Re-assigning a closed TcpStream and reading causes a segmentation fault

[rapha]

$ cat segv.rs    
use std::io::{Acceptor, Listener, net};

fn main() {
  let address = net::ip::SocketAddr { ip: net::ip::Ipv4Addr(127, 0, 0, 1), port: 1234 };
  std::task::spawn(proc() {
    let mut socket = std::io::TcpListener::bind(address).listen().accept();
    socket.write([0]).unwrap();
    println!("server: done");
  });
  let socket = std::io::TcpStream::connect(address);
  std::task::spawn(proc() {
    loop {
      let mut socket = socket; // Here is the problem! Fixed if assignment happens outside loop.
      println!("client: about to read");
      socket.read_byte().unwrap();
      println!("client: read successfully");
    }
  });
}
$ rustc segv.rs && ./segv
client: about to read
server: done
client: read successfully
client: about to read
Segmentation fault: 11
$

[huonw]

Thanks for the report; this is a rather annoying bug that's filed as #12041.

(It's just the proc + loop interaction, nothing specific to TcpStream etc.)