It might be, it might not be. Either way the OS doesn’t incur additional cost (since it’s doing it across applications already) and it provides for a more robust platform. In what world is it reasonable to ship an OS in this decade that is a less robust platform just because?
(I’ve hit it a few times in many-threaded tests, eg the rust test runner, in the latest it makes it basically impossible to test https://github.com/lightningdevkit/rust-lightning/pull/3680 on windows).
Sounds like Rust test framework is broken.
I guess there might be cost for applications that want to control the scheduling on their own but it could've been opt-in.
Nope, it just spawns too many threads all at once so you don’t get predictability on a platform like windows with two threads on a slow CPU. Windows really just is not a real operating system, it’s a toy that somehow convinced people to use.
That's my point, Rust test binary should only spawn as many threads as there are CPUs.
I hate Windows as much as anyone else but pretending that your tests will run faster if you inject a bunch of context switches into them is just naive.
