Implicit search ignores implicit definitions for a subtype imported from inside a parameter value

[yoohaemin]

Reproduction steps

Scala version:
2.13.10 (https://scastie.scala-lang.org/DXf5ct0bSCOWZkyrZtmDQQ)
2.12.17 (https://scastie.scala-lang.org/1ZQxsfQvRxeA1ulIZcTI6Q)

Note: 3.2.2 behaves as expected (https://scastie.scala-lang.org/o9DgmHrBQ0ymlC3KHnxfqg)

trait Module {
  trait X[A]
  object X {
    trait One[A] extends X[A]
  }

  implicit def ya: X.One[a]
}

class a

object Test {
  def ok(module: Module) = {
    val x = module
    import x._

    implicitly[X[a]]
  }

  def notOk(module: Module) = {
    import module._

    implicitly[X[a]] //fails
  }
}

Output

/path/to/source.scala:23:15: implicit error;
!I e: module.X[<empty>.a]
    implicitly[X[a]] //fails
              ^
one error found

Expected behavior

Code compiles

[som-snytt]

The matchesPt test in ImplicitComputation uses depoly which is commented why it's handling the dependent type as it does. No idea if this is a fundamental restriction or can be tweaked.

Scala 3 shows:

def ok(module: Module): Module#X[a]
def notOk(module: Module): module.X[a]

[som-snytt]

perhaps relatedly #5070

OH: "changing implicit resolution is too risky in a minor release"

[yoohaemin]

I see. Thanks for setting the expectation. Still for me it would be great if it can be fixed, my library makes use of this pattern pretty often.

[som-snytt]

Apologies, that was an old quote in a linked ticket, and I intended to highlight the irony that there is never a good time to change implicit resolution. But Scala 3 permits us to backport behavior under -Xsource:3. Also, maybe this is just a fixable bug.