Repository navigation
refactor(g-mv22): hardware-scoped SOURCE queue leases for ChannelId and M4 - #804
Conversation
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. |
Deploying arsonmix with
|
| Latest commit: |
20be733
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://97f9890e.arsonmix.pages.dev |
| Branch Preview URL: | https://refactor-g-mv22-physical-sou.arsonmix.pages.dev |
|
/arsonmix-ci ui 20be733 |
G-MV22 — Indonesian owner completion / continuity (19 Nov 2026 WIB)(1) Hasil untuk pengguna: Pilihan SOURCE/INPUT lama sekarang tidak akan diteruskan ke native ketika mask input/output, sample rate, atau buffer hardware berubah pada epoch yang sama. Native ChannelId maupun M4 harus memegang fingerprint fisik yang sama dari saat pilihan diantrekan, dikirim, hingga ACK + independent PCM readback. Urutan enumerasi pin/monitor tetap tidak menyebabkan false remap. Pemilihan source tetap latest-choice-wins dan antrean channel lain tidak tertahan oleh entri kedaluwarsa. (2) Status sebenarnya: MERGED MAIN melalui PR #804, squash SHA (3) Aman dipakai? YA untuk source code hasil merge dan perilaku GUI yang diverifikasi oleh CI. BELUM berarti installer Windows baru tersedia, hardware ASIO fisik telah diuji, atau multi-vendor capture telah dikualifikasi. Tidak ada perubahan DSP, VST3, driver, .smix, CSS/layout atau speaker auto-ARM. (4) Tindakan Anda: TIDAK PERLU MELAKUKAN APA-APA. (5) Bukti exact SHA:
(6) Langkah berikutnya: G-MV23 should audit lifecycle of SOURCE timeout/reconnect and queued-gesture cancel/retry without introducing any second state owner; prefer further pure ViewModel policy and remove redundant bridge responsibility proven by tests. Preserve the one approved |
G-MV22 — Same-epoch SOURCE physical-device safety + modular state ownership
Problem audited on merged G-MV21 main: SOURCE queues were scoped only by
deviceEpoch. A changed physical input/output mask, sample rate, or buffer size may retain the same epoch, permitting an old queued GUI source intent (or ACKed source readback settlement) to outlive its physical hardware mapping. ChannelId/M4 native source request lease construction was duplicated in the bridge, which made it easy to omit physical identity from one path.Implementation
snapshot.hardwareConfigurationtoken. Never recalculate hardware/pins in Channel Strip VM.sourceDispatchPlan()drops the old queued source (including on unchanged epochs).sourceQueueRequestLease()in the existing Channel Strip ViewModel. It validates actual ChannelId/revision/roster or M4 lane/revision before a single bridge transport emits, and freezes per-request identity/choice.sourceQueueReadback()now refuse to confirm an old ACKed source against the wrong canonical hardware token even if epoch, source channel, and pin numbers happen to match.UNADMITTED → M4promotion when the hardware is unchanged, FIFO latest-choice-wins, unrelated channel progress, strict ACK+independent readback, physical output authorization, one native DSP/PCM renderer and ONE approved/preview/.Compat: Existing pure test fixture callers that omit a physical token remain usable for isolated historical helpers; the actual GUI always supplies
hardware.hardwareConfigurationand refuses to emit if missing. No driver, DSP, .smix, CSS or layout changes. CI fast UI/native then one exact-SHA Full Chromium and Required Merge Gate; only merge if clean. Windows installer and physical multi-ASIO FAT not claimed.