[Errors] Empty object contextually typed by type with indexer

[mhegazy]

interface Foo { a }
interface Bar { b }

interface Object {
    [n: number]: Foo;
}

interface Function {
    [n: number]: Bar;
}

var o = {};
var f = () => { };

var v1: {
    //!!! Cannot convert '{}' to '{ [x: number]: Foo; }':
    //!!!   Numeric index signature is missing in type '{}'.
    [n: number]: Foo
} = o;  
var v2: {
    //!!! Cannot convert '() => void' to '{ [x: number]: Bar; }':
    //!!!   Numeric index signature is missing in type '() => void'.
    [n: number]: Bar
} = f;  

Expected: no errors

Actual: errors in the comments above

Verify test case:
tests/cases/compiler/augmentedTypeBracketAccessIndexSignature.ts
tests/cases/conformance/types/members/objectTypeWithCallSignatureHidingMembersOfExtendedFunction.ts

[JsonFreeman]

The actual behavior here matches the spec. It appears to be by design.

[JsonFreeman]

Another interesting example where we don't error, but probably should:

var v: { [s: string]: string} = {
p: 3, // Expect an error here
q: null
}

This is a result of best common type losing information, and the nontransitivity of assignability.

[sophiajt]

What's the action here? If we're aligned to spec, shouldn't this just be closed as by design?

[JsonFreeman]

As I recall from discussion, the spec needs to be updated to align with the current behavior. Specifically, I think section 3.8.1 needs to not pull in call/construct/index signatures for the augmented types.

And it is a breaking change.

[sophiajt]

I'm confused. You mention earlier that the "actual behavior here matches the spec. It appears to be by design." If so, is there a spec change?

[JsonFreeman]

Yeah my previous comments were wrong because I misunderstood what was going on. Ignore those. It needs a spec change per my recent comment.

[sophiajt]

Before we make this change, please confirm that the change does not impact real world code before we make the spec change.

[JsonFreeman]

Reopening this because the validation on real world code was not done yet.

[ahejlsberg]

Doesn't break any real world code we know of.

[RyanCavanaugh]

Is this fixed, or by design?

[ahejlsberg]

Fixed, in that it is a change from the old compiler. Spec has been updated to reflect behavior of the new compiler.