Different behavior between inline type and type alias

[ozum]

TypeScript Version: 3.4.0-dev.20190131

I'm trying to find base types of some class properties. For example, for a given X[], trying to find X.

I use same definition inline and as a type alias. Type alias has error whereas inline type works.

Search Terms: Type alias, different behavior, conditional

Code

Below is simplified version of the actual code. Generics A1 and A2 in class method whereRelated are equal, except A1 uses inline type and A2 uses a type alias.

interface RelationFields {
  x: A;
  y: A[];
  z: A[];
}
type Name = keyof RelationFields;
type ShouldA<RF extends RelationFields, N extends Name> = RF[N] extends A[] ? RF[N][0] : never;

class A {
  x: A;
  y: A[];
  z: A[];

  whereRelated<
    RF extends RelationFields = RelationFields,
    N extends Name = Name,
    A1 extends A = RF[N] extends A[] ? RF[N][0] : never, // Works
    A2 extends A = ShouldA<RF, N> // Type is same as A1, but is not assignable to type A
  >(): number {
    return 1;
  }
}

Expected behavior:

Either both A1 and A2 work or both have errors.

Actual behavior:

A1 works, but A2 has an error: ... Type 'RelationFields[N][0]' is not assignable to type 'A'.

Playground Link: See here

Related Issues: #15973 seems similar, but I can not be sure. If this is an expected behavior, an advise how to use a type alias is appreciated.

Thanks,

[AnyhowStep]

I've run into similar problems before. Not quite your exact situation, though.

My hack-y workaround is to add a comment saying I'm about to use a workaround, and then do,

//For some reason, I need to surround this with Extract<>
//TODO Investigate
A2 extends A = Extract<ShouldA<RF, N>, A>

This is unrelated but is there a reason you use RF[N][0] instead of RF[N][number]?

[ozum]

@AnyhowStep Thanks for the tip for Extract. I will wrap original alias with Extract until finding a direct solution:

type ShouldA<RF extends RelationFields, N extends Name> = Extract<RF[N] extends A[] ? RF[N][0] : never, A>;

[jack-williams]

I'm wondering if #29698 is also fundamentally related to this---inlining Record produces the correct errors.

[weswigham]

@jack-williams nope.