fix: super.x in a loop

[liuxingbaoyu]

Q	A
Fixed Issues?	Fixes #15945
Patch: Bug Fix?	√
Major: Breaking Change?
Minor: New Feature?
Tests Added + Pass?	√
Documentation PR Link
Any Dependency Changes?
License	MIT

[babel-bot]

Build successful! You can test your changes in the REPL here: https://babeljs.io/repl/build/55697/

[nicolo-ribaudo]

I think it's ok to start emitting let for the loop case even if it's ES6, given that anybody using super will also be compiling let/const anyway. However, would it be possible to only emit let if we are actually in a loop, to reduce risk?

[lightmare]

Checking direct parent is not sufficient. For example, wrap the loop body in if (true) { ... } and you get a var again.

[liuxingbaoyu]

Fixed, thank you!

[lightmare]

Found another issue, probably related to plugin order. plugin-transform-block-scoping will simply rewrite the let to a var, unless it's already capturing the scope because of something else.

This fails, but if you uncomment the bar() method, it passes: REPL

//---

const objects = [];

for (const proto of [{x: 0}, {x: 1}])
  if (true) {
    objects.push({
      __proto__: proto,
      p: proto,
      foo() {
        return super.x
      },
/*
      bar() {
        return proto.x
      },
//*/
    });
  }

for (const o of objects) {
  console.log(o.p.x === o.foo(), o.p.x, o.foo());
}
for (const o of objects) {
  console.assert(o.p.x === o.foo(), o.p.x, o.foo());
}

[liuxingbaoyu]

Amazing find, thank you!
This is actually a bug in plugin-transform-block-scoping, it's probably a regression, I'm surprised no one has reported it. :)
I will investigate.

old babel

var objects = [];
var i = 0;
for (var proto of [0, 1 ,2]) if (true) {
  let _obj = i;
  i++;
  
  objects.push({
    foo() {
      return _obj;
    }
  });
}

console.assert(objects[0].foo()===0,objects[0].foo());

repl

[nicolo-ribaudo]

Awesome, thanks @lightmare for the review :)

@liuxingbayou Do you mind splitting 8eda7f0 into a separate PR so that we have to separate changelog entries? 🙏

[nicolo-ribaudo]

Could you rebase? :)

[nicolo-ribaudo]

I opened #16058 to keep track of the other regression.