Normative: Remove [[VarNames]] from the global

[syg]

I noticed this due to a recent test262 test from @shvaikalesh: tc39/test262#3914

The explainer has been pulled into a proposal repo: https://github.com/tc39-transfer/proposal-redeclarable-global-eval-vars

This PR still serves as the spec draft.

[shvaikalesh]

@syg This is a totally awesome proposal! Would love to implement.

[bakkot]

To be clear, the only observable change here is that if you have something like

<script>
eval('var x;')
</script>
<script>
let x;
</script>

that is current an error (per spec), and with this PR it would not be?

[syg]

Re: #3226 (comment)

Yep, that's the intention. Same for functions introduced by sloppy eval also no longer erroring. That is, the following would no longer error with this PR.

<script>
eval('function x() {}')
</script>
<script>
let x;
</script>

Edit: Another importance difference is also that you shadow the binding. So in the function example above, x after let x evaluates will now be undefined, while globalThis.x remains the function.

This may be an unintuitive and undesirable result, but my feeling with that is, as with many things, "just don't do that".

[syg]

At the February 2024 TC39 it was decided that this PR be moved into its own proposal (which I will do after plenary).

[shvaikalesh]

Implemented in WebKit/WebKit#30121.

[syg]

Rebased.

[bakkot]

jmdyck points out that this undefines sec-hasvardeclaration, but I don't think there's really a good place for that to live now, so I'm OK with that.

[michaelficarra]

LGTM