Specify Symbol.prototype methods as [[Writable]]: false

[jdalton]

I'm not sure if this is mentioned in another place in the spec, I didn't see it, but I noticed all implementations have symbol methods like Symbol#toString as [[Writable]]: false.

Object.getOwnPropertyDescriptor(Symbol.prototype, 'toString')
// {writable: true, enumerable: false, configurable: true}

but

var s = Symbol();
s.toString = function() { return ''; };
s.toString(); // => "Symbol()"

[domenic]

This is just the usual confusion between wrapper objects and primitives. It's the same as

var s = false;
s.toString = function() { return ''; };
s.toString(); // => "false"

line 2 creates a useless wrapper object that is immediately discarded. It doesn't set {Symbol,Boolean}.prototype.toString.

[jdalton]

Ah yes, my bad.

var s = Object(Symbol());
typeof s; // => "object"
s.toString = function() { return '' };
s.toString(); // => ""

then

var o = {};
o[s] = 1;
Object.keys(o);
// => []
Object.getOwnPropertySymbols(o)
// => [Symbol()]

Kinda neat it works as an object too.

[claudepache]

line 2 creates a useless wrapper object that is immediately discarded

It's less cumbersome to think as: Primitives are frozen.

[domenic]

It's less cumbersome to think as: Primitives are frozen.

Except "modifying" them doesn't throw in strict mode.

[claudepache]

Except "modifying" them doesn't throw in strict mode.

It should. (It does in JSC and V8. If it doesn’t in ECMA262, I consider it as a bug.)

[allenwb]

@claudepache

Except "modifying" them doesn't throw in strict mode.

It should. (It does in JSC and V8. If it doesn’t in ECMA262, I consider it as a bug.)

No, Wrapper objects are not created frozen and hence there should be no strict mode errors associated with creating own properties on them. This is legacy behavior for Number, Boolean, String and modify it would be a web breaking change. As a new kind of primitive wrapper, it was conceivable that Symbol wrapper could be frozen. But that would then have created an internal inconsistency between Symbol wrapper objects and the other pre-existing kinds of wrapper objects..

[anba]

there should be no strict mode errors associated with creating own properties on them.

Except in most cases PutValue ends up in 9.1.9 [[Set]], which returns false if the receiver is a primitive value and the property is not present (or a data property). And then a TypeError is thrown. (I think this is what @claudepache meant.)

[claudepache]

I meant that it is more useful that it throws, so that the programmer is warned that their intention could not be satisfied. The wrapper object dance is a legacy specification detail. In practice, what the programmer sees is what @jdalton reported: attempt to define an own property on a primitive fails, just as for frozen objects.

[allenwb]

But 9.1.9 returns that false to step 6.c of 6.2.3.2 which only throws for strict references.

@claudepache is correct it would be better if it could always throw, but that isn't the legacy semantics outside of strict mode. But strict mode does give the more desirable error behavior.