- setuid/setgid must effect process
  + test syscalls (getuid) afterwards
  + test core file content

  + use UID/GID in access(2), chmod(2), chown(2), link(2)

- pgrp & session are process properties

- nice level is process property

- getppid(() must return process ID of parent process of the thread
  group leader
  + test core file content (psinfo, prstatus)
    The question is: do we really want this?  The "wrong" information
    actually provides some information one cannot get easily otherwise.

- rlimit should be process-wide and SIGXCPU should be sent if all threads
  together exceed the limit

- getrusage() must return resource utilization for the process

- sigaltstack must be per-thread (???)

   I've an interpretation request outstanding

- the scheduler must be thread group-aware, i.e., it has to give time to
  the thread group not proportional to the number of threads.