Normative: Verify if Binding is preserved for obj's SetMutableBinding

[leobalter]

Verify binding existence in the object's SetMutableBinding on Strict Mode.

Closes #2093
Ref #467
Ref tc39/test262#427

[bakkot]

Note that the HasBinding call is observable. For example, consider

let scope = { x: 1 };
let handler = {
  has(o, p) {
    if (p === 'x') {
      console.log('hit', Reflect.has(o, p));
    }
    return Reflect.has(o, p);
  },
};
with (new Proxy(scope, handler)) {
  (function () {
    'use strict';
    x = (delete scope.x, 2);
  })();
}

What should this print / do? In Chrome it prints hit, false and throws; in Firefox it prints hit, true and does not throw; in Safari it prints hit, true; hit, false and throws.

For a trickier example, consider

let scope = { x: 1 };
let first = true;
let handler = {
  has(o, p) {
    if (p === 'scope') {
      return false;
    }
    console.log('hit');
    if (first) {
      first = false;
      return true;
    }
    return false;
  },
};
with (new Proxy({ x: 1 }, handler)) {
  (function () {
    'use strict';
    x = 2;
  })();
}

where here the call to HasBinding is the thing which causes the property to not exist!

[anba]

Can't we perform this check as part of 8.1.1.2.5 SetMutableBinding as proposed in the es-discuss thread? That way it nicely lines up with other strict mode checks in environment records, like for example 8.1.1.2.6 GetBindingValue, step 4 or 8.1.1.1.5 SetMutableBinding, step 2.

[leobalter]

@anba @bakkot how does it look now?

We are still caught by Proxies [[HasProperty]] that can falsely report as an existing property. IMO, it sounds like not a problem for the specs but user land.

[anba]

LGTM!

FWIW the proxy issue actually also applies to 8.1.1.2.6 GetBindingValue.

[leobalter]

This has reached consensus today. We need to update the current tests - starting with those pointed out at tc39/test262#427 - and also verify observable implications with Proxies.

[syg]

lgtm editorially