Repository navigation
Description
Reconciled 2026-11-08 — GUI/implementation scope completed. The approved GUI has direct cable select/replug/undo interaction contracts in tests/pages_wire_v49_direct_controls_smoke.py and current WIRE/ACK plumbing in merged #418. Close this GUI interaction milestone only. Real output/audio-path WIRE correctness and source-only behavior are NOT certified here and continue under #412/#389.
Current main at audit:
c57720ea25585c861af89802e1365ffef7d03d17. No statement of physical hardware FAT.
Problem / verified source evidence (2026-11-04)
Owner screenshot shows cables visually present but not directly selectable or detachable. Source inspection of site/preview-v2/index.html on current main:
.routing-wire-overlayand.routing-wire-overlay patharepointer-events:none; rendered SVG cables are non-interactive.wireTray()only renders editable.wire-gainwhenarmedWireSource()matches an edge. With no source armed, a connected destination renders.wire-pot.is-readonly, making direct gain adjustment impossible.- Pointer replug starts from a connected receiver only when
armedWireSource()matcheswireFor(source,destination); user must first select OUT. replugWireEdge(sourceId, oldDestId, nextDestId)already provides atomic replug/remove with cycle/duplicate guards: reuse its single transaction model.- PR GUI M3 — 70px aligned meters, clip lamps, safe multi-wire inspection and operator controls #343 adds a multi-incoming route inspector, but this does not implement selecting a visible cable by clicking the cable. Extend, don't undo it.
Outcome / interaction contract
- Visible cable is a selectable object. Click/tap an actual rendered cable (with an invisible generous ~10–14px stroke hit target) to select an exact stable
edge.id. Show a quiet selected-cable highlight and compact route details: FROM, TO, role (primary/aux/split), gain/tap, and explicit Unplug. Never select via channel-index guess; identical incoming/outgoing visual representations must map to the SAME edge. - Drag the cable from its receiving jack without arming source OUT. If only one incoming route is present, pointerdown on occupied IN begins a candidate detach gesture; >~10px intentional movement starts drag. Drop on another valid IN = replug same edge preserving id/mode/level; drop in blank patch bay (no snap) = unplug that exact edge; pointercancel/Escape/outside app/invalid target = rollback unchanged. No disconnect on merely clicking IN, hover, or tiny movement. Provide immediate Undo.
- Multi-incoming IN: click/hover a crowded receiving socket surfaces per-route picker (reuse GUI M3 — 70px aligned meters, clip lamps, safe multi-wire inspection and operator controls #343 inspector); user explicitly chooses which wire, then may drag/unplug. Never drop/remove all wires or silently pick the first wire.
- Knob works without 'select source OUT'. With exactly one connected wire into a destination, the existing wire-pot directly adjusts its edge gain. When multiple routes arrive, selected cable determines editable knob. If no route is selected, show a small route count/access to inspector rather than a misleading editable knob. Existing legacy SEND-managed routes are identified as such; no duplicate gain/audio processing.
- Preserve route safety: explicit direct MAIN vs group/parallel SEND semantics; primary OUT cannot silently become unpatched; reject loop, duplicate send, invalid destination, stale route. Native implementation must issue one intent and wait for accepted ACK/snapshot before claiming LIVE; browser Pages stays a GUI reference, not PCM.
- Focus/keyboard: selected route reachable via route inspector; Enter/Space to inspect, Escape cancels gesture, keyboard-accessible Unplug with confirmation/Undo; visible focus and textual route identification.
- Path hit zones must not block fader/pan/knob, background bank selection or adjacent socket controls. Pointer capture and keyboard/modifier gestures must remain consistent.
Implementation boundaries
- Reference GUI owner:
site/preview-v2/index.htmlcurrent edge model/gesture/render,site/preview-v2/assets/mixer-wire-dots.cssplus existing V48 socket owner only if required. Do NOT add parallel global wire models, stylesheet override layers, new audio routers, or brute-force rerender during drag. - Native GUI adapter: PR GUI #271 — Native adaptation of approved PR #278 V8 Mixer Console (draft) #271 / issue [V6 GUI OWNER] Rebuild single-window operator-first LIVE Console — native ChannelId/ACK only #268 owns mapping these same semantics to native
ChannelId, actual capabilities and ACK; ENGINE [ENGINE SOLE OWNER] One native callback → full-fidelity ChannelId cutover (draft) #266 and FAT [INTEGRATION/FAT v7 G0] Canonical JUCE WebView ACK/PCM evidence and exact-SHA release gate #270 own DSP and integrated qualification. - Preserve AGENTS.md and
docs/CHECKPOINT_GUI_M1_WIRE_V48_BUTTONS_V23_20261103.md; no wholesale merge of draft V17 — Magnetic OUT→IN Wire Patch Bay: stable controls, soft curves & snap (GUI reference) #278 into main.
Acceptance scenarios (automate Playwright)
- Unarmed source, one incoming: click actual cable -> exact route selected -> drag occupied IN to another IN -> same edge attributes; drag to blank -> only targeted edge removed; Undo restores.
- Unarmed source, 2+ incoming: choose route B -> adjust ONLY B -> unplug B; A untouched; duplicate visual rendering never causes two selection/commit actions.
- Empty IN click and near-miss must NOT delete anything. Out-of-window drop, Escape, pointercancel, duplicate/loop/invalid destination must preserve graph.
- Across bank, clipped/portal references, MAIN/group/FX, 3 density modes at 1366/1680/1920; no blocked fader/pan/knob wheel.
- Verify both screen geometry/visible selection and serialized state; no GUI 'LIVE audio' claims.
Ownership / dependency: Coordinate with PR #343 before touching its multi-wire inspector. Add corresponding native integration test to #271 only when native capability exists. Screenshots from owner feedback dated 2026-11-04. This issue is a task contract, not a claim that code is already fixed.
Activity
masarray commentedon Nov 3, 2026
Cross-thread owner update (2026-11-04): master execution / release gate is #351. Additional P0 visual-truth bug to cover here: owner observes CH01 and CH02 OUT/IN colored despite believing no wire is connected, unlike neutral dark-grey channels. Current
wireTray(t)marksis-sourcefrom armed OUT;syncVisibleWireDots(svg)independently paintsdata-wire-visible/data-wire-arrivalfrom rendered SVG paths. Neither is necessarily a reliable indication of a committed enabled edge if view/filter/selection changes. Require a canonical state matrix: (1) no committed enabled edges + unarmed/unhovered => neutral dark grey regardless of track number; (2) armed/hover/snap => distinctly temporary state; (3) committed enabled edge => connected state with inspector showing exact edge; (4) hidden/offbank route must not become false 'unconnected'; (5) after unplug + Undo, style and graph agree. Test CH01 and CH02 specifically by clearing demo seed graph and adding/removing routes; verify computed CSS + serialized model. Do not hide a real route just to turn LED grey. Keep actual route selection/drop behavior from this issue as the primary task.