE/C++/ASIO/DSP, CSS/HTML/PNG, external FX, .smix schema, plugin editor or device driver. Source queue is UI transient only. Installed PCM remains authoritative.Status: KODE DIBUAT / DRAFT. Fast UI/native and one exact-head Chromium qualification pending before merge; Windows installer and physical dual-ASIO FAT NOT RUN. Tindakan Anda: TIDAK PERLU MELAKUKAN APA-APA.