Missing @implicitNotFound for ApplicativeError and MonadError

[satorg]

Check it out:

scala> import cats.syntax.all._

scala> none[Int].raiseTo[Option]
-- Error:
1 |none[Int].raiseTo[Option]
  |                         ^
  |                      Could not find an instance of Applicative for Option

But in fact, Applicative[Option] DOES exist in the scope.

The error message for the above should look like: "Could not find an instance of ApplicativeError[Option, Int]" instead.
The reason why it is incorrect is that ApplicativeError does not have @implicitNotFound annotation applied and compiler picks one up from its parent.

[armanbilge]

These errors frequently trip me up, and not just in Cats proper. E.g. if something in the CE hierarchy is missing then it says you are missing Monad (and not Concurrent or Async).

[djspiewak]

Scala 3 inherits these definitions while Scala 2 does not. This has generated… no small amount of confusion. :-) Basically the consequence of this is the fact that in any typeclass hierarchy which has one @implicitNotFound needs to have them on all classes.

[satorg]

Seems that Scala 2 also uses the inherited annotation (notice the sentence in parentheses):

could not find implicit value for parameter F: cats.ApplicativeError[Option,Int] (Could not find an instance of Applicative for Option)

But the overall error message is a way easier to decipher. At least it provides some clue where to start looking at.

[joroKr21]

Why does Applicative need an @implicitNotFound with such a generic message? How is it different than the default?

[joroKr21]

And is that an argument against inheritance of @implicitNotFound messages 🤔

[armanbilge]

Yes, I agree, I'm pretty sure we don't need these @implicitNotFound if the messages are so generic.

I experimented in a scastie (2.13) and it seems the default is to use the FQCN?
https://scastie.scala-lang.org/TUftkqyJSyyOg0XkkcntRw

trait Applicative[F[_]]
trait Foo[A]
implicitly[Applicative[Foo]]
// could not find implicit value for parameter e: Playground.Applicative[Playground.Foo]

Update: same on 2.12. Maybe some ancient scalac was much less friendly about these.

[armanbilge]

Lol, apparently this inheritance issue was known at the time! #581 (comment)

(The bad part here is that I think this would require annotating every type class with some kind of message. Either that, or wiring it up to Simulacrum somehow.)

[joroKr21]

Hmm so it's coming from simulacrum 🤔 , but the message never included the suggestion to import cats.implicits which is now obsolete anyway. 😄

[djspiewak]

Let's just remove the annotations. I don't think they're carrying any weight.

[armanbilge]

Does that mean making a change in simulacrum, and re-running it here? Not sure how that stuff works.

[djspiewak]

We can just do it manually. Cats no longer uses simulacrum because the latter isn't compatible with Scala 3. Travis wrote a simulacrum-scalafix which kind of does the same thing, but I don't think it is run on an ongoing basis. I would need to check.