Add in_place_scope_fifo
#855
Conversation
This matches the recent `in_place_scope`, but using `ScopeFifo`. The scope constructors have also been refactored to avoid duplication.
|
cc @rocallahan |
| /// | ||
| /// Unsafe because it must be executed on a worker thread. | ||
| unsafe fn complete<FUNC, R>(&self, owner: Option<&WorkerThread>, func: FUNC) -> R | ||
| fn complete<FUNC, R>(&self, owner: Option<&WorkerThread>, func: FUNC) -> R |
There was a problem hiding this comment.
Why is this not unsafe? I guess it never had to be...?
There was a problem hiding this comment.
Well, at least the comment "because it must be executed on a worker thread" is optional after #844. I didn't really change that functionality in this PR, just cleaning up.
I suppose there's a correctness issue if Some(owner) doesn't match the latch, but we do have debug_assert_eq for that, and it's not really a safety concern. Worst case is that we can't wake up the right thread. We control all the callers in this same module though, and they do trivially match from new to complete.
(We can't put the &WorkerThread directly in the scope or latch because it's not Sync.)
| }, | ||
| None => ScopeLatch::Blocking { | ||
| latch: CountLockLatch::new(), | ||
| }, |
|
r+ from my side :) @cuviper were you ready to merge? |
|
Sure! bors r=nikomatsakis |
This matches the recent
in_place_scope, but usingScopeFifo. Thescope constructors have also been refactored to avoid duplication.