Editorial: Allow ACs in Async{Function,Block}Start

[syg]

There is no use of Abstract Closures as async function bodies in ecma262 itself. This PR is a prereq to allow the JSPI specification to pass in an AC.

[bakkot]

LGTM but I want to see the spec text which uses this before we merge.

[syg]

LGTM but I want to see the spec text which uses this before we merge.

You mean in 262? As explained in #3452 (comment) this is a prereq for the JSPI spec. You can see a draft here.

[bakkot]

I meant the JSPI draft, yeah. Seems fine to use AsyncFunctionStart in this way there, though some of the other stuff (in particular the execution context map) is a bit suspicious.

[syg]

Seems fine to use AsyncFunctionStart in this way there, though some of the other stuff (in particular the execution context map) is a bit suspicious.

I actually like that in that it programmatically makes explicit which are the pausable ECs from JSPI's perspective. If you recall I don't want 262 to grant any embedder of 262 the power to stack switch any stack sandwiched between two ECs. And that EC map is like signing a waiver "I promise to only stack switch ECs that sandwich wasm stacks, which I've thought about and know to be switchable".

Edit: Oh also it traps for reentrant use cases, so it actually has some runtime semantics.