fix shutdown_timeout(0) blocking

[sharnoff]

Background

After a brief discussion with @Darksonn about the precise behavior given by various methods of shutting down a Runtime (dropping it, shutdown_timeout, shutdown_background), I was experimenting with the available options to try to determine the differences between them.

The issue

During this experimentation, I discovered that calling rt.shutdown_timeout(Duration::from_nanos(0)) actually does wait until all blocking tasks finish, contrary to what would be expected. This is notably different from if we passed Duration::from_nanos(1) instead, which has the effect of shutting down immediately. As best I can tell, this is a bug.

Here's a sample program on the playground that demonstrates this behavior.

The solution

The issue seems to be that the implementation of the runtime's shutdown channel's Sender::wait is returning true (indicating that the timeout duration has not been reached) when given a duration of zero. When that happens, we take extra steps that rely on those tasks having finished.

Other things

This might be a breaking change - I wouldn't expect anyone to be directly relying on the behavior of shutdown_timeout(0), but anyone using shutdown_background might suddenly find that certain tasks are unexpectedly terminated when it wasn't happening before. This is because the current behavior seems to be identical to simply dropping the Runtime, which is more lenient than what the documentation of shutdown_background indicates.

[carllerche]

Thanks! Is there any way to add a test so we don't regress in the future?

[sharnoff]

@carllerche Sorry for not getting back to you - I ended up having a really busy week. I'm glad to see that there's a test now :)

[Darksonn]

Don't worry about it! We are happy to have your contribution.