Method type declarations violate a es6 specification

[falsandtru]

var es5obj = {
    method: function(){}
};
var es6obj = {
    method(){}
};

'prototype' in es5obj.method; // true
new es5obj.method(); // ok

'prototype' in es6obj.method; // false
new es6obj.method(); // TypeError: es6obj.method is not a constructor

1. Methods cannot call with new keyword.
2. Methods must not have a prototype property.

Class declarations also have same problems.

class C {
    m() {
    }
}
var c: C;
c.m.prototype; // should be a compile error
new c.m(); // should be a compile error

14.3.9 In ECMAScript 2015, the function objects that are created as the values of the [[Get]] or [[Set]] attribute of accessor properties in an ObjectLiteral are not constructor functions and they do not have a prototype own property. In the previous edition, they were constructors and had a prototype property.

http://www.ecma-international.org/ecma-262/6.0/

[DanielRosenwasser]

It says functions created for get/set accessors, which these functions aren't, so it's not clear that this is actually a problem.

[falsandtru]

I could not find a just said section about this behavior. This spec is a closest explanation about this behavior. I thought this is spec because there is this behavior in chrome and firefox. But edge has not this behavior. Sorry, now I cannot know that which implementation reflects a spec.

[DanielRosenwasser]

now I cannot know that which implementation reflects a spec.

That's the fun of it though! 😄

[DanielRosenwasser]

@bterlson any clue what this section corresponds to specifically?

[vladima]

spec is clear and TypeScript should be more strict:
DefineMethod accepts optional parameter functionPrototype that is not passed during PropertyDefinitionEvaluation. Also internally DefineMethod calls FunctionCreate passing in Method as a kind of a function -> for such functions FunctionCreate will pass non-constructor as allocKind to FunctionAllocate and this will not set up [[Construct]] internal method

[falsandtru]

I checked v8 and spidermonkey bug tracker but it seems has not reports about this implementation. As a spec, I also agree with @vladima . Edge seems to have correct implementations.

[DanielRosenwasser]

A fix will need a way to propagate this information along with the signature or type. I think this also would be a breaking change.

[RyanCavanaugh]

Doesn't appear to be worth fixing