v2.15.0 — MIDI + OSC backends · cue-list and stage-layout persistence

A big batch: the backend halves of four new control surfaces (MIDI, OSC, cue-list preview/waveform, stage-layout persistence), a round of bugs found in post-merge review of the tap-tempo feature, and an ops fix for a real production incident this release itself walked into.

🎛️ MIDI + OSC backends (PR #100, #99)

Two new inbound control paths, both scoped deliberately narrow: the listener/dispatch logic only, reusing the exact same lighting functions the web UI and MCP already call (set_channel_values, scene activation, chase start/stop, /blackout) — no new lighting logic, just new triggers into existing code.

• MIDI — python-rtmidi listener thread with device auto-discovery and reconnect-on-hot-plug, CC/note → action mapping persisted to midi_mappings.json, GET/POST /api/midi/mappings, GET /api/midi/devices, GET /api/midi/state.
• OSC — UDP listener (default port 8000) routing /scene/<name>, /chase/<name>, /fixture/<id>/<channel>, /master, /blackout, /cue/go|pause|stop, plus outbound /state/* feedback on changes.

Both degrade gracefully — the server starts fine with no MIDI hardware or python-osc/python-rtmidi installed, same pattern as the existing audio engine's available flag.

Deliberately not in this release: the "MIDI" web UI tab, drag-to-bind UX, TouchOSC layout file, and verification against real hardware — those need hands-on review with an actual controller/iPad, not something CI can sign off on. Tracked as follow-ups on the original feature issues.

🎚️ Cue-list preview/waveform + stage-layout persistence (PR #102, #101)

Two more backend-only slices, each scoped after checking what already existed:

• Cue lists — POST /api/cue_lists/<id>/preview (apply whichever cue would be active at a given timestamp, without touching actual playback state — verified side-effect-free against a running cue list) and GET /api/cue_lists/<id>/waveform (peak/RMS data for an associated audio file, display-only). Re-timing a cue turned out to already work via the existing PATCH endpoint, so no new work was needed there.
• Stage layout — GET/POST /api/stage_layout persists fixture {x, y} positions to stage_layout.json. Group-creation-from-a-fixture-list already existed too (POST /api/groups), so this is purely position storage.

Same split as MIDI/OSC: the actual timeline/canvas UI is a visual/UX feature pytest can't judge and stays a human-reviewed follow-up.

📡 Wireless DMX transmitter channel, cross-referenced against a live scan (PR #89)

Follow-up to the RF Environment scan from the last release. QLC+ has no visibility into a wireless DMX transmitter's RF channel — that's a hardware-only setting on the transmitter's own display. This adds a small form where the operator reports what their transmitter (D-Fi Hub or similar) is set to, persisted server-side, and the RF scan now cross-references it against the live WiFi survey for a concrete suggestion ("Transmitter channel 9 is sitting near loud WiFi traffic") instead of just a generic quiet-window readout. Verified against Chauvet's actual published manuals rather than guessed.

🐛 Fixes

• Tempo API edge cases (PR #97) — bpm: NaN/Infinity was slipping past range validation and hitting a 500 instead of a 400; starting a tap chase with zero playable steps reported false success while silently killing any runner already in progress. Both fixed, plus a parametrized regression suite.
• Deploy pipeline never installed new Python dependencies (PR #103) — the first deploy after this batch's structlog/python-osc/python-rtmidi additions landed crash-looped lighting-control.service on the live Pi (ModuleNotFoundError: structlog). deploy.sh synced the code but never ran pip install. Fixed live via SSH, then fixed for real: deploy.sh now runs pip install -r requirements.txt against the venv on every deploy (fast/idempotent when nothing changed) so this surfaces as ordinary install output instead of a silent crash loop next time.

✅ Tests

730 unit tests + 73 integration tests, all green (plus 3 backup-script tests that need bash 4+ and only run in CI's Linux environment, not locally on macOS's stock bash). Every PR in this batch was independently re-verified before merge — diffs read by hand, not just CI trusted — including a cross-PR check that caught a real breakage: an already-merged test called _mock_chase_run() with a signature another PR had changed, which git's merge never would have flagged since the two changes never touched the same lines.

Upgrade

From your Mac:

git pull origin main
bash scripts/deploy.sh

deploy.sh now installs any new/changed Python dependencies automatically — no manual pip install step needed even though this release adds three new packages. Hard-refresh the browser (⌘⇧R / Ctrl⇧R) after deploying.

What's next

• PR #60 (multi-workspace switching) is still held, not part of this release — it collides with the boot-restore/last-look system (switching workspaces restarts QLC+, which the crash-restart-restore logic reacts to by replaying the previous workspace's last DMX snapshot onto the new one). Needs that fixed before it's safe to merge.
• The MIDI/OSC web UI tabs, TouchOSC layout, and the cue-list/stage-layout canvas editors are all queued as separate, human-reviewed follow-ups to the backend work in this release.
• Issue #47 (gunicorn/eventlet WSGI migration) and #25's remaining nice-to-haves are still open in the backlog.