I am writting a double-index ecs component storage, by wrap the sled api.
// Define a merge function trait like MergeOperator, let my lib user to write the function
pub trait MergeFn: Fn(EID, Option<Value>, Delta) -> Option<Value> + Send + Sync{}
I found that it is not possible to insert a trait-object(Rc<dyn MergeFn> in my case) into a Tree as its merge_operator. Because the set_merge_operator function's signature is strict to a static implement.
I design a trait to let my user write there own closure, an wrap them into a Rc<>. Function pointer is also a workround but use triat object is encouraged by the Book, see more in here.
PR content
Just change the signature to a Box<...>.
All test cases related are changed.
cargo test --features light_testing are passed.
1 more thing
(Not include in this pr)
In my scanery, merge function is considered as stateless, so it's a Fn trait, instead FnMut, also the merge function is init at the beginning of the program, and never changed after that. So I think the mutex to protect the function isn't needed. The merge function should be passed at the init pharse of the Tree when the program begin, and not protected by a Mutex.
Pros
No cost when set or call the merge function during run-time.
Cons
One tree can only hold one immutable merge function during its lifetime.
User must impl their own inner-mutable data in the merge function.
RuofengX
commented
9 months ago
Motivation
I am writting a double-index ecs component storage, by wrap the sled api.
I found that it is not possible to insert a trait-object(
Rc<dyn MergeFn>in my case) into a Tree as its merge_operator. Because the set_merge_operator function's signature is strict to a static implement.I design a trait to let my user write there own closure, an wrap them into a
Rc<>. Function pointer is also a workround but use triat object is encouraged by the Book, see more in here.PR content
cargo test --features light_testingare passed.1 more thing
(Not include in this pr)
In my scanery, merge function is considered as
stateless, so it's aFntrait, insteadFnMut, also the merge function is init at the beginning of the program, and never changed after that. So I think the mutex to protect the function isn't needed. The merge function should be passed at the init pharse of the Tree when the program begin, and not protected by a Mutex.Pros
Cons