make rng seed extendable #26
Conversation
|
does this PR make sense to you? |
|
|
||
| // Seed abstraction, so users can specify their own | ||
| #[derive(Debug, PartialEq, Eq)] | ||
| pub struct CellSeed(Cell<u64>); |
There was a problem hiding this comment.
I don't think the new-type struct abstraction is necessary, it imo should be omitted. It might be a good idea to replace it with a type alias.
pub type CellSeed = Cell<u64>;
|
As mentioned in #31 it might be more beneficial to get rid of interior mutability altogether and instead let the user wrap it in whatever interior mutability abstraction they want to use. |
| fn with_seed(seed: u64) -> Rng<Self> where Self: Sized { | ||
| Rng::from(AtomicSeed(AtomicU64::new(seed))) | ||
| } | ||
|
|
||
| fn get(&self) -> u64 { | ||
| self.0.load(Ordering::Relaxed) | ||
| } | ||
|
|
||
| fn set(&self, new_seed: u64) { | ||
| self.0.store(new_seed, Ordering::Relaxed) | ||
| } |
There was a problem hiding this comment.
If thread B reads a value between the time thread A reads it and the time it stores the value, thread A and thread B will get the same value from rng. I guess there are both use cases where that is ok and use cases where it is not ok, but I'm not positive about having an API that is easy to misuse.
(If you want to create atomic rng, I think you need to replace get/set in gen_u64 with fetch_add)
Cellis notSendthus it cannot be used as struct field in async environment.I tried to implement cursor, which needs to have its' data shuffled in a deterministic manner.
example: