This has probably been asked before, but I couldn’t find it from a quick search. Why doesn’t

shift

return

Nothing

? It’s declared as

suspend fun <B> shift(r: R): B

but it could instead be

suspend fun shift(r: R): Nothing

The reason I ask is specifically because of this use case:

var failure: MyFailure? = null
val effect = eagerEffect<MyFailure, String> {
    failure?.let { shift(it) } // "Not enough information to infer type variable B"
    "some result"
}

The fix is to be explicit about the

Nothing

type, which feels super weird:

val effect = eagerEffect<MyFailure, String> {
    failure?.let { shift<Nothing>(it) }
    "some result"
}

I think we had encoded it at some point with

Nothing

but due to how inference worked it was causing issues. It may be possible now. Technically

Nothing

is more correct than any type because after

shift

is called all statements and code afterward is unreachable.

This follows the same pattern as

emptyList

from Kotlin Std. Requiring

A

rather than

Nothing

. IIRC, this is the recommended pattern for inference in Kotlin but I might be mistaken.

That’s a good comparison, I hadn’t considered it 👍. I would also compare it to

error(...)

which isn’t generic and always returns

Nothing

. I can see why

emptyList

uses the generic, but my sense is that

shift

is more similar to

error

than to

emptyList

.

but my sense is that

shift

is more similar to

error

than to

emptyList

Yes, very valid. We hadn't made this comparison before 🤔 Good candidate for Arrow 2.0 improvement as @raulraja said. Feel free to make a ticket for this on the Github repo so we can keep track of it 👍

Thank you for the very complete issue report @Sam 🙏

https://github.com/arrow-kt/arrow/pull/2839 blob sunglasses