--noImplicitAny codefixes infer anonymous object types despite appropriate interfaces in (or out of) scope

[JoshuaKGoldberg]

TypeScript Version: 3.3.0-dev.20181212

Search Terms: noimplicitany infer code fix codefix quick suggested

Code

interface IBox {
    x: number;
    y: number;
}

const shiftBox = (box) => {
    box.x += 1;
    box.y += 1;
}

const box = {
    x: 0,
    y: 0,
};

shiftBox(box);

Expected behavior:

--noImplicitAny suggested fix on the box parameter:

const shiftBox = (box: IBox) =>

Actual behavior:

const shiftBox = (box: { x: any; y: any; }) => {

Related Issues: #13243 (parent tracking quick fixes) and #15114 (discussion on inference difficulties)

If there is an interface available that can satisfy an inferred type in a --noImplicitAny code fix, can we use that? Perhaps with an ordering of possibilities like:

1. Interfaces already available in the file, by how small they are
2. User-defined interfaces that could be imported, by how distant the file is
3. Module types already imported in a user file, by how distant the nearest import is

...where, if multiple interfaces could satisfy the best possibility of those three, we choose the one with the fewest fields?

In code bases that don't explicitly type when unnecessary (e.g. : IBox for variables), I'm finding the --noImplicitAny fixes to be a bit useless for anything other than primitives.

[sandersn]

Non-primitive inference is indeed a known hole in inferFromUsage.

I haven't published a roadmap for the inferFromUsage codefix, but I prototyped a fix for this last year. At least, I defined the way to choose the "closest type". The part I did not do was create a reliable way to find types that were in scope from a particular definition. The closest-type definition also didn't take scopes into account.