Sequencer event ordering is broken for NoteOn vel=0 events

[jet2jet]

FluidSynth version

(commit hash: 17fb48d )

$ ./fluidsynth --version
FluidSynth runtime version 2.2.1
Copyright (C) 2000-2021 Peter Hanappe and others.
Distributed under the LGPL license.
SoundFont(R) is a registered trademark of Creative Technology Ltd.

FluidSynth executable version 2.2.1
Sample type=double

Describe the bug

With using sequencer (fluid_sequencer_t), when sending 'Note on' event immediately after (i.e. same tick of) 'Note on with velocity = 0' event, the first 'Note on' event may sound nothing.

Expected behavior

'Note on' should always sound in that case

Steps to reproduce

I attached reproduce source code: main.cpp.zip
(I used VintageDreamsWaves-v2.sf2 Soundfont on reproduce.)

Extract main.cpp, build with gcc -o main main.cpp libfluidsynth.a -lglib-2.0 -lstdc++ -lm -fopenmp, and execute ./main.
The WAVE file out.wav is generated and it may play something weird sound.

Additional context

• If 'Note off' event is used instead of 'Note on with velocity = 0', it seems OK.
• I wonder that this problem occurs from 0d98c47 (not tested well, but v2.1.x had no problem)

[derselbst]

Since 2.2.0 events that share the same tick are ordered in a special way, see docs of fluid_sequencer_send_at(). This ordering currently considers NoteOn Vel=0 events as real NoteOns by mistake. NoteOns with velocity=0 are always treated as noteoff by the synth.

There is more than one way to fix this, I'll need to think about it.

[derselbst]

I think the best way would be to make fluid_event_noteon() return a NOTEOFF event when vel==0. I'll add a regression test for this as well.

[derselbst]

Implemented in #908, feel free to test.

[jet2jet]

Thanks for the fix.
I tested and confirmed that the issue was fixed.

Since 2.2.0 events that share the same tick are ordered in a special way, see docs of fluid_sequencer_send_at().

I understood that the order of events that have the same timestamp is undefined, but how do you think about 'on-off' effects such as Hold? I wonder that, there may be the case that Hold-off event is followed by Hold-on event with the same timestamp, which assumes it resets sounded notes and turn hold on again at the same time. (If it is a rare or not so good case, the current behavior would be fine.)

[derselbst]

By "hold-event" you mean regular Sustain and Sostenuto events? If yes: Just like other CC events, if you want them to be applied in a special order to achieve a special effect, you would need to use different ticks.

[jet2jet]

By "hold-event" you mean regular Sustain and Sostenuto events?

Yes.

Just like other CC events, if you want them to be applied in a special order to achieve a special effect, you would need to use different ticks.

I understood. Thanks for the reply.