New warning with 2.13.17: a type was inferred to be kind-polymorphic `Nothing` to conform to `F[_]`

[durban]

Reproduction steps

Scala version: 2.13.17

package com.example

trait Thing[F[_]]

object Repro {
  def foo[F[_]](i: Int)(implicit F: Thing[F]): F[Unit] = ???
  def bar[F[_] : Thing](i: Int): F[Unit] = foo(i) // <- warning here
}

[warn] .../Repro.scala:...:44: a type was inferred to be kind-polymorphic `Nothing` to conform to `F[_]`
[warn] Applicable -Wconf / @nowarn filters for this warning: msg=<part of the message>, cat=lint-infer-any, site=com.example.Repro.bar
[warn]   def bar[F[_] : Thing](i: Int): F[Unit] = foo(i)
[warn]                                            ^

Problem

I don't see a reason for this warning, but in case there is a good reason, I would definitely like to understand it.

Note: if I do foo[F](i) instead of foo(i), the warning disappears.

Note, that this is probably the same warning as reported here: https://contributors.scala-lang.org/t/scala-2-13-17-release-planning/6994/20

[durban]

A similar but different warning is also new with 2.13.17:

package com.example

trait Base1
class MyClass1[F[_]](val obj: MyClass2[F], val fa: F[Unit]) extends Base1

sealed trait Base2
class MyClass2[F[_]] extends Base2 {
  def unit: F[Unit] = ???
}

object Repro {
  def foo[F[_]](obj: Base2): Base1 = obj match {
    case obj: MyClass2[_] =>
      new MyClass1(obj, obj.unit) // <- warning here
  }
}

[warn] .../Repro.scala:...:7: a type was inferred to be kind-polymorphic `Any` to conform to `F[_]`
[warn] Applicable -Wconf / @nowarn filters for this warning: msg=<part of the message>, cat=lint-infer-any, site=com.example.Repro.foo
[warn]       new MyClass1(obj, obj.unit)
[warn]       ^

Note, that if I remove the second argument of MyClass1's constructor (i.e., if I do new MyClass1(obj)), the warning disappears.

I don't know, if the reason for this is the same as for the original report.

[gaeljw]

I have the same error with a concrete case when upgrading to 2.13.17, minified below:

//> using scala 2.13.17
//> using dep com.softwaremill.sttp.client::core:2.3.0

import sttp.client._

implicit val sttpBackend: SttpBackend[Identity, Nothing, NothingT] = ???

val someUrl = "http://example.com"
val response: Identity[Response[Either[String, String]]] = basicRequest.post(uri"$someUrl").send()

$ scala-cli -Xlint:infer-any test.sc

[warn] a type was inferred to be kind-polymorphic `Nothing` to conform to `F[_]`
[warn] val response: Identity[Response[Either[String, String]]] = basicRequest.post(uri"$someUrl").send()
[warn]                                                            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

[gaeljw]

For the record, here are the changes I had to make in my codebase:

- val response: Identity[Response[Either[String, String]]] = basicRequest.post(uri"$credentialsWsUrl").send()
+ val response: Identity[Response[Either[String, String]]] = basicRequest.post(uri"$credentialsWsUrl").send[Identity]()

- implicit val sttpBackendMocked: SttpBackendStub[Identity, Nothing, Nothing] = SttpBackendStub.synchronous
+ implicit val sttpBackendMocked: SttpBackendStub[Identity, Nothing, Nothing] = SttpBackendStub.synchronous[Nothing]

[som-snytt]

@gaeljw It's a lint. I'd expect people to either -Xlint:-infer-any or equivalent @nowarn, rather than change code -- unless the change is warranted or desired.

[gaeljw]

That's a possibility as well, for sure. I personally prefer to have as less as possible nowarn. I assume warnings (treated as errors in our codebases) are here for a good reason and prefer to make the compiler happy with additional code in such cases.

I'll admit that like OP I don't really understand what's behind the warning in this case though.

[gaeljw]

For the record, it's "caused by" scala/scala#11053 in order to fix #12044

[Prillan]

I'm having a really hard time understanding why this is a warning? It looks like our entire code base is doing something wrong suddenly. Is it? And if it's doing something wrong, could someone maybe explain what's wrong with inferring the type in situations like:

import cats.syntax.all._

object example {
  def f(x: Int, y: Int): Either[String, Int] = ???

  def thisIsApparentlyWorthyOfAWarning(vals: List[Int]): Either[String, Int] =
    vals.foldLeftM(0)(f)
}

a type was inferred to be kind-polymorphic Nothing to conform to G[_]

[joroKr21]

Also happens here: typelevel/cats-tagless#644

I'm not sure how a new warning fixes #12044 - if I understand the issue correctly, the problem is that type variable bounds were collected from all implicit conversions instead of only the one that is applicable. The correct solution would be to find a way to reset the bounds at the appropriate place but that's really hard.