Hi,
I think the issue here is not that pipes are not closed after fork/before exec but something else.

Signal::INT.trap do
  puts "Canceled by user"
  exit
end

puts "blocking before: #{STDIN.blocking}"
system ""
puts "blocking after: #{STDIN.blocking}"
puts "Press CTRL+C"
gets

Output:

blocking before: false
blocking after: true
Press CTRL+C

A call to system() changes the blocking state of STDIN
If i manually set the blocking state to false, it responds to Ctrl-C /SIGINT as expected.

Signal::INT.trap do
  puts "Canceled by user"
  exit
end

system ""
STDIN.blocking = false
puts "Press CTRL+C"
gets 

Output:

Press CTRL+C
^CCanceled by user

Good grief, my patch doesn't work anymore 😭

@forksaber ohhh, that's interesting!

Actually, STDIN, STDOUT and STDERR all got their blocking state changed.

Since signals are handled through the event loop, but STDIN is now blocking (unexpected) gets ends up blocking... and signals are never handled.

We should probably do what @RX14 said once: let's not change the blocking state of STDIN, STDOUT and STDERR. Yeah, they will be blocking, but we'll get rid of a ton of problems. Later when threads are implemented we can revisit the issue.

I'm closing this PR as it doesn't solve anything: the issue is that we can't set O_NONBLOCK on STDIN, STDOUT and STDERR and can't expect them to not be changed by other processes.