fp-ts
fp-ts copied to clipboard
Improving option.getOrElse type inference
🚀 Feature request
Current Behavior
Today, when using option.getOrElse
, the type of what's inside the option is inferred from the first argument (the or else argument). This leads to typescript easily erroring when returning an array or an option.none (non exhaustive list):
declare const foo: Option<ReadonlyArray<string>>
pipe(
foo,
option.getOrElse(() => []),
)
// ts error: Argument of type 'Option<readonly string[]>' is not assignable to parameter of type 'Option<never[]>'.
To fix this, I can use getOrElseW
but the resulting type will be not that great (readonly string[] | never[]
) or enforce the type like this:
pipe(
foo,
option.getOrElse<ReadonlyArray<string>>(() => []),
)
// -> no error when getOrElse is typed
Desired Behavior
It would be great that the inference would be done from the option that is passed to getOrElse
instead so that this would be valid code:
declare const foo: Option<ReadonlyArray<string>>
pipe(
foo,
option.getOrElse(() => []),
)
Suggested Solution
Use NoInfer
from ts-toolbelt. Here's the implementation if we don't want to add a new dependency:
type NoInfer<A extends any> =
[A][A extends any ? 0 : never]
and use it like this:
export declare const getOrElse: <A>(onNone: LazyArg<NoInfer<A>>) => (ma: Option<A>) => A
Who does this impact? Who is this for?
This can impact any user and would help greatly new users that don't understand why typescript is not happy.
Describe alternatives you've considered
None
Additional context
Your environment
Software | Version(s) |
---|---|
fp-ts | 2.16.1 |
TypeScript | 5.3.3 |
thank you for the issue I'm already using your proposed implementation with the intrinsic ts NoInfer
implementation
I've noticed though that this causes the application without piping, i.e.:
const x = getOrElse(() => 3)(O.some(3))
to infer x as unknown