I ended up reading through the links and can see the benefit in doing NaN boxing, however, what is the memory footprint of the fat enum being used currently? Not totally sure how or whether Rust does any sort of padding or whatnot on fat enums.

I ended up reading through the links and can see the benefit in doing NaN boxing, however, what is the memory footprint of the fat enum being used currently? Not totally sure how or whether Rust does any sort of padding or whatnot on fat enums.

Currently the size of Value is 24 bytes (size_of::<Value>()).

If you're looking for impl of NaN boxing in Rust Starlight has two different implementations of them for handling JS values: https://github.com/Starlight-JS/starlight/blob/dev/crates/starlight/src/vm/value.rs :)

If you're looking for impl of NaN boxing in Rust Starlight has two different implementations of them for handling JS values: https://github.com/Starlight-JS/starlight/blob/dev/crates/starlight/src/vm/value.rs :)

Awesome! thanks, for the suggestion. I have implemented NaN boxing locally, but the problem is refactoring the existing match value stuff (which we use a lot, this generates a couple of hundred of erros 😅 )

What about having ValueVariant enum which essentially can hold all the values NaN-boxed value can encode and then add JsValue::variant() which returns this ValueVariant. I'm pretty sure Rust and LLVM are able to optimize this.

What about having ValueVariant enum which essentially can hold all the values NaN-boxed value can encode and then add JsValue::variant() which returns this ValueVariant. I'm pretty sure Rust and LLVM are able to optimize this.

Yup. thought about this as a possible solution and it's probably the best solution, since refactoring will be a nightmare otherwise.

@HalidOdat any news on this?

@HalidOdat any news on this?

Sorry for not replying I have been very busy, but plan to start to do more contributions in the future.

I'll start working on this again. had a semi-working implementation of this that needs a rebase (it's pretty old). Hopefully I can get it working by this weekend :)

Will try to do some investigation on how to reduce the size of JsObject.

Would be nice to revisit this as part of our performance focus, I’m happy to team up with @jedel1043 if at some point you wanted to take another look at this

PR #4080 is the first step towards this.