New parameter type inference induces regression

[som-snytt]

reproduction steps

trait T[-X, +Y] {
  def apply(x: X): Y
  def g[Z](t: T[Y, Z]): T[X, Z] = ???
  def g[Z](m: Y => Z): T[X, Z] = ???
  def f[W](t: T[W, X]): T[W, Y] = t.g(this)
}

problem

SO.scala:5: error: type mismatch;
 found   : T[X,Y]
 required: W => ?
  def f[W](t: T[W, X]): T[W, Y] = t.g(this)
                                      ^

expectation

By commenting out the apply method, overload resolution proceeds as expected.

Explicitly annotating t.g[Y](this) also works around the behavior.

Maybe if assisted type inference ("higher-order function type parameter inference") fails during overload resolution, it could fall back to the old behavior and type check with unexpected type and correspondingly low expectations.

[som-snytt]

https://stackoverflow.com/questions/60677577/why-scalac-gets-confused-my-sam-type-not-implementing-function-when-it-overloads/

[SethTisue]

Note the SAM angle: the apply method brings the possibility SAM-conversion to Function1 into play.

Curious how minimized this is — does every type parameter matter here? Every variance annotation?

[som-snytt]

No idea if "compiles in dotty" implies "fixed in dotty".