Remove sync::Once::call_once 'static bound

[CAD97]

See https://internals.rust-lang.org/t/sync-once-per-instance/7918 for more context.

Suggested r is @alexcrichton, the one who added the 'static bound back in 2014. I don't want to officially r? though, if the system would even let me. I'd rather let the system choose the appropriate member since it knows more than I do.

git blame history for sync::Once::call_once's signature:

• std: Second pass stabilization of sync (Dec 2014)

  -    pub fn doit<F>(&'static self, f: F) where F: FnOnce() {
  +    #[stable]
  +    pub fn call_once<F>(&'static self, f: F) where F: FnOnce() {

• libstd: use unboxed closures (Dec 2014)

  -    pub fn doit(&'static self, f: ||) {
  +    pub fn doit<F>(&'static self, f: F) where F: FnOnce() {

• std: Rewrite the sync module (Nov 2014)

  -    pub fn doit(&self, f: ||) {
  +    pub fn doit(&'static self, f: ||) {

   The second layer is the layer provided by `std::sync` which is intended to be
   the thinnest possible layer on top of `sys_common` which is entirely safe to
   use. There are a few concerns which need to be addressed when making these
   system primitives safe:
  
     * Once used, the OS primitives can never be **moved**. This means that they
       essentially need to have a stable address. The static primitives use
       `&'static self` to enforce this, and the non-static primitives all use a
       `Box` to provide this guarantee.
  

The author of this diff is @alexcrichton. sync::Once now contains only a pointer to (privately hidden) Waiters, which are all stack-allocated. The 'static bound to sync::Once is thus unnecessary to guarantee that any OS primitives are non-relocatable.

As the 'static bound is not required for sync::Once's operation, removing it is strictly more useful. As an example, it allows attaching a one-time operation to instances rather than only to global singletons.

[rust-highfive]

r? @bluss

(rust_highfive has picked a reviewer for you, use r? to override)

[CAD97]

Note also that parking_lot:0.6's Once takes &self for its Once::call_once implementation: https://docs.rs/parking_lot/0.6/parking_lot/struct.Once.html#method.call_once

Differences from the standard library Once

• Only requires 1 byte of space, instead of 1 word.
• Not required to be 'static.
• Relaxed memory barriers in the fast path, which can significantly improve
  performance on some architectures.
• Efficient handling of micro-contention using adaptive spinning.

[CAD97]

The only reason I can see that this would be problematic is if it causes inference issues. I don't know how this might cause inference breakage, but I've long since given up trying to predict what changes to signatures can cause inference breakage at this point.

[alexcrichton]

Thanks for the PR @CAD97! I think this should be a safe change to apply backwards-compatibly and thanks for doing the digging here to see where the 'static came from! It was indeed needed long ago for safety but nowadays it no longer needs it, so I believe this should all be safe to merge. To be extra sure though let's...

@rfcbot fcp merge

[rfcbot]

Team member @alexcrichton has proposed to merge this. The next step is review by the rest of the tagged teams:

• @Kimundi
• @SimonSapin
• @alexcrichton
• @dtolnay
• @sfackler
• @withoutboats

No concerns currently listed.

Once a majority of reviewers approve (and none object), this will enter its final comment period. If you spot a major issue that hasn't been raised at any point in this process, please speak up!

See this document for info about what commands tagged team members can give me.

[rfcbot]

🔔 This is now entering its final comment period, as per the review above. 🔔

[alexcrichton]

@bors: r+

Alrighty!

[bors]

📌 Commit 0f3f292 has been approved by alexcrichton

[Mark-Simulacrum]

@bors rollup