Dropping EventLoopHandle after running a request causes the program to hang.

[abrodersen]

Dropping EventLoopHandle before dropping Web3<T> causes a hang if at least one request has been issued. Example code:

extern crate web3;

use web3::futures::Future;

fn main() {
  let (eloop, http) = web3::transports::Http::new("http://localhost:8545").unwrap();
  let web3 = web3::Web3::new(http);
  let _accounts = web3.eth().accounts().wait().unwrap();

  drop(eloop);
}

This became an issue when I created a struct that owned both EventLoopHandle and Web3<T>. If my struct fields were in the wrong order, the program would hang when attempting to drop my struct. If I switched the order of the fields, the program would run to completion.

[tomusdrw]

Good catch it's really interesting why it fails.

[aleksuss]

I have the same issue

[aleksuss]

FYI: But if I use an event loop from tokio everything is OK

[yangby-cryptape]

After test, I think it is hang at EventLoopHandle::Drop.

Move the event_loop into another thread and ignore it can avoid this bug (uncleanly exit?).

[yangby-cryptape]

Finally, I have known what happened.

It hangs at event_loop.turn(None);, before checking the done flag.

When we want to shutdown the event_loop, we stop send any new message, and set the done flag.
But the event_loop will check the done flag after next message, so it hangs at waiting for next message.
So, it hangs.

The simplest way (but stupid?) to fix this bug is: after set the done flag, send to it a new message.

[tomusdrw]

Perhaps we could supply a timeout to turn?

[synlestidae]

@yangby-cryptape is correct I think. I added self.remote.spawn(|_| Ok(())) before the thread join. It seems to fix it on my end.

#144

[tomusdrw]

Resolved via #144