Andreas Raab writes:
 > Of course, if I'd ever learn how to spell "timeout" correctly, I might 
 > find out that that's not the problem. Back to the drawing board... (but 
 > I'm still not convinced that the problem isn't somewhere in the nested 
 > unwind-protect stuff).

It definately looks like it's caused by the changes to Delay>>wait.

  wait
	"Schedule this Delay, then wait on its semaphore. The current 
         process will be suspended for the amount of time specified 
         when this Delay was created."

	self schedule.
	[delaySemaphore wait] ifCurtailed:[self unschedule].

If I comment out the "self unschedule" it works reliably. If I replace
the "self unschedule" with "Processor yield" the bug occurs but takes
a few more iterations (about 10 instead of 1 or two).

I'm going to leave this bug and go back to finishing up the next
Exupery release. I can make my test work by increasing the delay so
that the profilers never kill themselves.

Bryce