* possibly add -n to dump the linenumber
* maybe add optimization useing seek() in file
* maybe add smarter memory management allocating the buffers only once
* work out if the arithmetic holds for large values
* automaticly (configure) select the best way to seed the random number
  generator
* add error checking of streams using ferror()
* maybe implement linked list contruction for arrays that might grow fast and
  need a lot of realloc()'s
* implement sanity checks for testing program
* improve docs
* implement internationalisation with GNU gettext
* do something with the NEWS file?
* maybe move source stuff to a src directory
* setup cvs server
* add option for other field delimiter (other than newline)
* try using temporary files with pointer into files
* add options:
  --dbm  use Berkeley DB for operation instead of using memory. This is useful when FILE is
         too large to fit into memory. Little cpu/memory usage but much slower
  -s, --skip-blank  skip blank lines


* options:
  --pick[=count]   pick at random count lines (lines can be picked more than once)
  --count[=count]  select count number of lines (lines will not be chosen more than once)