WeakSet<T> should be WeakSet<T extends object>

TypeScript Version: nightly (2.3.0-dev.20170324)

Code

const s = new WeakSet<string>();
s.add("foo");

Expected behavior:

An error, since this will fail at runtime.

Actual behavior:

No error.

[falsandtru]

I did it, but @mhegazy reverted that fix in #15278. His reverting doesn't comply with the spec of ecma. We should comply with the spec.

@mhegazy says object constraints of WeakSet causes errors in @types/lodash, but WeakMap is not. Why? lodash has correct constraints for WeakMap like https://github.com/DefinitelyTyped/DefinitelyTyped/blob/master/types/lodash/index.d.ts#L12898. So WeakMap already have object constraints. WeakSet also should be so.

[mhegazy]

We reverted the change to avoid breaking @types/lodash. we should reply it, but first we need either 1. remove these definitions from @types/lodash alltogahter, or 2. relax the rule on the compiler side that all declarations have the same constraint.

The definition in lodash is at the end of the file, not visible on github view:

// Backward compatibility with --target es5
declare global {
    interface Set<T> { }
    interface Map<K, V> { }
    interface WeakSet<T> { }  // this breaks if lib.es6.d.ts has WeakSet<T extends object>
    interface WeakMap<K extends object, V> { }
}

[falsandtru]

Now that problem seems to have gone away.

[mhegazy]

so what happened to lodash use of this API?

[falsandtru]

A last fix is DefinitelyTyped/DefinitelyTyped#21502 and #19756. We need to accept some breaking changes, or never fix this bug.

[mhegazy]

As i mentioned earlier, we need to first:

relax the rule on the compiler side that all declarations have the same constraint.

[falsandtru]

Is there a proposal issue of that?

[mhegazy]

nope.

[falsandtru]

Can you make it to resolve this issue?

[mhegazy]

feel free to file a new issue for it.

[falsandtru]

Why you don't do it?

[mhegazy]

here you go #20018

[falsandtru]

Thanks.