Need a dummy overload to improve inference

[stephanedr]

TypeScript Version: 3.x, 4.0.0-beta, nightly

Search Terms: dummy fake overload infer

Code

class FieldEvent<T extends Field> {
    constructor(readonly field: T, readonly message: string) { }
}

class Field {
    listen<T extends FieldEvent<Field>>(type: new (...args: any[]) => T, listener: (event: T) => void): void { }
}

class BeforeSetEvent<T extends number | string> extends FieldEvent<ValueField<T>> {
    constructor(field: ValueField<T>, public newValue: T) { super(field, ""); }
}

class AfterSetEvent<T extends number | string> extends FieldEvent<ValueField<T>> {
    constructor(field: ValueField<T>, readonly oldValue: T) { super(field, ""); }
}

class ValueField<T extends number | string> extends Field {
    constructor(public value: T) { super(); }

    // listen<U extends FieldEvent<Field>>(type: new (arg: this) => U, listener: (event: U) => void): void;
    // listen<U extends FieldEvent<Field>>(type: new (...args: any[]) => U, listener: (event: U) => void): void;
    listen<U extends FieldEvent<Field>>(type: new (...args: any[]) => U, listener: (event: U) => void): void {
        super.listen(type, listener);
    }
}

let field = new ValueField(0);
field.listen(FieldEvent, (event) => event.field);                           // event: FieldEvent<any>
field.listen(BeforeSetEvent, (event) => { event.field; event.newValue; });  // event: BeforeSetEvent<any>
field.listen(AfterSetEvent, (event) => { event.field; event.oldValue; });   // event: AfterSetEvent<any>

Expected behavior:
event to be better inferred than any.

Actual behavior:
event inferred as any.

But if we uncomment the 2 overloads, then event is correctly inferred:

field.listen(FieldEvent, (event) => event.field);                           // event: FieldEvent<ValueField<0>>
field.listen(BeforeSetEvent, (event) => { event.field; event.newValue; });  // event: BeforeSetEvent<0>
field.listen(AfterSetEvent, (event) => { event.field; event.oldValue; });   // event: AfterSetEvent<0>

What's surprising (so the reason I consider it as a bug):

• The 1st overload doesn't match any signature.
• But it influences the way the other overload is inferred.

Note that if we replace (arg: this) by () or another type, then:

field.listen(FieldEvent, (event) => event.field);                           // event: FieldEvent<Field>
field.listen(BeforeSetEvent, (event) => { event.field; event.newValue; });  // event: BeforeSetEvent<string | number>
field.listen(AfterSetEvent, (event) => { event.field; event.oldValue; });   // event: AfterSetEvent<string | number>

Playground Link: Playground Link

Related Issues: I first posted #36226 but was not able to solve my issue, until I found the (arg: this) trick.

[RyanCavanaugh]

I'm not really sure if this is a "bug" per se, but agree that it's not ideal. A straightforward fix without worse side effects may not exist