CreateDataProperty is sometimes called with ? but can't throw

[jyasskin]

https://tc39.github.io/ecma262/#sec-internalizejsonproperty, https://tc39.github.io/ecma262/#sec-createdatapropertyorthrow, and a few other places use "? CreateDataProperty(...)", but its definition appears to use just true and false return values to say whether it has succeeded.

This is related to #253.

[shvaikalesh]

O may be an exotic Proxy object with defineProperty trap that throws or breaks invariants. Please, see [[DefineOwnProperty]].

[jyasskin]

In that case, the far more common use of CreateDataProperty() without either ! or ? should be fixed.

[shvaikalesh]

I have checked all (49) usages of CreateDataProperty and there are no calls without ! or ? that may receive Proxy as first argument.

[ljharb]

@shvaikalesh any user-provided object, including "this" values, could be a Proxy.

[shvaikalesh]

All ! / ?-less calls operate on objects created in the same abstract operation or method, mostly arrays or plain objects. They do not get any user-passed stuff.

[shvaikalesh]

may receive Proxy as first argument.

sorry for the confusion, I meant first argument of CreateDataProperty

[ljharb]

Does the HTML spec use that abstract operation at all?

[annevk]

@ljharb it does at the end of https://html.spec.whatwg.org/multipage/infrastructure.html#structuredclone preceded by ?. Although I think in HTML's case it cannot really be called on a proxy so maybe that is wrong?

[ljharb]

There are currently 4 remaining uses of ? with CreateDataProperty in the spec:

• 1 in CreateDataPropertyOrThrow (this is fine)
• 2 in InternalizeJSONProperty, which are intentional and noted as such explicitly
• 1 in OrdinarySetWithOwnDescriptor

Since [[DefineOwnProperty]] could throw, the ? on CreateDataProperty seems correct to me - but I could easily be wrong here. As I see it, this issue should be closed with one of two outcomes:

1. Removal of the ? on the [[DefineOwnProperty]] call inside CreateDataProperty, with an accompanying assertion that O must be an ordinary object (or whatever assertion is applicable to ensure it can't throw), and then removal of the ? on all CreateDataProperty calls
2. No further change is needed.

I'm going to close it assuming the latter; but will happily reopen it if needed.

[annevk]

HTML seems to have changed to use ! meanwhile, FWIW.