children CPU times

[Henry Cejtin]

Ah, I was not quite right about the child times: the child times includes any
grand children times.  This explains why it  includes  children  of  the  gcc
(like the assembler and the loader).

Things  are  a bit complicated by the following strange detail of Unix: every
process has a parent process, which is the process that will get  your  time,
on  your exit, as part of its children's times.  (It is also the process that
can wait for you to exit and get your status.)  If your  parent  exits  while
you  are  still  running,  then  your  parent  process  gets set to process 1
(typically init).  This init process is always waiting for things to  finish,
and  when  it  gets  one  it  knows  nothing  about  (not a child, but a long
descended grand* child), it just ignores it.

The effect here is that if I have a child who has a  child,  and  that  grand
child  of  mine  does a lot of expensive computation, and if that grand child
outlives my direct child, then all of its running time, when it  exits,  will
be added to process 1's child times, not mine.

The  reason  for  the strangeness is that if your parent exits before you do,
then no one would be waiting for your exit status.  To make things cleaner in
the  kernel,  you  are  assigned to some one who will always accept your exit
status (process 1).  It also means that the total time of process 1 (and  its
children)  is the total amount of CPU time used on the machine by all process
that have exited since boot up.