TSX files doesn't suggest component properties

[nschurmann]

This are the only props suggested in the autocomplete, @types for react router are installed, so it should suggest at least path and component.

[DanielRosenwasser]

Repro:

import * as React from "react";
import { Route } from "react-router"

class App extends React.Component {
	render() {
		return (
			<div>
				<Route path="" /*1*/ />
				<Route exact /*2*/ />
				<Route exact={true} /*3*/ />
			</div>
		);
	}
}

For the 3 markers, (1) has a completion that includes "" which is totally nonsensical, and all 3 exhibit the explained behavior.

[weswigham]

Non-JSX repro of the same issue:

interface Foo {
    a?: number;
    b?: number;
}

function f<T extends Foo>(x: T) { }

f({/*1*/});

branch with a candidate fix (so I don't forget) is here. Currently it has some issues with the ThisType<T> tag that I need to investigate before opening a PR and seems to regress services behavior for

window.addEventListener("error", event/*1*/ => {
	event.filename;
});

but not

interface EventListenerWithOverloads {
	addEventListener(type: "error", listener: (this: Window, ev: ErrorEvent) => any, options?: boolean | AddEventListenerOptions): void;
	addEventListener(type: "beforeunload", listener: (this: Window, ev: BeforeUnloadEvent) => any, options?: boolean | AddEventListenerOptions): void;
}

declare var ev: EventListenerWithOverloads;

ev.addEventListener("error", event/*1*/ => {
	event.filename;
})

which I'll need to look into as well. (The first uses a mapped type and an index, while the second is overloads. It's possible the first won't work right without full generic inference, which is exactly what I'm trying to avoid because of the other issues it causes...)

I've long held that getContextualType needed to not be reentrant with inference, and with that branch, that is the case. The effect is that completions are a lot better (they come from the uninferred signatures now) and that literal types are retained in calls with errors and that an inferred boolean type can no longer contextual type itself. We've had a bunch of issues that essentially boil down to this same underlying flaw in contextual typing - I'll need to gather them all up along with their test cases.

[weswigham]

This has been fixed since #36556 went in.