Reference & SharedReference design issue wrt. memory consumption

[lbarthon]

I created a bit earlier this week the #1952 issue, which has been closed, but I'm still seeing a problem.

The "memory leak" naming was incorrect. Per my understanding, what happens is that if a struct B has a SharedReference<A, ()>, the pointer to the data that needs the reference (in that case, ()) is only cleaned up when the finalize_callbacks run for that said instance of A - see code here https://github.com/napi-rs/napi-rs/blob/main/crates/napi/src/bindgen_runtime/js_values/value_ref.rs#L137-L142

This means that if I have an object A that lives for a very long time (potentially even global), and I have many objects B that live for a short period of time, the pointer to what those objects hold will only be cleaned up when A is cleaned up, which might never happen. This will work, and will not be a problem in any use case where A doesn't live for too long (although it'll use more memory).

This reference dropping would need to work bilaterally. If B has a SharedReference on A, when B is dropped it needs to be dropped, and vice-versa. It seems that napi-rs misses half of that logic.

I tried implementing it myself, adding a Drop impl on SharedReference, but without success. Any implementation would also need a way to know if the pointer has been dropped already...or not, which seems like a hassle.

In the meantime, do you see a way for us to use lifetimes more efficiently? I might just declare my global in napi-exposed functions, and go from there. Thanks!