Infer to partially homomorphic types (such as Pick<T, K>)

[ahejlsberg]

With this PR we improve type inference for partially homomorphic types such as Pick<T, K>. Now, when inferring to a mapped type { [P in K]: X }, where K is a type parameter K extends keyof T, we make the same inferences as we would for a homomorphic mapped type { [P in keyof T]: X }.

Fixes #29765.

[weswigham]

We should probably consider handling unions of indexes here.

[weswigham]

Specifically for along the lines of:

declare function f20<A, B, K extends keyof A | keyof B>(obj: Pick<A & B, K>): T;

which instantiable of the original f20 could easily produce.

[ahejlsberg]

We don't do that for ordinary homomorphic types so I'm not sure why we'd do that here. Also, not clear how we would divide the inferences between A and B.

[weswigham]

We do do that for ordinary homomorphic mapped types. It's why constraintType is passed into inferFromMappedTypeConstraint.

[weswigham]

Like, in this inference function we break down keyof A | keyof B, and we break down (T extends keyof A) | (U extends keyof B), I don't see why we would not also break down T extends (keyof A | keyof B).

[weswigham]

-                     if (extendedConstraint && extendedConstraint.flags & TypeFlags.Index) {
-                         inferToHomomorphicMappedType(source, target, <IndexType>extendedConstraint);
-                         return true;
+                     if (extendedConstraint) {
+                         return inferFromMappedTypeConstraint(source, target, extendedConstraint);

maybe? Probably shouldn't return, since making inferences directly to the constrained type parameter as well as the things referenced in its constraint should both be possible.

[ahejlsberg]

I see what you're getting at. We really need to repeat the whole process for the constraint.