Issues starting MIDI playback, ALSA w/ hardware MIDI synths

[m0rb]

Hi Devs!

I've been encountering playback issues on multiple engines that use this library (namely GZDoom and Raze); every once in a while, when a problematic MIDI track loads, it'll start flooding the console with a decrementing sequencer undderrun and an error number (error -22);

ex:

Alsa sequencer underrun: -1067117 ticks!
Alsa sequencer did not accept event: error -22!
Alsa sequencer underrun: -1067194 ticks!
Alsa sequencer did not accept event: error -22!
Alsa sequencer underrun: -1067271 ticks!
Alsa sequencer did not accept event: error -22!

Switching to any given Software MIDI library plays the tracks without issue.

I'm currently fully synced with origin/master (commit 50ad730, 1.1.13); GZDoom and Raze are also relatively as fresh. Host system is Fedora 40 x86_64.

I've got a SoundCanvas SC-8850 that works as intended with just about anything I throw at it. This also happens with a cheap USB-MIDI adapter hooked up to a MT-80s.

Thus far it seems like all music tracks in DOOM PWADs (DOOM, DOOM2) do not trigger this issue. MAP01 on sunlust.wad reliably fails.

Thanks for your time!

[witchymary]

Can confirm this as well. I'm using nuked-sc55 instead, but the same happens with some selected tracks. If I play the midi through aplaymidi it works without issue.

[ZDHellser]

Also dealing with this issue. I'm using Manjaro (Zetar 25.0.2) with Pipewire. All music plays fine through QMidiPlayer including GS midis. But ZMusic fails with Sunlust (GZDoom) and E1L2 (among others) for Duke Nukem 3D (RAZE). I have a Roland UM-ONE II with a SC-88ST.

Updated 5-22-25: Other source ports don't have this problem. Tested with Woof! and Chocolate Doom.

Update 5-28-25: I just remembered I still have the Windows version of GZDoom. I can ensure that the MIDI handling itself is fine as GZDoom plays songs just fine under Wine - and if Wine is handling MIDI differently, good for them. But ZMusic is doing something different with ALSA directly that ALSA just does not like.

[RoqueDeicide]

I've had a look at AlsaMIDIDevice::PumpEvents() function code, and it appears, that, when the snd_seq_event_output returns an error code, the function doesn't increment buffer_ticks or Position values, which causes it to keep trying to shove the same event, that's not getting accepted by the device, over and over and over again, and the song playback gets stuck. Either the error handling needs to be moved to the end of the while loop, or removed entirely.

I'm on Windows, so I can't really fix this myself.

[coelckers]

If this is caused by an error return it would be interesting to see the actual error that started the entire mess.
And what does error -22 mean anyway? The Alsa docs are typical Doxygen shit that is as vague as it may come because nobody ever bothered to write meaningful explanations.
This will have to be fixed by someone running Linux

[Cacodemon345]

This should be fixed now as of commit 0fb7f18.

[m0rb]

Time sure has flew by!

I'm unable to reproduce on GZDoom g4.15pre-643-g5e9e2302f (latest commit to repo) built against a fresh checkout of ZMusic. ALSA version is 1.2.13.

Starting, stopping, and changing tracks was flawless on the SC-8850. Used SUNLUST.WAD as a testcase again and nothing logged to console.

I'll leave it to the maintainer's discretion to close out the issue. Thanks for everything!