feat: support first class function via builtin function

[HerrCai0907]

This is the first pr to support first class function.
Firstly we need to support allocating function object in heap instead of data area.

[CountBleck]

Are you sure we shouldn't wait until we use Wasm GC?

[HerrCai0907]

Are you sure we shouldn't wait until we use Wasm GC?

I think we can do it separately. First class is based on object module, it don't and shouldn't rely on wasm GC or __new. It only means we have a Function class and can manage it as normal object.

[CountBleck]

By "first class function", do you mean closures? Those might be easier to implement in Wasm GC, since the heap doesn't have to be dealt with manually, and closure objects for child blocks would include the parent closure objects (at least the way I see it).

[HerrCai0907]

Those might be easier to implement in Wasm GC, since the heap doesn't have to be dealt with manually, and closure objects for child blocks would include the parent closure objects (at least the way I see it).

We have memory allocation abstraction, and then we have GC object abstraction based on memory allocation. Those thing can be replaced by
WASM GC. But closures should be something based on those thing. When we design closures, we should not consider where and how to create object. It should be same as new XXX().

[CountBleck]

It should be same as new XXX().

That's actually my point. The XXX is backed by a Class, which is backed by a ClassPrototype, which is backed by an AST node (ClassDeclaration). Sharing code with this would be difficult (at least, I think so), and not sharing code would add another arcane part of the compiler that'd be harder to maintain or iterate on (as far as I can see).

Meanwhile, with Wasm GC, we could just make GC structs without touching AS's type system (which would also be simplified). As a result, it would (theoretically) be much more convenient to write, debug, and maintain.

[CountBleck]

Also, the current test should already work: see here.

[HerrCai0907]

Also, the current test should already work: see here.

Yes. This pr will not support closure. It only wants to support store function instance in heap instead of in data section. I think it will be the first step to support closure.

[CountBleck]

I don't quite understand what you mean.

[CountBleck]

Oh, do you mean creating a new Function object for each returned inner function, instead of a single object?

[HerrCai0907]

This is binaryen bug which fixed in latest binaryen.

# compiler/first-class-function

- compile debug
  SUCCESS (251.300 ms)

- compare fixture
  SUCCESS (1.495 ms)

- compile release
  SUCCESS (397.126 ms)

- compare fixture
  SUCCESS (0.117 ms)

- instantiate debug
  SUCCESS (1.620 ms)

- instantiate release
  ---
  RuntimeError: table index is out of bounds
      at wasm://wasm/651ca9aa:wasm-function[14]:0xc45
  FAILURE (1.120 ms)
      at wasm://wasm/651ca9aa:wasm-function[13]:0xb6d

  ---

[lcswillems]

@HerrCai0907 Great news you are working on this!! 🙏 @CountBleck Is Wasm GC ready?

[CountBleck]

Is Wasm GC ready?

Nope. Keep in mind that I'm a procrastinating 10th grader who has to deal with an annoying amount of English homework :P

I'll try to get in more work towards the Wasm GC transition in my spare time, but the progress won't really be visible until I make a PR to Binaryen.