>  This appears to work, but now I'm getting increasingly unpredictable
>  behavior.  Csound terminates at a certain part of my score for no
>  obvious reason, I'm still working out what to post about it.  When I
>  play the .csd file from another Pd test patch, it terminates at
>  different places, or it makes no sound, or it plays fine; I can't
>  determine what is changing to make it behave differently each time.
>  I'll post some info when I can zero in on it.

Here's another .csd file that maxes CPU almost immediately.  I thought
I solved the problem by giving all the notes decimal numbers that
increase according to noteoff times and using mode 5 for turnoff2;
note that the arguments for p4 of i10 increase throughout the score,
giving the illusion that it's always turning off the "oldest" note.
The reason I did it this way was simply to control tempo in realtime;
Csound doesn't warp incoming realtime events, and if I speed up after
sending a note, the duration needs to change accordingly.  Pure Data's
qlist object handles tempo change well, so I'm using noteoff messages
instead of durations.
At first it seemed to work, but alas, I think I'm back to breakage.  I
get the same behavior using this orc/sco setup as with csoundapi~ in
Pd, so it has to be Csound.

Thanks for looking.
-Chuckk