From: Thomas Hafner <thomas@...>
Date: 2007-04-14T08:55:08+09:00
Subject: chaining processes, Process.waitpid

Hello,

for chaining processes in a way, that the output of one process feeds
the input of another one like in that example ...

#\\
# transformation 'foo' -> 'fee' -> 'tee':
chain_processes(lambda { exec('echo', 'foo') },
                lambda { exec('sed', 's|o|e|g') }) do
  system('sed', 's|f|t|g')
end
#//

... I came to that solution:

#\\
def chain_processes(*processes, &block)
  saved_stdin = $stdin.clone
  threads = []
  this, *remaining = processes + (block ? [ block ] : [])
  while [] != remaining
    rd, wr = IO.pipe
    child = fork
    if !child
      rd.close
      $stdout.reopen(wr)
      exit this[]
    end
    wr.close
    $stdin.reopen(rd)
    threads << Thread.new { Process.waitpid(child) }
    this, *remaining = remaining
  end
  
  this[]
  threads.each { |t| t.join }
  $stdin.reopen(saved_stdin)
end
#//

That seems also to work without Process.waitpid(child). When is it
really necessary to wait for the children? How should an example look
like which breaks without using Process.waitpid or Process.wait?

Regards
  Thomas