Move `std::thread_local::*` into `std::thread`

[carllerche]

Propose moving the contents of std::thread_local::* into std::thread

Rendered

[CloudiDust]

+1 to std::thread_local -> std::thread::local.

Though this is a breaking change, it is also a small one, and if I am reading #911 correctly, then there may be breaking changes to std::thread_local (specially s::t::Key) anyway.

[blaenk]

Good idea.

[liigo]

Both std::thread::local and std::thread::tls are good places.

[jmesmon]

tls is an initialism that expands to thread_local_storage, so it'd be a little funny, one could read it as std::thread::thread_local_storage.

On std::thread::local : are we really sure that the only thing that will ever be local to threads is storage? And does local actually describe what we're using? (local is an adjective in it's usage in tls, storage is a noun)

All that said: the current name suffers from some of my nit-picking as well, so it isn't really an excuse not to change what we have now to one of the proposed items.

[aturon]

This is a great idea! The fact that thread_local ended up as a separate top-level module was mostly an accident of stabilization history.

A submodule called local is a reasonable idea; I want to chew on it a little bit more. But strong 👍 to merging in some fashion.

[CloudiDust]

@jmesmon When people talk about "thread locals" (note here local is used like a noun), I tend to assume that they are referring to "thread local storage/variables", though technically there are other things that are local to threads. So I am fine with std::thread::local.

Also, I believe we can "repurpose" the module a bit, and actually put other thread local entities inside std::thread::local if we want to.

[aturon]

Since there's very little in thread_local today, I wonder if we could instead import the contents with a prefix: LocalKey and ScopedKey perhaps?

[CloudiDust]

@aturon, LocalKey and ScopedKey are good names. I do wonder if we have plans to add things to std::thread_local/std::thread::local? If not, then we can deprecate std::thread_local and move the two types to std::thread.

[aturon]

@carllerche Can you update the RFC with the LocalKey and ScopedKey proposal? It was discussed by the core team today and is otherwise ready to be accepted.

[aturon]

This RFC, which is a minor organizational tweak, has been approved. The final decision was to use LocalKey and ScopedKey; I updated the RFC accordingly (as per @carllerche's request on IRC).

Tracking issue