Testing MIDI hotplug
ShowMidi reopens a connection when its device comes back (see the Hotplug
notes in Plugins). Two scripts test that on
Windows: one drives a virtual device unattended, the other watches a real one
while you plug it by hand. Both start their own sr start on a scratch show,
on port 18010 by default, and send CC and panic messages only.
Unattended, with a virtual device
This needs loopMIDI installed; the script uses its driver to create and close a port.
uv sync --extra midi
uv run python scripts/midi/hotplug.py --interval 2.0 --phases cycles,flaps,fast
| Phase | What it does | Passes when |
|---|---|---|
cycles |
Unplugs and replugs, waiting for each status change. | Every replug reconnects, a CC into the input comes back out as a panic, and the server stays up. |
flaps |
Random up and down spells, many shorter than the interval. Then one steady plug. | The server survives and the steady plug works end to end. |
fast |
Unplugs and replugs inside one interval. | The stale connection is noticed. It currently fails; see below. |
The script prints one JSON line per step and exits 1 if a phase failed.
By hand, with a real device
Pass the device's port name without its trailing number. The script prints a line when a connection opens or closes, when any key or knob reaches the input, when rtmidi reports reopening a half-closed input, and when the server exits or stops answering.
- Press a key, and check that it arrives.
- Unplug, wait one interval, replug, wait, and press a key. Repeat a few times.
- Unplug and replug as fast as you can, then press a key.
Results, 2026-10-11
Windows 11 (10.0.26200), Python 3.12.10, python-rtmidi 1.5.8 (RtMidi 5.0.0, WinMM).
At f8ae176, before the fix:
| Device | Interval | Result |
|---|---|---|
| Virtual | 0.5 s | Cycle 1 passes. Cycle 2 reconnects with its input dead, and the next unplug crashes the process with 0xC0000374. |
| Launchkey 49 | 2 s | The first replug logs MidiInWinMM::openPort: a valid connection already exists! and no input arrives. The next unplug hangs the server: the process stays alive but stops answering HTTP. |
With the fix that retires a client whose close fails:
| Device | Interval | Result |
|---|---|---|
| Virtual | 0.5 s | 5 of 5 cycles pass. 40 flaps pass, recovering in 0.48 s. |
| Launchkey 49 | 2 s | 3 of 3 cycles pass, with input after each. 2 of 2 quick wiggles are detected. No crash and no hang. |
On both builds: the fast phase at 2 s fails 5 of 5 times. Status reads
connected while input and output are dead, until the next unplug that lasts
longer than an interval. The Launchkey's USB re-enumeration outlasted the
interval, so this did not occur by hand.