There's no programmatic way to tell the test runner to run serially

[brson]

I'm working on bindings for a library that cannot be called in parallel and I have no mechanism for enforcing this in my tests. Just a way to create a global handle to a worker task would be enough.

[bblum]

A cheapo solution would be

#[cfg(test)] fn actual_test_1() { ... }
#[cfg(test)] fn actual_test_2() { ... }
...
#[cfg(test)] fn actual_test_N() { ... }

#[test]
fn run_all_tests() {
    actual_test_1();
    actual_test_2();
    ...
    actual_test_N();
}

but the disadvantage of that should be pretty obvious.

Just a #[serial_test] tag would be overkill, though, since you wanna say what other tests they should be serialized with-respect-to.

One good way to do this would be to use a sync::Mutex, and then have each of the test cases run while holding the mutex, though you couldn't store it as a mutable global. Maybe stash it in TLS?

[emberian]

Triage bump.

[brson]

This was never nominated. Unmilestoning.

[reem]

Triage. This still would be useful and is still unimplemented.

[brson]

We have a variety of global synchronization solutions now so this is not the dire problem it was. I don't think its necessary to do anything here.