Regression for method dispatch in 7mind/izumi

[WojciechMazur]

Compiler version

Last good release: 3.6.0-RC1-bin-20240719-af933c4-NIGHTLY
First bad release: 3.6.0-RC1-bin-20240723-46ff151-NIGHTLY

Bisect scripts failed to build the compiler for some revisions, the first bad commit could be any of:
3b3b8a7
637b1d3
2cfc561
875af44
dec395b

Minimized code

trait IO2[F[_, _]]:
  def apply[A](effect: => A): F[Throwable, A] = ???

trait  syntax:
  def IO2[F[_, _], A](effect: => A)(implicit F: IO2[F]): F[Throwable, A] = ???
  def IO2[F[_, _]: IO2]: IO2[F] = ???

class Test extends syntax:
  class X[F[_, _]: IO2]:
    def fails = IO2(println("hello world!"))
    def works = IO2(println("hello world!"))(using summon[IO2[F]])

Output

[error] ./src/test.scala:10:17
[error] Ambiguous overload. The overloaded alternatives of method IO2 in trait syntax with types
[error]  [F[_$5,_$6]](using evidence$1: IO2[F]): IO2[F]
[error]  [F[_$3,_$4], A](effect: => A)(implicit F: IO2[F]): F[Throwable, A]
[error] both match arguments (Unit)
[error]     def fails = IO2(println("hello world!"))
[error]                 ^^^

Expectation

Should compile

[Gedochao]

cc @kasiaMarek @tgodzik

[tgodzik]

Wait, that's not possible. None of these commits touch the compiler itself. It's only related to the presentation compiler module

[tgodzik]

Could this just be flaky?

[dwijnand]

In af933c4...46ff151 #21226 looks like the best candidate, with #21248 as a fallback. Those commits mentioned don't even seem to be part of the commit range. Is the commit message auto-generated by a script? Because I think there's a bug in the script.

[dwijnand]

Btw, this continues to not compile for me locally on main, which is a2c53a1. cc @EugeneFlesselle who's been active on the given prio change changes.

[EugeneFlesselle]

The issue minimizes to:

trait P[X]
given P[Int] with {}

def foo[F]()(implicit p: P[F]): F = ???
def foo[F: P]: () => F = ???

def works = foo()(using summon[P[Int]])
def fails = foo()

It is introduced by Switch mapping of context bounds to using clauses in 3.6 #21257.
The 2nd foo definition was/is equivalent to

• def foo[F](implicit ev: P[F]): () => F = ??? before the changes,
• def foo[F](using P[F]): () => F after the changes.

With the 2nd version, we have the same error both before and after the changes:

11 |def fails = foo()
   |            ^^^
   |Ambiguous overload. The overloaded alternatives of method foo with types
   | [F](using x$1: P[F]): () => F
   | [F]()(implicit p: P[F]): F
   |both match arguments ()
   |----------------------------------------------------------------------------
   | Explanation (enabled by `-explain`)
   |- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
   | There are 2 methods that could be referenced as the compiler knows too little
   | about the expected type.
   | You may specify the expected type e.g. by
   | - assigning it to a value with a specified type, or
   | - adding a type ascription as in instance.myMethod: String => Int
    ----------------------------------------------------------------------------

I believe this is an intended change in behavior of #21257. Unless it is causing too many regressions, this should probably be a won't-fix. \cc @WojciechMazur @Gedochao

---

A trivial, but unideal, user-code fix would be to manually make the context bound an implicit argument in the 2nd def IO2. A better fix would be to, as the error message suggests, use an ascription:

def fails = (IO2: IO2[F])(println("hello world!")) // now works

[Gedochao]

I believe this is an intended change in behavior of #21257. Unless it is causing too many regressions, this should probably be a won't-fix.

@EugeneFlesselle agreed. Closing this as won'tfix