Re: [fluid-dev] Crash caused by hot-plugging audio

[josh]

Hello Ebrahim,

Quoting Ebrahim Mayat <address@hidden>:
> Hello all
>
> I happened to stumble upon a way to make the new fluidsynth crash.
> This is a minor bug which I can live with but I'll report it anyway.
>
> Fluidsynth can be launched and executed to obtain audio output from my
> computer's speakers. But when I try to patch the computer's audio
> output socket with an audio mixer's input socket, I get the following
> error:
>
> > >
> > ** (process:532): CRITICAL **: fluid_synth_alloc_voice: assertion    
> > `fluid_synth_is_synth_thread (synth)' failed
>
> Secondly, if the audio ports are patched beforehand and fluidsynth is
> run, there is no problem injecting MIDI-triggered audio into the mixer
> but if the plug is then pulled from the computer's audio out, the same
> error occurs.
>
> Thirdly, there is no problem with audio hot-plugging if I run qsynth
> linked to fluidsynth-1.0.9.
>
> (All this was done on OS X by the way.)
>
> E


Uhh oh.  This sounds like the audio driver starts to get called by  
another thread, than the original audio thread.  I was afraid that  
might happen and I'm not exactly certain how to resolve it with the  
current architecture.  Many FluidSynth functions attempt to determine  
whether they are being called from synthesis thread context or not.   
Right now the method of detecting this, is to assume the first thread  
to call the audio synthesis routines is the audio synthesis thread.   
If another thread gets used afterwards, then that is a problem.  I'll  
have to think about this one a bit.  I'm curious why it is happening,  
but I can see that it may also possibly occur with Jack as well, since  
FluidSynth doesn't actually create or control the audio thread in that  
case.

For that particular function (fluid_synth_alloc_voice) I could remove  
the assertion that it is the synthesis thread, though that could also  
mask potential threading issues in an application using FluidSynth.  A  
check could also be done in the synthesis functions to see if the  
calling thread has changed.

If anyone has any better ideas of how to handle this with the current  
FluidSynth API, I'm all ears!

Cheers,
Josh