Added a method that allows for conditional mutable access #132
Conversation
| return None; | ||
| } | ||
| Some(unsafe { ptr.as_mut().unwrap() }) | ||
| } |
There was a problem hiding this comment.
🤔 This might actually work.
However, some more argument / (non-formal) proof that this is actually OK and safe would be in place here.
Also, instead of having 3 unsafe blocks just one after another, maybe we can just put all three things into a block.
| drop(guard); | ||
| assert_eq!(1, Arc::strong_count(&ptr)); | ||
| } | ||
|
|
There was a problem hiding this comment.
I think a check for the try_mut access, calls for a dedicated test (I know there's one „hidden“ in this long method, but we probably should have a test where it can be found easily). Especially, making sure it is not accessible in all the cases listed above.
| @@ -67,6 +68,9 @@ pub unsafe trait RefCnt: Clone { | |||
| /// This must not be called by code outside of this crate. | |||
| unsafe fn from_ptr(ptr: *const Self::Base) -> Self; | |||
|
|
|||
| /// Returns the reference count of the pointer. | |||
| unsafe fn ref_cnt(ptr: *const Self::Base) -> usize; | |||
There was a problem hiding this comment.
Now, adding a new method into the trait would be a semver-breaking change.
Worse, the crate also offers (under a feature) RefCnt for Weak… and I don't think this method is valid (eg. makes sense) for that type.
| if ptr.is_null() { | ||
| 0 | ||
| } else { | ||
| 1 |
There was a problem hiding this comment.
This doesn't look correct. This implementation is for things like ArcSwap<Option<Arc<X>>. And the inner Arc can have multiple references.
|
Also, I'd suggest you do some measurements. This method will likely be expensive (the wait_for_readers thing). |
Adds a new method
try_mutwhich if there is only a single reference to theArcSwapgrants mutable access to the insides