How can I proxy a panic message from a thread pool to the test endpoint?

help
1

I have some thread pool. It creates some async runner and executes a provided future. It also uses catch_unwind to catch panics. If a panic occurs, a thread from the pool shuts down and sends via a sync channel the caught data. The main test endpoint periodically checks all spawned (by this endpoint) handles and resume_unwind if it gets a panic. Like

pub(crate) type JobResult = thread::Result<()>;

pub(crate) struct ExecutorPoolJoinHandle {
    was_joined: bool,
    sync_receiver: SyncReceiver<JobResult>,
}


pub(crate) fn join_timeout_maybe_resume_panic(
        &mut self,
        timeout: Option<Duration>,
    ) -> Result<(), RecvTimeoutError>
    {
        assert!(!self.was_joined);

        let res = if let Some(timeout) = timeout {
            match self.sync_receiver.recv_timeout(timeout) {
                Ok(res) => res,
                Err(RecvTimeoutError::Timeout) => return Err(RecvTimeoutError::Timeout),
                _ => panic!("{BUG_MESSAGE}"),
            }
        } else {
            self.sync_receiver.recv().expect(BUG_MESSAGE)
        };

        self.was_joined = true;

        if let Err(err) = res {
            panic::resume_unwind(err);
        }

        Ok(())
    }

And it sees panics. It also fails tests. But in the RustRover test's window, I don't see the panic message. If I run it with the console, I can see something like it:

---- runtime::epoch_gc::test::test_epoch_gc_concurrent stdout ----
Test test_epoch_gc_concurrent started!

thread 'orengine-test-executor' panicked at src/runtime/epoch_gc/test.rs:116:9:
assertion left == right failed
left: 4
right: 24

Looks like all messages are forwarded to the stdout. How to forward them to the stderr?

2

This is not the first place to look. Often, Rust is set up to capture test output and only show it if the test fails. I would start by investigating whether test capturing is set up correctly.

With cargo, the relevant flag is --no-capture.

3

I don't think you can. The test running takes control of the global panic hook so you should not change it and the Err from catch_unwind can't be extracted into anything meaningful. Best you possibly can do it panic at that point and say the thread panicked so look at std out for its message.

4

But the test fails, and I can't see the output. I would panic one more time with the message, but it is Box<dyn Any + Send> and can't be downcast to String or &str.

Yes, it works and shows the panic message, but I want to repair the output for IDEs. I write my own async runtime, and it is bad if users can't debug their programs because of this issue. I can rewrite the test module to show panics from the main test endpoints, but I want to show all panics (even in other threads but spawned by one endpoint).

5

Honestly, it sounds like an IDE bug.

6

Before choosing a "hacky" solution, I decided to test all the theories I had. It turned out that panics are displayed correctly if threads are not reused. Moreover, you need to avoid reusing even those threads that did not catch any panics. I don't know why this happens. If you know how to actually fix this bug while still reusing threads, please supplement my answer.

7

This topic was automatically closed 90 days after the last reply. We invite you to open a new topic if you have further questions or comments.