Panic on None when joining on thread which is called yield_now

[zonyitoo]

Minimal test case:

use std::thread;

fn main() {
    let h = thread::spawn(|| {
        thread::yield_now();
    });

    h.join().unwrap();
}

This piece of code will panic with output:

thread '<main>' panicked at 'called `Option::unwrap()` on a `None` value', ../src/libcore/option.rs:367
playpen: application terminated with error code 101

Tested on playpen with stable Rust.

[alexcrichton]

I can reproduce this on playpen but not locally unfortunately.

The assertion being tripped here indicates that pthread_join returned but the return value of the thread was not filled out, so this is potentially a data race as well (hence tagging with I-nominated and T-libs). This may have to do with pthread_join returning an error but us not detecting it because the assertion only happens in debug mode. A minimal reproduction of this behavior, however, does not indicate that pthread_join is returning an error.

This would probably be easier to debug if we could reproduce it somewhere where we could debug it, but I would also be happy for now just leaving the assertion in place but switching to using Mutex or something to prevent a possible data race.

@zonyitoo did you discover this through the playpen itself, or was this something that happened locally?

[alexcrichton]

Oh wait after a little more investigation, it looks like this program exhibits similar behavior. It looks like `sched_yield causes the thread to be terminated due to playpen considering it a bad system call.

The behavior here appears to be that the thread is immediately terminated and yet pthread_join returns 0 (success). This means that in the standard library we never hit the code that stores the return value of the thread, hence the assertion being hit.

I'm removing the nominated/unsound tags as if the thread is being killed this isn't a safety issue, it seems like it's just a bug that may want a better error message.

[zonyitoo]

I discovered it on playpen itself. It seems to works well on my laptop.

So it likely is a bug in playpen?

[tbu-]

In some sense, yes. You could ask the playpen maintainers to add sched_yield to the list of safe system calls.

[zonyitoo]

Ok, closing and going to ask playpen maintainers to fix this.