program dies silently when doing I/O on a defunct child Process

[japaric]

Scripts to reproduce

// loopback.rs
use std::io::stdio::stdin;

fn main() {
    for line in stdin().lines() {
        match line {
            Ok(line) => print!("{}", line),
            Err(e)   => fail!("whoops: {}", e),
        }
    }
}

// process-fail.rs
use std::io::process::Process;
use std::io::timer::Timer;

fn main() {
    let mut timer = Timer::new().unwrap();
    let periodic = timer.periodic(1000);

    let mut child = Process::new("./loopback", []).unwrap();
    let mut count = 10;

    println!("write to child process in")
    while count > 0 {
        println!("{}", count);
        count -=1;

        periodic.recv();
    }

    match child.stdin.get_mut_ref().write_str("Hello world!\n") {
        Ok(_)  => println!("succeed"),
        Err(_) => fail!("whoops"),
    }

    println!("end of main");
}

Steps to reproduce:

$ ./process-fail
$ killall loopback  # before countdown ends

Output:

Backtrace: None
Exit code: 141

Version:

rustc 0.10-pre (e6468a8 2014-03-24 10:01:57 -0700)
host: x86_64-unknown-linux-gnu

[alexcrichton]

What are you expecting to happen? When you send a TERM signal to a process without a registered handler, it terminates promptly. The runtime currently doesn't register any TERM/INT handlers, that's up to you.

[japaric]

What are you expecting to happen?

I was expecting the runtime to return some Err() upon requesting an action from a defunct process. I also tried polling the state of the defunct process using signal(0) (as mentioned in the documentation), but this returned Ok() as well.

When you send a TERM signal to a process without a registered handler, it terminates promptly. The runtime currently doesn't register any TERM/INT handlers, that's up to you.

Even if the runtime dosn't "catch" the SIG{TERM,INT}, is it not possible at all to fail!() or return Err() in any of the two cases mentioned above?

[alexcrichton]

Oh whoops, sorry! I had my understanding backwards, and apparently this isn't a bug on OSX which caused more confusion for me.

Anyway, closing as a dupe of #13124, it's the same cause.

[alexcrichton]

Oh wow, I've done it again. I'm sorry, again! I apologize for not investigating more thoroughly.

This is indeed separate from #13124