Relax orderings in std::sync::once

[jeehoonkang]

Currently, in std::sync::once, every access to the shared objects are SeqCst, while it doesn't need to be. I relaxed orderings to Acquire, Release, and AcqRel.

r? @alexcrichton

[rust-highfive]

Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @alexcrichton (or someone else) soon.

If any changes to this PR are deemed necessary, please add them as extra commits. This ensures that the reviewer can see what has changed since they last reviewed the code. Due to the way GitHub handles out-of-date commits, this should also make it reasonably obvious what issues have or haven't been addressed. Large or tricky changes may require several passes of review and changes.

Please see the contribution instructions for more information.

[alexcrichton]

Thanks for the PR! This has actually been proposed before and my own personal opinions at least haven't changed much.

Mind reading over some of the discussion there and posting your thoughts on it?

[jeehoonkang]

@alexcrichton Thank you! I haven't notice that there was a closed PR on this issue. I will discuss further in that issue.

[aidanhs]

Hi @jeehoonkang @alexcrichton, just wondering what the status of this PR is after the (brief) exchange on #31650 - is this waiting on review, or are we looking for a bit more justifying background?

[alexcrichton]

@jeehoonkang do you think SeqCst as-is today is wrong? (in the LLVM implementation rather than the standard) If not, do you think this has a technical benefit over the current construction?

[jeehoonkang]

@aidanhs @alexcrichton The current implementation with SeqCst is correct. What I want to argue is that the reason it is correct is exactly the same as the reason a version with Release and Acquire is correct, because currently LLVM's SeqCst is broken and it doesn't give you a better reasoning principle than Release and Acquire.

That said, I kinda agree with the conclusion made in #31650 that we keep SeqCst. Without a formal proof, it would be just safer to keep the strongest ordering, which is SeqCst. If performance difference is not that great, I agree that we don't merge it now.

I agree to close this PR. Maybe later we can reopen, once someone clearly explains why Release and Acquire is safe in once.

[alexcrichton]

Ok sounds good to me, thanks regardless @jeehoonkang !