Port to windows-sys

[kennykerr]

I noticed that all of the tokio dependencies have already made the switch to windows-sys, so I thought I'd take a stab at porting tokio itself just to confirm that there aren't any gaps.

https://github.com/kennykerr/tokio/pull/1/files

All tests pass as far as I can tell. There are some doc dependencies on winapi which I'm not too sure about, otherwise I'd submit a PR. Anyway, just thought I'd point you to my port in case someone wants to pursue it further.

[Darksonn]

Considering that all of our dependencies have switched, I think it is reasonable that we switch too.

[davidbarsky]

Speaking not as a reviewer, but instead a consumer of Tokio, windows-sys is moving to rust's not-yet-stabilized raw-dylib. This is particularly impactful for my employer, as vendoring windows-sys requires stripping out all the #[link] arguments for complicated, Buck-related reasons. Would it be possible to have a feature hidden behind tokio_unstable that breaks MSRV to use a version of windows-sys that uses raw-dylib?

(I'm genuinely unsure whether rustflags can influence dependency resolution, and if it can, whether it is a good idea from a maintenance perspective.)

[kennykerr]

raw-dylib support is modeled after tokio_unstable and is offered via the windows_raw_dylib flag.

microsoft/windows-rs#2164

[davidbarsky]

I didn't realize that you were the one responsible for that change! Thanks so much for doing that. raw_dylib support in windows-sys is going to make vendoring windows-sys a lot a lot easier.

[Darksonn]

Would it be possible to have a feature hidden behind tokio_unstable that breaks MSRV to use a version of windows-sys that uses raw-dylib?

If windows-sys plans to unconditionally use features that require a very recent MSRV, then that's something we should be very aware of when deciding whether we should use it. But it sounds like this is not the case?

[kennykerr]

Right, this will continue to be an opt-in for windows-sys via the windows_raw_dylib flag for the foreseeable future to avoid requiring a major MSRV bump.

[kennykerr]

I went ahead and ported the examples and docs as well. Here's a PR: #5204

Feel free to close if you would prefer to manage the port yourself. 😊