Bad handling of thread join deadlock in standard library

[eefriedman]

use std::sync::mpsc::Sender;
use std::sync::mpsc::Receiver;
use std::thread::JoinHandle;
fn main() {
    let (c1s, c1r) = std::sync::mpsc::channel::<JoinHandle<_>>();
    let (c2s, c2r) = std::sync::mpsc::channel::<JoinHandle<_>>();
    let (c3s, c3r) = std::sync::mpsc::channel::<()>();
    let t1 = std::thread::spawn(move || {
        let () = c1r.recv().expect("D").join().expect("A");
    });
    let t2 = std::thread::spawn(move || {
        let () = c2r.recv().expect("E").join().expect("B");
        c3s.send(());
    });
    c1s.send(t2);
    c2s.send(t1);
    c3r.recv().expect("C");
}

Gives (on playground):

thread '<unnamed>' panicked at 'called `Option::unwrap()` on a `None` value', ../src/libcore/option.rs:325
note: Run with `RUST_BACKTRACE=1` for a backtrace.
thread '<unnamed>' panicked at 'B: Any', ../src/libcore/result.rs:785
thread '<main>' panicked at 'C: RecvError', ../src/libcore/result.rs:785

There are two problems here:

1. The error message is absolutely terrible.
2. For the sake of safety, we might need to do something other than panic if this happens. This could affect crates which require that join() never returns early, like crossbeam.

[Mark-Simulacrum]

Possibly in #44112 or otherwise, but this seems to have since been fixed -- this now deadlocks.