JoshCheek
commented
8 years ago Update: This program, when run will hang indefinitely. When interrupted from the shell (both fish and bash do this), it kills the process, its child, and its grandchild. When interrupted with SIGINT, from another process, it kills the parent and the child, but not the grandchild. The documentation assumes a deeper knowledge of Unix than I have. Trying to read explanations of processes / process groups / sessions / etc has not improved clarity. It may be possible to prevent it from hanging, though, which would be something, at least.
require 'timeout'
require 'rbconfig'
# helper methods
def annotate(description, *allowed_exceptions, &block)
success = false
puts "\e[33mPRE #{description}\e[39m"
block.call
success = true
rescue *allowed_exceptions
success = true
ensure
puts "\e[3#{success ? 2 : 1}mPOST #{description}\e[39m"
end
def close_all(*streams)
streams.each { |io| io && !io.closed? && io.close }
end
# setup streams
parentout, childout = IO.pipe
parenterr, childerr = IO.pipe
childin, parentin = IO.pipe
# clean up when exiting the program
at_exit do
annotate("ensure streams closed") { close_all parentout, childout, parenterr, childerr, childin, parentin }
end
# the child program has its own child, both of which will hang waiting on the pipe to be closed
parentin.write <<-RUBY
puts "hello, world"
read, write = IO.pipe
if fork
write.puts "parent"
else
write.puts "child"
end
read.read
RUBY
# evaluate the code in a child process
child_pid = Kernel.spawn(RbConfig.ruby, '-', in: childin, out: childout, err: childerr)
# close streams talking to the child so that it could exit
[parentin, childout, childerr, childin].each(&:close)
# print anything we see the child print
consumer_thread = Thread.new do
parentout.each_line { |line| p stdout: line }
parenterr.each_line { |line| p stderr: line }
end
# wait for child to complete, timing out if it does not
begin
Timeout.timeout 2 do
annotate("wait for the child we started") { Process.wait child_pid }
annotate("wait for all children", Errno::ECHILD) { Process.wait }
annotate("joining consumer thread") { consumer_thread.join }
end
rescue Timeout::Error
annotate("killing child") { Process.kill "KILL", child_pid }
annotate("joining consumer thread") { consumer_thread.join }
end
In TextMate, when I run this code, it hangs indefinitely, even though there is a timeout set. Probably something in the parent process waits (probably either the consumer or EvaluateByMovingFiles). Or maybe the fork prevents the timeout from working correctly. Regardless, users shouldn't have to go grepping and killing processes on their own, SiB should respect the timeout.
There is a surprisingly large number of processes spawned to accomplish SiB on my machine. I've analyzed them and know where they all come from, except for why there is three of the bottommost invocation: