Constraints which are generic mapped types should have a constraint of the same mapping over their generic's constraint

[weswigham]

Code

function f<
    T extends { member: string },
    TParam extends Readonly<T>
>(param: TParam, mutableSet: T): TParam {
    const x: string = param.member;
    return param;
}

Expected behavior:
Code works

Actual behavior:
Property member does not exist on type TParam.

When you call f, you must pass a first (and second) argument which provide a member named member (this is enforced), so it's common sense that within f you should always have access to the field of TParam named member. It seems like the issue today is that when calculating a generic's constraint, we don't consider a mapped type's generics' constraints when calculating its constraint.

[jakebailey]

Just looking at old bugs; this one was fixed all the way back in eeabd52 as part of #26698.