Latching scene B plays a voice on Scene A

[Andreya-Autumn]

Scene A has a sine osc and is monophonic
Scene B has an audio in osc from Scene A
Latching scene B should be silent, but here, it latches both scenes for some reason.

[Andreya-Autumn]

Simpler repro:
Load a fresh Surge.
Set it to Dual scene mode.
In one of the scenes, turn latch on and off again.

If I latch scene A, the voice counter reads 2. Turning latch off again it does down to 1, and indeed it still makes sound. Each time I turn latch on/off, another extra voice lingers. Switching off dual doesn't end them, but switching to scene B and turning latch on/off there does.

If I latch scene B, the behavior is the same, except it doesn't stack extra voices. Each time you latch on it goes to 2, then back to 1. Switching off dual does stop them from sounding.

[baconpaul]

tagging into 1.3

i bet the latch-end-voice-reaper isn't properly scene aware

[Andreya-Autumn]

An afternoon of casually building Surge from different commits I think I've found it. It appears at #6606, and only happens with the added MIDI channel-to-Scene option not engaged. I guess something about the scene channel mask code is a bit borky...

[baconpaul]

My guess is this state

surge/src/common/SurgeSynthesizer.cpp

Line 4023 in 3c4c9fc

if (release_if_latched[0])

is getting set improperly in that case (that is play a or b is wrong) and I would start with a print there

[Andreya-Autumn]

I tried looking at it, but I'm in a bit over my head as you might expect. 😅 I messed around with XCode breakpoints a little, and I think you're right PlayB stays true in some cases where it shouldn't, but discerning why is where I'm lost. Wish I could be of more help. 🙃

[baconpaul]

We have confirmed the diff that breaks it and the line that’s wrong. That’s 90% of the fix. I’m sure this one will now be easy to do the last 10%. Awesome bug hunting!

[baconpaul]

Oh also do you know about git bisect?

[Andreya-Autumn]

No! Lemme guess, it helps find a breaking commit? I should read up on that probably. :)

[baconpaul]

Right. The problem is entirely with the ch23 turn off thing. That basically breaks latch.

It breaks latch for an annoying reason though.

surge/src/common/SurgeSynthesizer.cpp

Line 4094 in 8d791a1

if (playA && (storage.getPatch().scene[0].polymode.val.i == pm_latch) && voices[0].empty())

there's the code which plays in latch mode. As you can see it uses 'channel 1' to mean scene A and 'channel 2' to mean scene b.

So it turns out this is also broken in 1.2.3 in MPE mode, where latch does the exact same thing.

The answer here is to add an argument to playNote for 'ignore scene' which ands onto the channel mask then pass it form that latch initiation point.

[baconpaul]

Ahh great and i can grab it in a regtest now. Will have this soon.

[baconpaul]

yup got it.