constructors that return result<>

[OmarTawfik]

Today, the following syntax is illegal, since constructors have no return types:

resource blob {
    constructor(data: list<u8>) -> result<blob, error-type>;
}

However, this is a perfectly valid use case for languages like JS and C++, as result<> return type are lowered into blob directly, using exceptions to propagate error-type values if they occur. This is what happens already with normal methods that return result<>.

Thoughts? I wonder if it is possible to enable this use case. Without it, APIs are forced to use alternative static methods like create: static func(), which is not standard, and hurts readability/discoverability.

Moved from bytecodealliance/wit-bindgen#1114

[lukewagner]

Totally agreed; we should definitely do this. I believe #285 makes this same point, so perhaps we can dup this to that?

Spec-wise, the change is very simple; we just relax the validation rules for [constructor]...-named functions to also allow result. Most of the work would be in wit-bindgen and related tools. Let me know if anyone is keen to work on this!

[OmarTawfik]

Thank you. I missed #285. I can close this one as a dup.