Move some things to std::sync::poison and reexport them in std::sync

[GrigorenkoPV]

Tracking issue: #134646

r? @tgross35

I've used sync_poison_mod feature flag instead, because sync_poison had already been used back in 1.2.

[tgross35]

Could this skip just get applied to specific blocks, like in https://github.com/rust-lang/rust/blob/addbd001ec56741829f20a3000892f8620dd0843/library/core/src/unicode/mod.rs? Applying to the whole file would probably make it a bit too easy for things to get messy.

[GrigorenkoPV]

Then pretty much every block of uses would have to be annotated, because otherwise rustfmt will regroup them.

IMO this looks even noisier and less accurate with all the additional attributes and inability to leave two blank lines between the groups (rustfmt removes double empty lines).

[tgross35]

I don't think this plays with rustdoc nicely. Searching for Mutex now gives the absolute std::sync::poison::Mutex path, where it has an "unstable" mark. But if you open it in std::sync::Mutex, it is fine.

These should probably be marked doc(inline) for now. But @rust-lang/rustdoc is there something we can do better for stable types reexported from unstable modules? This is also a problem for other types too, e.g. FromBytesUntilNulError looks like it is unstable because it got moved from core::ffi to core::ffi::c_str.

Edit: opened #134702

[GrigorenkoPV]

These should probably be marked doc(inline) for now.

Done. This helped with the search, but not with the things like MutexGuard's doc's reference to Mutex::lock being converted to the absolute (unstable) std::sync::poison::Mutex path.

[tgross35]

I am not sure what you mean, could you post a screenshot (edit: on the rustdoc issue, #134702)? Unfortunately we probably should wait on a Rustdoc fix then, this would be a lot of API to have misleading documentation.

[tgross35]

We are at least two years off from an edition where we would change these defaults, so I don't think this needs to be addressed at this point. Just reexporting from self::poison should be fine, things might change between now and the next edition.

[GrigorenkoPV]

#134702 got closed, but #134702 (comment) is still a problem.
I guess we can manually qualify all the links in docs to lead to std::sync::* until the std::sync::poison module is stabilized, but that sounds a bit suboptimal.