First-class functions

[dcodeIO]

Introduces a wrapper class Function<T> to represent references of a function type, including preparation for closure environments and an initial implementation of Function#call supporting calls with substituted this arguments.

Turned out to be a much larger change than initially expected due to tidying up types, also anticipating that we'll have both internal and external references eventually. Similarly, what previously was FunctionTarget now is represented by a runtime class, and there is an additional opportunity now to also get rid of Signature by deriving from the class.

In general this is getting somewhat too messy for my taste in that the code wasn't initially designed with first-class-everything in mind, so refactoring this even more by also introducing Object might help and making the compiler-side integration more natural might be good.

• I've read the contributing guidelines

[MaxGraey]

is it generated in compile time or just not yet implemented?

[MaxGraey]

Perhaps will be gread extend nameof<T>() to nameof<T>(value: T) and do something like this return nameof(this)

[dcodeIO]

There is a difference to JS here in that a Function class represents unique signatures. There can be multiple functions with the same signature, and signatures don't have names. So far, the implementation here does what JS does for anonymous functions, essentially dropping any names that would have to be baked into the binary otherwise.

[dcodeIO]

There might be a way to make .name a builtin if that's important, but I assume that it turns out that we need to ship a full symbol map (names of all the functions potentially called indirectly) with each binary to achieve this. Imagine for example that there is a variable one can assign multiple functions to, then we'll have to look up the name at runtime somehow, for example by mapping from indexes to names. Not sure if the extra bytes to ship are worth it.

[MaxGraey]

Yeah. it definitely make sense do only if we could deduce this name in compile time and only if used request this like:

const fn = (a: i32, b: i32, c: i32) => {}
const name  = fn.name;  // could statically reflect as "fn"

function foo(fn: Function<(a: i32, b: i32, c: i32) => void>) {
  const name = fn.name // name is unknown. So output `"anonymous"` or `""`
}

const funcs = [(a: i32, b: i32, c: i32): void => {}, function fn(a: i32): void {}]
const name0 = funcs[0].name // name0 is ""
const name1 = funcs[1].name // name1 is "fn"
funcs[1] = function boo(): void {}
const name1 = funcs[1].name // name1 is "boo"

But yeah, it required dependency analysis, so probably make sense deduce only simple cases

[MaxGraey]

definitely we need extra pass for binaryen)

[dcodeIO]

There's still quite a bit more to do here in order to get (mutable) captures to work. Good next steps would be to complete #1260 to simplify compilation in general, and WebAssembly/binaryen#2586 so we can have a closureify pass on our end that replaces accesses to locals with loads from a function environment to avoid recompiling (and tracking state while doing so). Looks like this'll take a while and overall this PR might become something more like a dev branch, unless you guys want this in master already for your own experiments?

[MaxGraey]

I guess it make split to several stage (logically finished parts) like this one. WebAssembly/binaryen#2586 could be long story and I'm worry we could loose focus so better merge all useful parts asap

[torch2424]

This is awesome, thanks for cleaning this up 😄 👍

[torch2424]

This PR looks great! Super excited for this! Nothing huge / not requesting anything, but does this mean we will also be able to access a function "this" now? :o That would be awesome if so!

[dcodeIO]

There are a couple tests for Function#call, showing what's possible with this arguments. For instance, one can reference an instance method now, pass it around and Function#call it outside of the class by providing this, and the same also works for explicitly annotated this arguments.

[dcodeIO]

Merged as per consensus at the WG meeting. It is likely that there are remaining edge cases that need to be covered, so give it a try with your codebase and let me know if there are issues.

[github-actions]

🎉 This PR is included in version 0.14.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀