Repository navigation
Conversation
|
/arsonmix-ci ui 14a17bd |
CI efficiency workstream instruction — owner request, independent audit 18 Nov 2026Goal: speed up practical MVVM development while preserving EXACT-HEAD reliability and release quality, not merely delete tests. This is nonblocking guidance to the PR #768 working thread. PR #767 was already MERGED; audit current main and reconcile #768 against that source before any further work. DO NOT reimplement the overlapping Device Setup draft/selector work or silently merge both variants. Measured GitHub evidence: normal fast UI PR run #44196523503 completed in ~34 seconds and deliberately SKIPPED the full Chromium job; exact-SHA owner-dispatched full UI #44189797846 completed in ~8m45s and was SUCCESS. Thus the repository already avoids running full Chromium on every PR commit. There is no evidence it is necessary to remove essential coverage or replace CI framework. Task split: Complete/reconcile the MV-2 functional PR first. If optimizing CI, use a separate, narrow CI-only PR, after an explicit audit of Safe optimization candidates:
Acceptance: CI-only PR with measured wall/runner-time improvement without reduced scope; PR source tests for changed-files matrix (GUI JS, bridge, CSS/asset, native C++, docs, release, missing/pending/failing runs); tested exact-SHA workflow paths; report what was NOT run. If measurement shows fast lane is already sufficiently efficient, DO NOT make cosmetic YAML refactors—return to shipping real GUI modularization. Operator report: HASIL, STATUS, AMAN DIPAKAI?, TINDAKAN ANDA, BUKTI PR/SHA/CI before/after, BERIKUTNYA. No new audio installer is implied by workflow changes. |
MV-2 reconciliation: SUPERSEDED — DO NOT MERGE THIS PROn final verification, PR #767 independently merged to main as Exact alternative HEAD: Reviewed unique ideas for follow-up from #768, excluding duplicated drafts/selectors:
Any needed corrections will be forward-ported as minimal edits against merged |
Why (real feature extension, not file shuffle)
After merged #765, native Device Setup still changed backend/capture/playback/rate/buffer/physical-pin draft directly inside DOM callbacks, and its Apply-enabled rules were duplicated in rendering. On combined non-separate JUCE devices, selecting another output left capture on the previous endpoint, preventing Apply.
Delivered
setupInitialDraft,setupEditDraft,setupUiModelown ephemeral editable selection and UI affordances.setupBeginreuses the samesetupValidSelectionpredicate used by the Apply button (one rule)./preview/bridge only translates real DOM events to pure model operations and paints their results; retains singlearMixActiondispatcher,NativeDeviceApplyCoordinator, ACK, v6 readback, and G0/M4/ChannelId PCM distinctions.Measurable baseline and delta
web/approved-native-bridge.js: 4228 → 4184 lines (-44); removed inline draft mutator and duplicate option/Apply-state decisions.web/approved-device-view-model.js: 225 → 338 lines (+113); explicit pure unit-test seam (without JUCE/ASIO).Required evidence before merge
/preview/(oneuidispatch after fast CI).