`unique symbol`s from the global `SymbolConstructor` widen way too eagerly

[weswigham]

Given

const a = Symbol.toStringTag;

class F {
    [a](){ return "" }
}

(new F())[a];

we currently issue a Element implicitly has an 'any' type because expression of type 'symbol' can't be used to index type 'F'.(7053) error on the access.

There should be no errors, as with

const a = Symbol();

class F {
    [a](){ return "" }
}

(new F())[a];

since Symbol.toStringTag is a unique symbol just like what Symbol() makes.

[fatcerberus]

Wait, why is a well-known symbol typed as unique symbol? That feels wrong to me.

[weswigham]

Because it's a specific, unique symbol. As a matter of fact, we don't even have the concept of a "well-known" symbol anymore - unique symbols subsumed them as the more general concept (the term well known no longer appears anywhere in the checker's code, just some old comments!).

[fatcerberus]

I guess I just always saw unique symbol as this special nominal type that every time it's mentioned, refers to a distinct type, so it feels weird to apply that to values that are predefined by the runtime. IOW unique symbol implies to me a value that can't be forged