Request for @functionExpression attribute to opt-out from arrow functions

[DZakh]

After the PR #6945 function expressions were changed to arrow functions to improve readability and bundle-size. But it also broke this and arguments binding to the function context, which could be accessed via %raw("this") when needed for some advanced logic.

I need it for rescript-schema, so I think introducing an attribute @functionExpression to opt-out from arrow functions would be a good solution.

[DZakh]

@cknitt @cometkim

[cometkim]

Doesn't @meth work?

We also have @this binding. Afaik, it works for both regular function and FFI

[DZakh]

Doesn't @meth work?

We also have @this binding. Afaik, it works for both regular function and FFI

I'll check

[fhammerschmidt]

Unfortunately, they do nothing here: https://rescript-lang.org/try?version=v12.0.0-alpha.3&code=DYUwLgBAZhC8EAoCUcB8EDeAoCFSQHM5EABMACwEsBnCAQxVnQGEB7AO2tdADphWCCAEQAVENUhCIAamn0kOCIKF0hC3PgjliCEgFtw2hmghtO3EHwHCxEiFNnzF5YaoUBfLEA

or here:
https://rescript-lang.org/try?version=v12.0.0-alpha.3&code=DYUwLgBAZhC8EAoCUcB8EDeAoCEACYAFgJYDOOEokA5nBAIZoQDCA9gHamugB0wr1BACIAKiFKQhEANTSGSCoKH0hC3HgC24QhSoRCdRrHRtO3EHwHCxEiFNnyKhYSoUBfLEA

[cometkim]

I've been exploring this for a bit. rescript-schema seems to be a somewhat special case. I believe most apps have a more explicit and idiomatic solution for binding context without relying on JS this.

ReScript's @this attr support is only compatible with the Js_OO.Callback type, not plain functions because that is the safe way to bind this in "callbacks".

It wouldn't be hard to implement for plain functions either, since there are already primitive implementations, but it would be inherently an unsafe way to write code, so even if we add it, it should be marked as an "unsafe" feature.

[cknitt]

I've been exploring this for a bit. rescript-schema seems to be a somewhat special case. I believe most apps have a more explicit and idiomatic solution for binding context without relying on JS this.

Agreed.

It is important that we get rescript-schema to work with v12 though. Is that not feasible at all without this, @DZakh? If so, then I think we should add the requested feature.

[cometkim]

It seems like there's a lot of this dependency, and removing it and going back to the type t pattern would require a lot of boilerplate and some breaking changes. rescript-schema looks like they're trying to avoid that and provide a more JS-friendly API.

My suggestion is to expose a primitive API instead of attributes. We already have the necessary implementations; %unsafe_to_method. If we provide a proper namespace for it, we can do something like this:

Js.Unsafe.thisCall((this, arg1, arg2) => {
  ...
})
// -> (('this, 'a, 'b) => 'result) => (('a, 'b) => 'result)

[cknitt]

@DZakh Would @cometkim's suggestion work for you?

This is something that you could already test now. %unsafe_to_method is currently exposed as Js_OO.unsafe_to_method.

[DZakh]

Js_OO.unsafe_to_method works for me.

[fhammerschmidt]

Should we document this somehow?