Thanks for the thorough reply. I'm doing one safer, even than "voice stealing": each Spork is, at most, reconfiguring a single global oscillator.

Attached is the entire source. Lines 128-142 contain initialization code that creates a Shred to make an LED blink by sending a MIDI message within a time loop, and another Shred that sets the frequency of an LFO, polls it every ten milliseconds, and sends some Serial output based on the LFO position.

Anytime that the parameters controlling the blinking rate and the LFO rate/phase have occasion to change (when MIDI events come in to change them), the "old" Shreds are Machine.Remove'd and they're recreated immediately following. You can see this at lines 343 and 589.

Notice that the LFOMod function and the Blink function aren't creating anything new (with the exception of local arguments being instantiated), but if you think those local arguments could be causing my problem I'll eliminate them too.

Please see the attached.

rs

On Mon, Jun 23, 2014 at 11:59 AM, Perry R Cook <prc@cs.princeton.edu> wrote:

I have lots of programs that spork hundreds or thousands
of shreds without this error.  So, without seeing your
code, my guesses (and questions) are as follows:

Are you running this on a small memory architecture? Since
you say headless, I'm guessing maybe Raspberry Pi?  It
could be that you're running out of memory, so see next
question.

Does the shred that you're sporking declare new UGs or
require memory to be allocated (arrays, lots of string
manipulation, etc.)?  In general, this is a bad idea if you
can avoid it.  ChucK does no garbage collection, which
means that you need to be cautious about declaring memory.

So even a shred as simple as:

fun void mySine()  {
    SinOsc s => dac;
    Math.random2f(100,1000) => s.freq;
    second => now;
    s =< dac;
}

will eat up memory, because even though the SinOsc
is unchucked and never computes again, the memory
for that structure is still around.  We want, someday
to make ChucK a proper garbage collecting language,
but it's hard, especially for real-time systems.

If this turns out to be your issue, then one way to
handle this is to make a global pool of fixed
resources and have your shred grab from that.
Like classic "round robin voice stealing" in
synthesizers:

SinOsc s[100];
0 => int next2Use;

fun void mySine() {
    next2Use => int thisOne;
    1 +=> next2Use;
    if (next2Use > 99) 0 => next2Use;
    s[thisOne] => dac;
    Math.random2f(100,1000) => s[thisOne].freq;
    second => now;
    s[thisOne] =< dac;
}

// to test:
while (1) {
    Math.random2f(0.01,0.1) :: second => now;
    spork ~ mySine();
}

I just fired up a few dozen of these running in
parallel, VM showed 2600 total shreds running,
no problem (and no clicks or dropouts!!).

Hope this helps, if none of this applies, then you
may have discovered a strange bug.  Source code please,
and we can all scratch our heads on it.

 Thanx!!  PRC


Today's Topics:

   1. 147th shred removed...segmentation fault? (Ryan Supak)
_______________________________________________
chuck-users mailing list
chuck-users@lists.cs.princeton.edu
https://lists.cs.princeton.edu/mailman/listinfo/chuck-users