profiling bug understood

[Stephen Weeks]

> As to doing it in ML, I thought about it, but I  don't  think  it  can  quite
> work.   The  problem  is  that  ML  signal handlers are only called at `safe'
> points, which isn't right for profiling stuff.  If you go  to  C  code  which
> grinds  for  a  while,  then  during that time no ML handler will be invoked.
> This can easily take more than .01 CPU seconds,  and  that  would  mean  tick
> lost.  I don't see any solution to that.  A bit sad.

Agreed.  Too bad.

It doesn't work anyways because of the safe points.  This skews where you get
the ticks.