Inference doesn't find common base type for union with null or undefined - TypeScript 4.8 regression

[dragomirtitian]

Bug Report

🔎 Search Terms

inference common base type null undefined

🕗 Version & Regression Information

• This changed between versions 4.7.3 and 4.8-beta

⏯ Playground Link

Playground link with relevant code

💻 Code

function equal<T>(a: T, b: T) { }

let v = null!;

// Object types with common base types

type B = { foo: string }
type D = { foo: string; bar: number }

// Error in 4.8 TS can't find common type ❌
// 4.7 T was undefined | B
equal(v as B, v as undefined | D)

// ok T is B ✅
equal(v as B, v as D)
// ok T is B | undefined ✅
equal(v as B, v as B | undefined)

🙁 Actual behavior

In 4.8 TypeScript can't find common base type between B and undefined | D. In 4.7 this used to be B | undefined.

This also happens for literal types and their respective base types as well as for arrays as shown in the 'Other Examples' in the playground link. I didn't include the code as the bug is the same, it's just different instances we found in our code.

🙂 Expected behavior

Typescript finds the type B | undefined for T as it did in 4.7.

[dragomirtitian]

Also reported in Bloomberg feedback #49835

[jakebailey]

This also bisects to #49119.

[jakebailey]

Also interestingly, if you write equal(v as D | undefined, v as B) (so, flip the argument order), T is inferred as D | undefined.

I'm a little confused as to why the argument order matters here, but given that the inference if you swap the arguments is the same in 4.7, it's nothing new.

Playground Link

[jakebailey]

I pinned this down to this part of the PR: https://github.com/microsoft/TypeScript/pull/49119/files#diff-d9ab6589e714c71e657f601cf30ff51dfc607fc98419bf72e04f6b0fa92cc4b8R21147

This part of the code doesn't consider the fact that any of the types could be unions that contain null or undefined, only types which are exactly null or undefined. I have a fix which correctly retains null and undefined which fixes this issue but doesn't make things consistent in argument order (so, same as 4.7). Currently verifying that this is right and that I can't also fix the other case too, since it seems incongruent with other inferences.

[jakebailey]

My fix doesn't seem to be the right one; we should probably discuss the intended inference behavior here.

EDIT: Well, maybe it is, after fixing a few things: Playground Link

[ahejlsberg]

Fixed by #50021.