T wider in lazy branch of T | (() => T) for known T

[pckilgore]

🔎 Search Terms

T | (() => T)
"T | (() => T)" inference
inference of lazy generic type
lazy generic inferred too wide

🕗 Version & Regression Information

• This is the behavior in every version I tried, and I reviewed the FAQ for entries about callbacks and functions and laziness and sum types.

⏯ Playground Link

https://www.typescriptlang.org/play?#code/C4TwDgpgBA8mwEsD2A7AhgGwyAMmgXiADwAqAfFALxQlQA+UAFIwJRUUksBQokUAsiBLho1AN5Q0ALigAjJEgwQ0KKAF8uAY1QBnYFABmq6o2Ay4iVJmx5CRQcMhk2lCmI1cjjCdKjAATgCuEAA0UDoAFkiBGAAmAGJoCBgyAIxqbAD0mVAAIhA6CP5oskqeKMwuFN6SMgHBYZHRcQZJKVAA5MARCDpQsUgFUChI+q3JYQDuPZoRUL1yaLEd6ixZOQBCS0A

💻 Code

type OptionallyLazy<T> = T | (() => T)
type MyType = { a: boolean }
const fn = (t: OptionallyLazy<MyType>) => {}

fn({ a: true, shouldFail: 'it does, good'})
fn(() => ({ a: true, shouldFail: 'this does not fail, which is bad' }))

🙁 Actual behavior

No type error for shouldFail on last line above.

🙂 Expected behavior

Type error for shouldFail on last line above, same as line above it.

[fatcerberus]

() => MyTypeWithExtraProps is assignable to () => MyType, so this is working as intended. The error you see in the first case is due to the excess property check (EPC), which only applies to object literals directly assigned to a type. The error message for EPC is currently misleading; see #55152.

[RobertSandiford]

Discussed in some detail here #55113

Objects are allowed to have additional properties, however special checks are applied in some locations just to catch probable bugs. As said in other places, this is more of a warning than a type error.

Handbook: excess-property-checks

[pckilgore]

Ahh, appreciate the notes. I wish the error message for those checks were more clear they are a lint, and not a type system property. I've been operating under this misunderstanding for years writing TS!

Any known good workarounds? I guess I can write a Lint rule to require annotation of a lazy value? This caused a very insidious misspelling bug.

Thanks again.

[RobertSandiford]

Ahh, appreciate the notes. I wish the error message for those checks were more clear they are a lint, and not a type system property. I've been operating under this misunderstanding for years writing TS!

Any known good workarounds? I guess I can write a Lint rule to require annotation of a lazy value? This caused a very insidious misspelling bug.

Thanks again.

I don't see any option other than explicit typings. There is an open suggestion (#12936) for Exact types that are always checked this way, maybe one day that will happen.

[fatcerberus]

I wish the error message for those checks were more clear they are a lint, and not a type system property. I've been operating under this misunderstanding for years writing TS!

You’re certainly not alone. Thats why I’m glad for #55152.