Extended Rc uniqueness methods #14908
Conversation
| /// count is 1 (this pointer is unique). | ||
| #[inline] | ||
| #[experimental] | ||
| pub fn maybe_unique<'a>(&'a self) -> Option<&'a mut T> { |
There was a problem hiding this comment.
This isn't safe because aliasing references can be created before or after calling maybe_unique. It would need to take &'a mut self and it would also need to prevent doing this if any Weak<T> reference exists because it could be upgraded. There have been implementations of this before and I think the previously proposed approaches need to be considered too.
There was a problem hiding this comment.
Thank you, fixed now.
There was a problem hiding this comment.
Please note that maybe_unique does check for weak() count to be 1.
|
I'd suggest rewriting |
|
Thanks @eddyb. It's fixed now. I was hesitating to do that because |
| // additional reference of either kind. | ||
| if self.strong() != 1 || self.weak() != 1 { | ||
| *self = Rc::new(self.deref().clone()) | ||
| if self.maybe_unique().is_none() { |
There was a problem hiding this comment.
This is not what I meant. You should be able to match on the result of maybe_unique.
(though, in retrospective, I may have underestimated the complexity of this change)
There was a problem hiding this comment.
That's what I initially tried, but Rust doesn't allow reassigning *self in the the match arm of maybe_unique() because the latter freezes self.
There was a problem hiding this comment.
You have to return early in the other branch and mutate outside of the match.
There was a problem hiding this comment.
Reverted make_unique() back to the original implementation after a brief discussion with @eddyb. The unsafe{} can be replaced by 2 calls to maybe_unique(), but that may not be as efficient.
|
@kvark I think I understand the use-case for Can you spell out how |
|
@aturon Thanks for having a look at this PR. I don't have a |
|
These should not be methods on The current state of my PR #15399 actually provides the same functionality under different names, along with a 3rd function to unwrap a uniquely-owned If you have no objection, I'm inclined to re-submit just that one commit as a PR (that would obsolete this PR). |
|
@kballard Thanks, I trust you to take care of this. Besides, I'm long past the point where I needed it. |
maybe_unique()allows to mutate the referenced data if the pointer is guaranteed to be the only one.is_last_standing()allows to check if this pointer is the last strong one.One possible application of these methods is a shared resource factory. The resource doesn't know how it is built or destroyed, it may even be method-less. The factory knows how to create a resource and how to destroy it. Upon creation, the factory stores
Rc<T>internally and returns a cloned reference. Later, at a designated (by the user) moment, the factory can sweep over known references and delete only those resources that are no longer needed by anyone else (here comesmaybe_uniqueoris_last_standing).