Lost time while profiling

[François Pinard]

Hi, people.

This Python program uses almost 100% of the CPU for a few minutes on a
small input sample.  It will surely consume many hours on full input.
While running it under the Python 2.2.1 profiler, the profiler reports a
total number of CPU seconds which account for about 40% of the wall-clock
time.  I wonder where the remaining time has gone!  Did others observe
similar discrepancies while profiling?  Why is the loss so spectacular? :-)

P.S. - This happens with no profiler bias, that is, zero per the
documentation.  First calibrating the profiler makes the results even
more surprising, as it then accounts for maybe 15% of the elapsed time.

-- 
François Pinard   http://www.iro.umontreal.ca/~pinard