Should all new constructors throw when called as a function?

[gibson042]

Every constructor except PluralRules is specified to be callable as a function and supported as such:

$ eshost -se 'typeof Intl === "undefined" ?
  "Intl: undefined" : 
  "Collator,NumberFormat,DateTimeFormat,PluralRules".split(",").map(name => {
    let src = `Intl.${name}()`, result;
    try {
      result = eval(src).constructor.name;
    } catch ( err ) {
      result = err.name;
    }
    return `${src}: ${JSON.stringify(result)}`;
  }).join("\n")'
#### Chakra, JavaScriptCore, SpiderMonkey, V8
Intl.Collator(): "Collator"
Intl.NumberFormat(): "NumberFormat"
Intl.DateTimeFormat(): "DateTimeFormat"
Intl.PluralRules(): "TypeError"

#### XS
Intl: undefined

PluralRules is intentionally different per tc39/proposal-intl-plural-rules#24 , but the door was left open to bring it in line with the rest of spec. Can we reach consensus on whether or not the behavior of Collator, NumberFormat, and DateTimeFormat when called as a function is to be treated as the ECMA 402 convention or as an abandoned legacy, and document accordingly in https://tc39.es/ecma402/#locale-and-parameter-negotiation ?

[ljharb]

It seems better to make them behave like the class keyword, otherwise they won’t be ergonomically self hostable in JS by implementations.

[gibson042]

Well, that's already the case for Intl.{Collator,NumberFormat,DateTimeFormat}, so no matter what such implementations will require code similar to export const NumberFormat = makeCallableFromClass(NumberFormatClass, {name: "NumberFormat"}). This is just a question of how uniformly such code would have to be applied, and also guidance to keep the spec itself predictable through future changes.

[anba]

General consensus (in ECMA-402 and ECMA-262) is that any new built-in constructor should require new. Only pre-ES6 constructors (such as Intl.{Collator, NumberFormat, DateTImeFormat}) support a legacy mode which allows to call them without new.

[sffc]

Is there any action required on this issue? Do we need to document the decision somewhere? Or can we just close it?

[leobalter]

I believe it's fine to close this, we can discuss (IMO directly in a new PR) if we want to add a note to the specs about this, like in #463.