fix: close https://github.com/cloudflare/workerd/issues/3620

[rxliuli]

ref #3620

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}

[rxliuli]

---

I have read the CLA Document and I hereby sign the CLA

---

[kentonv]

Won't this match plain objects, and thus bypass the next line below? So we won't end up recursively applying Stubify to the contents of objects?

I guess the problem you're experiencing (in your bug report) is that Date is matched by the line below, which then tries to recursively Stubify the properties of Date, and for some reason it is applying to all the prototype properties like toString, toDateString, toTimeString, and so on. I'm sort of surprised by this though, if [K in keyof T] actually matches prototype properties, why isn't it constantly getting hung up on Object.prototype properties which all objects have? Why only Date.prototype properties?

If this is just the way it works... then I guess what we actually need to do here is define a variant type that's narrower than Serializable which includes just the builtin types that are explicitly handled, like Date.

[rxliuli]

Currently, in the definition of Serializable, except for the Stub/Stubable type which I am not sure what it is doing, the others seem like they should not use Stubify wrapping.

https://github.com/cloudflare/workerd/blob/5fe608c86ffa23efa21c9cd06136e5e79e653b55/types/defines/rpc.d.ts#L38C1-L71C16

For example, the RegExp type is also incorrect

[kentonv]

I may be misreading as I don't know typescript super-well, but doesn't this part of Serializable match any plain object that is serializable?

    | {
        [K in keyof T]: K extends number | string ? Serializable<T[K]> : never;
      }

And won't that mean that your newly-added line will map any such plain object type to itself, thus bypassing the next line which would otherwise have recursively stubified the contents?

[kentonv]

Per comment, I think this breaks things as-is.

Removing per @kentonv's comments

[kentonv]

@penalosa any thoughts on this?